Asme - Journal of Vibration and Acoustics - July 2003

Transactions Journal of Vibration
of the ASME and Acoustics ®
Editor
LAWRENCE A. BERGMAN Published Quarterly by The American Society of Mechanical Engineers
Assistant to the Editor
LINDA CONWAY
VOLUME 125 • NUMBER 3 • JULY 2003
Past Editors
F. EHRICH TECHNICAL PAPERS
T. CONRY
D. J. INMAN 249 Spectrum of High-Frequency Acoustic Noise in Inviscid Liquid-Linear
DESIGN ENGINEERING DIVISION Approximation for Spherical Waves
L. Likhterov and A. Berman
Associate Editors
J. CUSUMANO „2003… 252 Theory on Pitch Noise and Its Application
S. DYKE „2005… Yukio Nakajima
G. FLOWERS „2005…
M. I. FRISWELL „2004… 257 Investigation of the Sound Transmission into an Advanced Grid-Stiffened
J. GINSBERG „2005… Structure
R. P. S. HAN „2003… Jeffrey S. Vipperman, Deyu Li, Ilya Avdeev, and Steven A. Lane
J. MAIN „2005…
267 Optimal Stiffener Design for Interior Sound Reduction Using a Topology
D. QUINN „2005…
R. PARKER „2005… Optimization Based Approach
C. H. TAN „2004… Jianhui Luo and Hae Chang Gea
J. WICKERT „2003… 274 New Evaluation Method on Gear Dynamics Using Continuous and
NOISE CONTROL AND ACOUSTICS Discrete Wavelet Transforms
DIVISION Yuji Ohue and Akira Yoshida
R. KELTIE „2004…
R. OHAYON „2003… 282 A Stochastic Model for Simulation and Diagnostics of Rolling Element
Bearings With Localized Faults
BOARD ON COMMUNICATIONS
Chair and Vice-President J. Antoni and R. B. Randall
OZDEN OCHOA 290 Nonlinear Dynamic Analysis of a Rotor Shaft System With Viscoelastically
OFFICERS OF THE ASME Supported Bearings
President, REGINALD VACHON Nabeel Shabaneh and Jean W. Zu
Exec. Director 299 Nonlinear Parameter Estimation in Rotor-Bearing System Using Volterra
VIRGIL R. CARTER Series and Method of Harmonic Probing
Treasurer Animesh Chatterjee and Nalinaksh S. Vyas
R. E. NICKELL 307 Effect of Thrust Magnetic Bearing on Stability and Bifurcation of a
PUBLISHING STAFF Flexible Rotor Active Magnetic Bearing System
Managing Director, Engineering Y. S. Ho, H. Liu, and L. Yu
THOMAS G. LOUGHLIN
317 Torsional Vibration Analysis of Complicated Multi-Branched Shafting
Director, Technical Publ.
Systems by Modal Synthesis Method
PHILIP DI VIETRO
Chun-Ping Zou, Duan-Shi Chen, and Hong-Xing Hua
Manager, Journals
JOAN MERANZE 324 Study on the Dynamics of a Rotor in a Maneuvering Aircraft
Production Coordinator Fusheng Lin and Guang Meng
RAY RAMONAS 328 Control of Self-Excited Vibration of a Rotor System With Active Gas
Production Assistant Bearings
MARISOL ANDINO Jinhao Qiu, Junji Tani, and Taekyu Kwon
Transactions of the ASME, Journal of Vibration and 335 Vibration of Flex Circuits in Hard Disk Drives
Acoustics (ISSN 1048-9002) is published quarterly
(Jan., April, July, Oct.) by The American Society of J. A. Wickert
Mechanical Engineers, Three Park Avenue, New York, NY
10016. Periodicals postage paid at New York, NY and 343 Vibration Control of a Traveling Suspended System Using Wave
additional mailing offices. POSTMASTER: Send address Absorbing Control
changes to Transactions of the ASME, Journal of
Vibration and Acoustics, c/o THE AMERICAN SOCIETY OF M. Saigo, K. Tani, and H. Usui
MECHANICAL ENGINEERS, 22 Law Drive, Box 2300,
Fairfield, NJ 07007-2300. CHANGES OF ADDRESS must be 351 Dynamic Stiffness Formulation and Its Application for a Combined Beam
received at Society headquarters seven weeks before and a Two Degree-of-Freedom System
they are to be effective. Please send J. R. Banerjee
old label and new address.
STATEMENT from By-Laws. The Society shall not be
responsible for statements or opinions advanced in papers
359 Effect of Damping on Asymmetric Systems
or ... printed in its publications (B7.1, Par. 3). Paolo Gallina
COPYRIGHT © 2003 by the American Society of Mechanical
Engineers. Authorization to photocopy material for
internal or personal use under circumstances not falling
within the fair use provisions of the Copyright Act is granted
by ASME to libraries and other users registered with
the Copyright Clearance Center (CCC) Transactional
Reporting Service provided that the base fee of $3.00 per „Contents continued on inside back cover…
article is paid directly to CCC, 222 Rosewood Dr.,
Danvers, MA 01923. Request for special permission or bulk
copying should be addressed to Reprints/Permission
Department. INDEXED by Applied Mechanics Reviews and
Engineering Information, Inc. Canadian Goods and
Services Tax Registration #126148048
„Contents continued…
Journal of Vibration and Acoustics Volume 125, Number 3 JULY 2003
365 The Effect of Actuator and Sensor Placement on the Active Control of Rotor Unbalance
Marty E. Johnson, Luiz P. Nascimento, Mary Kasarda, and Chris R. Fuller
374 Robust Control of a 2D Acoustic Enclosure
Hemanshu R. Pota, Ian R. Petersen, and Atul G. Kelkar
384 Active Vibration Suppression With Time Delayed Feedback
Rifat Sipahi and Nejat Olgac
389 Bayesian Model Screening for the Identification of Nonlinear Mechanical Structures
Gaëtan Kerschen, Jean-Claude Golinval, and François M. Hemez
TECHNICAL BRIEFS
398 Closed-Form Exact Solution to Hⴥ Optimization of Dynamic Vibration Absorbers „Application to Different
Transfer Functions and Damping Systems…
Toshihiko Asami and Osamu Nishihara
405 Determining Stability Boundaries Using Gyroscopic Eigenfunctions
Anthony A. Renshaw
407 Rotor Isolation for Vibration and Noise Reduction
Timothy A. Brungart and Eric T. Riggs
The ASME Journal of Vibration and Acoustics is abstracted and

indexed in the following:
Acoustics Abstracts, Aluminum Industry Abstracts, Applied Science & Technology
Index, AMR Abstracts Database, Aquatic Science and Fisheries Abstracts, Civil
Engineering Abstracts, Compendex (The electronic equivalent of Engineering Index),
Corrosion Abstracts, Current Contents, EEA (Earthquake Engineering Abstracts
Database), Electronics & Communications Abstracts, Engineered Materials Abstracts,
Engineering Index, Enviroline (The electronic equivalent of Environment Abstracts),
Environment Abstracts, Environmental Engineering Abstracts, Environmental Science
and Pollution Management, Excerpta Medica, Fluidex, Health & Safety Science
Abstracts, INSPEC, Materials Science Citation Index, Mechanical & Transportation
Engineering Abstracts, Mechanical Engineering Abstracts, METADEX (The electronic
equivalent of Metals Abstracts and Alloys Index), Pollution Abstracts, Referativnyi Zhur-
nal, Shock & Vibration Digest, Solid State and Superconductivity Abstracts, Steels Alert
Spectrum of High-Frequency
Acoustic Noise in Inviscid
L. Likhterov
Liquid-Linear Approximation for
A. Berman
Institutes for Applied Research
Spherical Waves
and Dept. of Biotechnology Engineering,
The high-frequency asymptotics of the acoustic noise spectrum is considered for the case
Ben-Gurion University of the Negev,
of spherically symmetric waves propagating in an unbounded inviscid liquid. Using the
Beer-Sheva, 84105, Israel
Kirkwood and Bethe hypothesis regarding kinetic enthalpy, the Euler equations, the equa-
tion of state in the Tait’s form and following linearization allow the kinetic enthalpy and
‘‘reduced’’ pressure to be obtained. The Fourier transform yields the spectral density of
acoustic energy which proves to be inversely proportional to the square frequency and
decreases approximately by 6 decibels per octave with increase of a frequency.
关DOI: 10.1115/1.1570446兴
1 Introduction ⳵␳ 2 ␳ ⳵␯ ⳵␳
⫹ ␯⫹␳ ⫹␯ ⫽0 共 the continuity equation兲 (2)
It is known that the acoustic noise spectrum has some slope in ⳵t r ⳵r ⳵r
the high range of frequencies. The typical Knudsen sea-noise
spectral slope is about ⫺5 decibels per octave 关1兴, 共see p. 336兲. with addition of the Tait’s equation of state for water
冉冊
Acoustic noise spectra generated by oscillating bubbles have been
studied most completely. Analysis of the acoustic energy gener- p⫹B ␳ n
⫽ , (3)
ated versus frequency 关2兴 suggests that the ⫺5 decibels per octave p 0 ⫹B ␳0
wind dependent ambient noise slopes of the Knudsen curves are
caused by the shorter lifetimes of high-frequency bubbles, rather where B⫽300 MPa⬇3000 atm and n⫽7.
than significantly lower peak pressure. The average of several As has been noted by Vogel, Bush and Parlitz 关7兴, the equation
acoustic spectra from a single energetic spill was shown a slope of of state given by Rice and Walsh 关8兴 may be used but this equa-
⫺5 decibels per octave over the frequency range up to 8 kHz tion can not easily be incorporated into the Gilmore 关9兴 model that
according to the same authors. The experimental data cited by has been developed on the basis of the Kirkwood and Bethe hy-
Longuet-Higgins 关3兴 show that deep water bubble noise spectra in pothesis 共limited to ongoing waves兲 described in the monograph
high-frequency range 共over 10 kHz兲 have the slope of ⫺共5–7兲 by Cole 关10兴.
decibels per octave. Pumphrey and Crum 关4兴 have obtained acous- Related more rigorous and general results can be obtained for
tic power spectra for rain drops falling onto a lake in slightly the radiated pressure from a bubble taking into account the heat
windy conditions which have the slope approximately ⫺5 deci- conduction in the fluid outside the bubble and the effects of the
bels per octave. In the spacious review by Prosperetti and Oguz liquid compressibility 关11兴. However, the Kirkwood and Bethe
关5兴, the comparison between measured and computed underwater theory gives good results and therefore found wide use for study-
noise spectra demonstrates the decrease of sound levels about ing the generation and propagation of shock waves. Beside that, it
⫺5.5 decibels per octave. According to Urick 关6兴, 共see p. 209兲, the is assumed that the considered process will be isothermal. In the
spectral slope of a spectrum of deep-sea noise is ⫺共5– 6兲 dB/ analysis that follows, the above-mentioned hypothesis is em-
octave in the frequency band from one up to hundred kilohertz, ployed. In terms of the enthalpy, dh⫽dp/ ␳ , the sound speed, c 2
and further, the thermal noise leads to the 6 dB/octave increase of ⫽ ⳵ p/ ⳵␳ , and above-mentioned hypothesis which postulates that
the spectrum level. However, such a rise gives the unlimited the kinetic enthalpy propagates in medium with the velocity c
acoustic energy (E ac⫽ 兰 ⬁0 Sd ␻ , where S is the spectral density and ⫹␯:
冋册冋冉冊册
␻ is the frequency兲. It is of interest to elucidate analytically the
asymptotic behavior of an acoustic spectrum of spherical waves ⳵ ⳵ ␯2
⫹ 共 c⫹ ␯ 兲 r h⫹ ⫽0, (4)
spreading in unbounded inviscid medium for following compari- ⳵t ⳵r 2
son with experimental data.
the Euler equations 共1兲 and 共2兲 can be written according to Cole
关10兴 as
2 Initial Equations ⳵␯ ⳵␯ ⳵h
The initial system of governing equations is the Navier-Stokes ⫹␯ ⫽⫺ , (5)
⳵t ⳵r ⳵r
equations which for inviscid medium are reduced to the Euler
equations 共the spherical case is considered and the radial velocity ⳵␯ 2 ␯ 1 ⳵ h
is denoted ␯ r ⫽ ␯ omitting the index兲: ⫹ ⫽ 2 . (6)
⳵r r c0 ⳵t
⳵␯ ⳵␯ 1 ⳵p
⫹␯ ⫽⫺ 共 the motion equation兲 (1) It should be noted that Eq. 共4兲 is not an equivalent to 关 ⳵ / ⳵ t⫹(c
⳵t ⳵r ␳ ⳵r
⫹ ␯ ) ⳵ / ⳵ r 兴 (r ␸ )⫽0, where ␸ is the flow potential, but its use al-
Contributed by the Technical Committee on Vibration and Sound for publication
lows the relatively simpler result to be obtained.
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002; According to the Tait’s equation, the enthalpy, h, can be ex-
Revised February 2003. Associate Editor: R. F. Keltie. pressed in following form:
Journal of Vibration and Acoustics Copyright © 2003 by ASME JULY 2003, Vol. 125 Õ 249
c0 According to the formula 共2.101兲 from the handbook of Kamke
h⫽ 共 p n⫺1/n ⫺1 兲 , (7) 关12兴, it should be assumed u(r)⫽rA, and Eq. 共17兲 reduces to the
n⫺1 *
linear equation
where c 20 ⫽n/ ␳ 0 (p 0 ⫹B) is the sound speed in undisturbed me-
dium and p ⫽p⫹B/p 0 ⫹B is the ‘‘reduced’’ pressure. The den- u ⬙ ⫹k 2 u⫽0, (18)
*
sity, ␳, will then be determined as
whose general solution 共for k⬎0) is
␳ ⫽ ␳ 0 p 1/n . (8)
* u⫽C 1 cos kr⫹C 2 sin kr. (19)
3 Determination of Enthalpy
Therefore, the solution of Eq. 共17兲 can be written in the form:
From Eq. 共6兲, the radial derivative of the velocity is:
1 1
⳵␯ 1 ⳵ h 2 ␯ A⫽C 1 cos kr⫹C 2 sin kr, (20)
⫽ ⫺ , (9) r r
⳵ r c 20 ⳵ t r
and this expression can be substituted into the motion equation 共5兲 and, consequently, the solution of Eq. 共15兲 is
that gives
1 ⫺␻t
⳵␯
⳵t
⫹␯ 2 ⫺
c0 ⳵t 冋
1 ⳵h 2␯
r
⫽⫺
⳵h
⳵r
. 册 (10)
h⫽ 共 C 1 cos kr⫹C 2 sin kr 兲
Since the enthalpy must remain limited, the constant C 1 has to

r
e . (21)
Now, differentiating Eq. 共10兲 with respect to r, differentiating the be equated to 0. Using the expression for enthalpy 共7兲, the ‘‘re-
continuity equation 共6兲 with respect to t, and subtracting the sec- duced’’ pressure is determined as
冋册
ond from the first, it follows
共 n⫺I 兲 h n/n⫺1
冋册
⳵␯ ␯ ⳵ h ⳵h 4 2 ⳵ h 1 ⳵ h
2 2 2 2 p ⫽ ⫹1 , (22)
* c 20
⫺ 2 ⫹ ⫹ ␯ ⫽ 2⫹ 2 2. (11)
⳵r c0 ⳵t⳵r ⳵r r2 ⳵r c0 ⳵t
therefore, in the first approximation,
The square of the continuity equation is
冋册冋册
p⫹B n
⳵␯ 2 4 ␯ ⳵␯ 4 ␯ 2 1 ⳵ h 2 ⫽1⫹ 2 h⫹ . . . (23)
⫺ ⫹ 2 ⫽ 4 , (12) p 0 ⫹B c0
⳵r r ⳵r r c0 ⳵t
and neglecting p 0 as compared with B, the acoustic pressure is
and Eq. 共11兲 becomes
冋册
n sin kr ⫺ ␻ t
1 ⳵ h 2 4 ␯ ⳵␯ ␯ ⳵ 2 h 2 ⳵h ⳵ 2h ␯ ⳵ 2h p⬇B C2 e . (24)
⫺ ⫹ 2 ⫹ ⫽⫺ 2 ⫹ . c 20 r
c0 ⳵t
4
r ⳵r c0 ⳵t⳵r r ⳵r ⳵r c0 ⳵t2
(13) The constant C 2 can be determined from the condition that p
⫽p 1 at t⫽0 and r⫽0 taking into account that limr→0 关 sin kr/r兴
Using Eq. 共6兲 once more, Eq. 共13兲 can be rewritten in the follow- ⫽k. Then, the expression for the pressure can be written as
ing form:
冋册
p 1 sin kr ⫺ ␻ t
⳵ 2h 2 ⳵ h 1 ⳵ 2h 1 ⳵ h 2 4 ␯ ⳵ h 8 ␯ 2 ␯ ⳵ 2h p⬇ e . (25)
⫹ ⫹ ⫽ ⫺ ⫺ 2 ⫹ 2 , k r
⳵ r 2 r ⳵ r c 20 ⳵ t 2 c 40 ⳵ t r c 20 ⳵ t r c0 ⳵t⳵r
(14) 4 Acoustic Spectrum
where all nonlinear terms are collected in the right-hand part of The Fourier transform of an aperiodic function has the form:
this equation.
The evaluation of the nonlinear terms shows that the first term
contains a multiplier c ⫺4
0 , the second and fourth terms are propor-
p共 t 兲⫽ 冕 ⬁
⫺⬁
L共 i ␻ 兲e i␻td ␻ , (26)
tional to c ⫺1
0 M
⫺1
共M is the Mach number兲 and the third term
corresponds to the spherical expansion which can be considered where
冕
as sufficiently small at large distances from the source of pertur- ⬁
1
bations. Therefore, these terms may be neglected if the linear L共 i␻ 兲⫽ p 共 t 兲 e ⫺i ␻ t dt (27)
approximation is considered, and the linear equation follows for 2␲ ⫺⬁
determination of the enthalpy:
is the complex function of the frequency, ␻.
⳵ 2h 2 ⳵ h 1 ⳵ 2h Assuming that p(t)⫽0 at ⫺⬁⬍t⬍0, the integral in Eq. 共25兲
⫹ ⫹ ⫽0. (15) can be calculated making use of the formula 3.351.3 from Grad-
⳵ r 2 r ⳵ r c 20 ⳵ t 2
stein and Ryzhik 关13兴共containing there in a exponent of a power
Assuming h⫽A(r)e ⫺ ␻ t , where A is a function of r only, the coefficient n⫽0)
冕
equation for determination of A(r) is obtained: ⬁ 1
d 2 A 2 dA L 共 i ␻ 兲 ⫽K e ⫺ 共 1⫹i 兲 ␻ t dt⫽K , (28)
共 1⫹i 兲 ␻
⫹ ⫹k 2 A⫽0, (16) 0
dr 2 r dr
where the coefficient K⫽1/2␲ p 1 /k 关 sin kr/r兴 is introduced for
where k⫽ ␻ /c 0 . brevity.
Multiplying this equation by r, this equation 共connected with The spectral density of an acoustic energy is given by the
the Bessel’s equation兲 becomes: modulus square of L(i ␻ ):
rA ⬙ ⫹2A ⬘ ⫹k 2 rA⫽0. (17) S共 ␻ 兲⫽ 兩 L共 i ␻ 兲兩 2, (29)
250 Õ Vol. 125, JULY 2003 Transactions of the ASME

therefore, with the accuracy up to numerical multiplier, the spec- 关3兴 Longuet-Higgins, M. S., 1990, ‘‘Bubble Noise Spectra,’’ J. Acoust. Soc. Am.,
tral density changes inversely proportional to the frequency 87, pp. 652– 661.
关4兴 Pumphrey, H. C., and Crum, L. A., 1990, ‘‘Bubble Noise Spectra,’’ J. Acoust.
square:
Soc. Am., 87, pp. 142–148.
S⬃ ␻ ⫺2 . (30) 关5兴 Prosperetti, A., and Oguz, H. M., 1993, ‘‘The Impact of Drops on Liquid
Surface and the Underwater Noise of Rain,’’ Ann. Rev. Fluid Mech, 25, pp.
577– 602.
5 Conclusion
关6兴 Urick, R. J., 1983, Principles of Underwater Sound, 3rd edition, McGraw-Hill
The result obtained shows that a linearization of initial equa- Book Company.
tions gives the decrease of the spectral density with the frequency 关7兴 Vogel, A., Bush, S., and Parlitz, U., 1996, ‘‘Shock Wave Emission and Cavi-
growth approximately 10 log 2⫺2⬇⫺6 decibels per octave for tation Bubble Generation by Picosecond and Nanosecond Optical Breakdown
in Water,’’ J. Acoust. Soc. Am., 100共1兲, July, 148 –165.
spherical waves propagating in inviscid medium. This may be
关8兴 Rice, M. H., and Walsh, H. M., 1957, ‘‘Equation of State of Water to 250
found useful for acoustic signal processing on background of an Kilobars,’’ J. Chem. Phys., 26, pp. 824 – 830.
ambient noise and separation and classification of acoustic 关9兴 Gilmore, R. F., 1952, ‘‘The Growth and Collapse of a Spherical Bubble in a
sources. Viscous Compressible Fluid,’’ Calif. Inst. Tech. Rep., 26-4.
关10兴 Cole, R. H., 1948, Underwater Explosions, Princeton U.P., Princeton, NJ.
References 关11兴 Temkin, S., 1999, ‘‘Radial Pulsation of a Fluid Sphere in a Sound Wave,’’ J.
Fluid Mech., 380, pp. 1–38.
关1兴 Medwin, H., and Clay, C. S., 1998, Fundamentals of Acoustical Oceanogra-
关12兴 Kamke, E., 1959, Differentialgleichungen, Lösungsmethoden und Lösungen,
phy, Academic Press.
关2兴 Updegraff, G. E., and Anderson, V. C., 1991, ‘‘Bubble Noise and Wavelet Vol. 1, Akad. Ver., Leipzig.
Spills Recorded 1 m Below the Ocean Surface,’’ J. Acoust. Soc. Am., 86, pp. 关13兴 Gradstein, I. S., and Ryzhik, I. M., 1965, Tables of Integrals, Series and Prod-
2264 –2279. ucts, Academic Press, New York.
Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 251

Theory on Pitch Noise and Its
Yukio Nakajima
Bridgestone Corporation
Application
3-1-1, Ogawahigashi-Cho, Kodaira-Shi, Tokyo
A theory on pitch noise of a tire has been developed based on the assumption that noise
187-8531, Japan
is approximated by a series of Dirac delta-function. The proposed theory predicts the
e-mail: [email protected]
existence of special lug shapes where tire noise becomes almost zero. In addition a tire
noise simulation system is developed based on this theory to predict tire noise from CAD
data. The prediction of overall noise of various tires with practical patterns is verified to
be in good agreement with the measurement. 关DOI: 10.1115/1.1569513兴
1 Introduction Assume the lug shape can be described by

Noise nuisance problems have increased due to the increase of y i 共 x 兲 ⫽g i 共 x 兲 ⫹ ␰ i . (3)
traffic volume and have become a community noise environment
problem. In order to solve the noise problem, regulations have g i (x) is the function of lug shape and ␰ i are coordinates deter-
become strict and manufacturers have invested in noise reduction mined by pitch variation. The peak amplitude of tire noise spec-
research. Vehicle manufacturers have been particularly successful trum is reduced by a pitch variation treatment 关7兴. The spectrum
for significant noise reduction in engine, exhaustion and transmis- with pitch variation treatment is widely distributed in a frequency
sion systems. Tire manufacturers have also studied a tire noise range like the white noise which has the same spectrum amplitude
reduction 关1–5兴. A tire, however, is becoming a dominant noise in every frequency. The tonal noise, hence, can be improved by a
source in a vehicle system and a tire causes 80 percent of total pitch variation treatment.
vehicle noise in steady rolling and 23 percent in acceleration 关6兴. Substituting Eq. 共3兲 for Eq. 共1兲 and integrating it with respect to
Research on tire noise reduction has dealt with tread pattern, the meridian direction, tire noise F(t) is obtained. By applying a
sidewall shape, construction, and material. This paper is related to Fourier transformation, F(t) can be described by noise amplitude
the research on tire pitch noise generated from tread pattern. In a n at nth harmonics as follows:
冕
various design elements of tread pattern as shown in Fig. 1, the W
N
lateral groove called the lug groove has the biggest effect on tire
pitch noise. Pitch refers to the distance between one groove on a
F共 t 兲⫽
0
f 共 t,x 兲 dx⫽ 兺ae
i⫽1
n
⫺i ␻ n t
(4)
tire’s tread and the corresponding point of the groove adjacent to
where W is half of tire contact width and ␻ n is given by
it along the tire’s circumference. Pitch noise is the noise generated
by the sequence of grooves. In this paper a theory on pitch noise 2␲n
is developed to predict the effect of the lug angle on tire pitch ␻ n⫽ . (5)
T
noise and a pattern to decrease tire pitch noise is proposed. Fur-
thermore a tire noise simulation system is developed based on this T is a cycle of rotation. Then a n can be obtained by the following
theory to predict tire noise from CAD data. The prediction of inverse Fourier transformation:
冕冕冕
overall noise of various tires with a practical pattern is verified to T W T
be in good agreement with the measurement. The proposed theory 1 1
a n⫽ F 共 t 兲 e i ␻ n t dt⫽ f 共 t,x 兲 e i ␻ n t dtdx. (6)
and the prediction system can be applicable to not only a tire but ␲ 0 ␲ 0 0
also a timing belt or gear where gear or tooth pitch causes noise.
Substituting Eqs. 共1兲 and 共3兲 for Eq. 共6兲,
冕
2 Tire Pitch Noise 1
N
W
2.1 Theory on Tire Noise. Tire pitch noise is approximated

a n⫽
␲ 兺e
j⫽1
␰ j /V
0
w 共 x 兲 e i ␻ n g j 共 x 兲 /V dx (7)
by the sequences of the following Dirac delta-function:
N
is obtained. In Eq. 共7兲 the term 兺 Nj⫽1 e ␰ j /V is related to pitch varia-
i ␻ n g j ( x ) /V
f 共 t,x 兲 ⫽ 兺 w 共 x 兲 ␦ 共 t⫺t 共 x 兲兲
i⫽1
i
tion and the term 兰 W 0 w(x)e
Assume that w(x) is given by
dx is related to lug shape.
(1)
y i共 x 兲 1
t i共 x 兲 ⫽ 共 i⫽1 . . . N 兲 w共 x 兲⫽ 兵 ␣ 共 W⫺x 兲 ⫹ ␤ x 其 (8)
V W
where w(x) is noise amplitude in the meridian direction repre- where ␣ and ␤ are noise amplitude at the tire center (x⫽0) and
sented by x, y i (x) is the ith lug position in the circumferential noise amplitude at the edge of tire footprint (x⫽W), respectively.
direction measured by a reference point, and V is the velocity. ␦ is 2.2 Theory on Magic Angle †1–2‡. Simplified tire pattern
Dirac delta-function defined by is modeled by assuming that the lug is a straight line with the
冕 h 共 t 兲 ␦ 共 t⫺t i 兲 dt⫽h 共 t i 兲 . (2)

same lug angle ␪. The shape of the lug is described by
g i 共 x 兲 ⫽tan ␪ x 共 i⫽1 . . . N 兲 . (9)
h(t) is an arbitrary function and t i is included in the range of Furthermore, assume that noise amplitude is the same in the me-
integral. ridian direction, w(x)⫽1. Substituting Eq. 共9兲 for Eq. 共7兲 and
integrating it, the following equation is obtained:
冏冉冊冏
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Dec. 2001; 2V ␻ nW
Revised Nov. 2002. Associate Editor: R. Keltie. 储 a n 共 ␪ 兲储 2 ⫽ sin tan ␪ d n (10)
␻ n tan ␪ 2V
252 Õ Vol. 125, JULY 2003 Copyright © 2003 by ASME Transactions of the ASME
Fig. 1 Tire noise and design elements of tread pattern
where 储 • 储 2 is the Euclidian norm and d n is the term related to the where L(⫽TV) is the circumference length of a tire. This equa-
pitch variation described by tion expresses the effect of the lug angle on pitch noise. P n ( ␪ )
冐兺冐
N
possesses multiple zeros where noise is never generated. We call
1 this lug angle the magic angle described by
d n⫽ e ␰ j /V . (11)
␲ j⫽1
2
冉冊
The ratio of noise with lug angle ␪ and noise with zero lug angle
is derived by using Eq. 共10兲 as follows: iL
␪ iM ⫽tan⫺1 共 i⫽1•••⬁ 兲 . (13)
冏冉冊冏
nW
储 a n 共 ␪ 兲储 2 2V ␻ nW
P n共 ␪ 兲 ⫽ ⫽ sin tan ␪ 2.3 Theory on Magic Shape. If ␣ is not equal to ␤ in Eq.
储 a n 共 0 兲储 2 ␻ n W tan ␪
冏冏
2V
共8兲, the optimized lug shape, which minimizes pitch noise, is not
⫽
sin 冉
␲ nW
L
tan ␪ 冊 , (12)
straight anymore. Hence, the optimized lug shape called the magic
shape must be determined by the optimization technique 关8兴. As-
sume that all lug shapes g j (x) are the same in all pitches and
␲ nW g j (x) is tangent to ␪ 2M 共second magic angle兲 at tire center. We
tan ␪
L approximate g j (x) by
Fig. 2 Magic shape

Fig. 3 Prediction of pitch noise of a tire with practical pattern
P rection described by w(x), noise amplitude is proportional to

g j 共 x 兲 ⫽g 共 x 兲 ⫽x tan ␪ 2M ⫹ 兺 bx
i⫽2
i
i
(14) w(x)•⌬x. By summing up every pattern edges, tire noise in time
domain can be described by the right graph of Fig. 3. In the case
where P is the order of polynomials. of a small lug angle noise becomes large, since noise amplitude
The design variables in the optimization are b i which are the becomes large in the time domain due to large ⌬x, and vice versa.
parameters to express lug shape in Eq. 共14兲. The objective func-
tion is to minimize tire pitch noise. Let’s consider the example
where W, L, n, P, ␣ and ␤ are 150 mm, 2000 mm, 50, 5, 1.0 and 4 Experiment and Discussion
0.5, respectively. In this example noise amplitude at the center Tire noise is measured in an anechoic room by a 3-m indoor
area is two times larger than that at the shoulder area. After ap- drum the surface of which is paved by a safety-walk. A micro-
plying various optimization techniques, the sequential quadratic phone is placed on the tire centerline and the distance from a tire
programming generated the best solution in this example. to the microphone is one meter and 0.25 meter height from drum
The optimized lug shape is shown by the dotted line in Fig. 2 surface. The tire is a passenger radial tire 共175R/70R13兲.
and the magic angle is shown by solid line as a reference. The In order to verify Eq. 共12兲, tires with lug angles of 0, 25 and 40
optimized lug shape is different from the magic angle ( ␪ 2M degrees measured form the meridian direction are prepared by
⫽46.8 degrees) and the tangential slope of the optimized lug be- hand cut. Lugs are grooved at the center area to eliminate the
comes smaller at the shoulder area than at the center area. This is effect of contact shape. The number of pitches is 60 and there is
because the noise amplitude at the shoulder area is small and the no pitch variation treatment. Measured data are the average of
amplitude needs to be increased by decreasing the lug angle at the overall noise at 40, 60, 80 and 100 km/h and indicated by black
shoulder to match the amplitude at the center area. Hence, the circles. The prediction of the proposed theory shows very good
magic shape depends on the value of ␣, ␤ and the footprint shape. agreement with measurement as shown in Fig. 4. Furthermore, in
The amplitude of noise with the magic shape is one hundredth of order to verify the existence of the magic angle defined by Eq.
the noise with the magic angle. 共13兲, a few patterns with different lug angles near the first magic
angle ␪ 1M 共56.3 degree兲 are grooved on a tire. Tire noise is mea-
3 Noise Prediction System of Practical Pattern sured near the leading edge and the average of overall noise at 40,
The noise prediction system of a tire with the practical pattern 60, 80 and 100 km/h is shown in Fig. 5 by a black circle. The
has been developed by calculating f (t,x) in Eq. 共1兲 from CAD theory indicated by solid line is in good agreement with measure-
data 关2,9兴. Considering that tire noise with the Dirac delta- ment and the propose theory is still valid near the magic angle.
function is generated just when the lug rolls in and out of ground, Then, the predictability of the tire noise prediction system is
f (t,x) can be easily calculated from CAD drawing as shown in verified by tires with a practical pattern. Tire noise 共overall with
Fig. 3. In order to consider the relationship between tire pattern A-weighted sound level兲 is measured in an anechoic room by a
and the edge of footprint in a rolling tire, it is convenient to move 3-m indoor drum with a safety-walk and tire speed is 100 km/h. A
the edge of the footprint in the circumferential direction on the tire microphone is placed on the tire centerline and the distance from
pattern. Assume that we observe a rolling tire at each time interval a tire to the microphone is one meter and 0.25 meter height from
of ⌬t. When tire velocity is V, the edge of the footprint moves on the drum surface. Tires are passenger radial tires with various
the tire pattern by the length V⌬t. If noise amplitude is the same sizes from 165R13 to 225/55R15. The following equation is used
in the meridian direction, noise amplitude is proportional to ⌬x for the prediction:
which is the lateral component of the length of intersection points 200
generated by the lug edge and the footprint edges at t and t
⫹⌬t. If noise amplitude has the distribution in the meridian di-
prediction⫽ 兺储a 储 A
n⫽1
n 2 n . (15)

Fig. 4 Prediction of lug angle vs tire pitch noise
Fig. 5 Verification of magic angle „measurement…
n is the number of harmonics, A n is the parameter for A-weighted

sound level and 储 a n 储 2 is the amplitude at nth harmonics. In this
prediction w(x) are assumed to be the same in the meridian di-
rection and both the leading and trailing edges. Predicted overall
tire noise is compared with the measurement as shown in Fig. 6.
The prediction is in good agreement with the measurement and
the tire noise prediction system is verified to be effective for pat-
tern design for noise reduction.
In order to improve predictability, a database needs to be estab-
lished including such data as distribution of noise amplitude in the
meridian direction, noise amplitude at the leading and trailing
edges, relationship between noise and material characteristics, and
relationship between noise and block stiffness, etc.
5 Summary
The theory on pitch noise of a tire has been developed based on
the assumption that noise is approximated by a series of Dirac
delta-function. Proposed theory predicts the effect of lug angle on
tire noise and the existence of special lug angles named magic
angles where tire noise becomes zero. The proposed theory was
Fig. 6 Comparison between prediction and measurement verified to quantitatively predict the effect of lug angle on tire

noise level and the existence of magic angles, where tire noise is W ⫽ half of tire contact width or lug width
theoretically zero, by the noise measurement of a grooved tire. x ⫽ meridian direction on tire
The theory was extended to pitch noise where noise amplitude yi ⫽ length measured from a reference point
has distribution in the meridian direction. In this case, the shape of ␣ ⫽ noise amplitude at tire center
the lug in minimizing tire noise was no longer a straight line and ␤ ⫽ noise amplitude at tire shoulder
became a complicated shape. The optimized lug shape, called the ␰j ⫽ position of lug in the circumferential direction
magic shape, was obtained by using optimization technique. Fur- ␪ ⫽ lug angle
thermore the tire noise prediction system was developed in order ␪ iM ⫽ ith magic angle
to apply this idea to the practical pattern. The prediction was
verified to be in good agreement with the measurement.
The proposed theory and the prediction system can be appli-
cable not only to a tire but also to a timing belt or gear where gear References
or tooth pitch causes noise. 关1兴 Nakajima, Y., 1980, ‘‘Theory on Pitch Noise 共I兲,’’ Internal Report, Bridgestone
Corporation.
Nomenclature 关2兴 Nakajima, Y., 1981, ‘‘Theory on Pitch Noise 共II兲,’’ Internal Report, Bridge-
stone Corporation.
储 a n储 2 ⫽ amplitude of noise of nth harmonic number 关3兴 Nakajima, Y., Inoue, Y., and Ogawa, H., 1993, ‘‘Application of the Boundary
Element Method and Modal Analysis to Tire Acoustics Problems,’’ Tire Sci.
An ⫽ the parameter for A-weighted sound level Technol., 21, p. 66.
bi ⫽ parameter to express lug shape 关4兴 Koike, H., Oshino, Y., and Tachibana, H., 1998, ‘‘Identification of Source of
f (t,x), F(t): ⫽ noise in time domain Tire/Road Noise 共in Japanese兲,’’ Souonseigyo, 22, p. 11.
g i (x) ⫽ function of lug shape 关5兴 Mukae, T., et al., 1978, ‘‘Clarification of Mechanism of Tire Noise Generation
共in Japanese兲,’’ Report of JARI, No. 51.
L ⫽ length of tire circumference 关6兴 JATMA, 1997, On Noise due to Tire and Road, 共6th edition兲共in Japanese兲.
n ⫽ harmonic number 关7兴 Nakajima, Y., and Abe, A., 2000, ‘‘Application of Genetic Algorithms of Op-
N ⫽ number of pitches timization of Tire Pitch Sequences,’’ Japan Journal of Industrial and Applied
T ⫽ cycle of rotation Mathematics, 17, pp. 403.
关8兴 Vanderplaats, G. N., 1984, Numerical Optimization Techniques for Engineer-
ti ⫽ time when lug at x contacts on ground ing Design with Applications, McGraw-Hill.
V ⫽ velocity 关9兴 Yoshida, M., and Takayama, M., 1994, ‘‘Method of Simulating Tire Tread
w(x) ⫽ weighting function Noise and Simulator Therefore,’’ USA Patent #4,838,330.

Investigation of the Sound
Jeffrey S. Vipperman
Assistant Professor, Associate Mem. ASME,
Transmission into an Advanced
Deyu Li
Graduate Research Assistant
Grid-Stiffened Structure
The noise transmission behavior of an advanced grid-stiffened (AGS) composite structure
Ilya Avdeev has been investigated by combining numerical and experimental methods. Structural-
John Swanson Graduate Fellow, acoustic coupling was found to be light, permitting separate analysis of the structure and
Student Mem. ASME acoustic cavity. Finite element analysis permitted the resonant frequencies of acoustic
cavity and structure to be calculated, which play an important role for noise transmission
Department of Mechanical Engineering, through the structure. Acoustic mode shapes permitted internal coincidence frequencies to
University of Pittsburgh, be estimated and provided insight into modal pressure distributions, when considering
Pittsburgh, PA 15261 payload location. Experimental structural and acoustic modal analysis permitted the
resonant frequencies and damping ratios for the structure and cavity to be determined,
which in turn were used to corroborate the FEA model. Finally, direct measurement of the
Steven A. Lane noise transmission was performed based on noise reduction spectrum (NRS), which is
Aerospace Engineer, calculated from spatial averages of the RMS acoustic pressures inside and outside of the
Air Force Research Laboratory, shell. It was found that the NRS was dominated by acoustic resonances, which were
Space Vehicles Directorate, marked by sharp dips in the NRS curve. Internal coincidence of the axial wavenumbers
Kirtland AFB, NM 87117 was also found to be a significant mechanism for noise transmission. External coinci-
dence and ring frequencies were found to provide less of an impact on the overall NRS for
the structure. 关DOI: 10.1115/1.1569511兴
Introduction frequencies on measured noise transmission. These prior studies

Novel composite structures that have application for use as ex- illustrate the need for a detailed examination of the physical
pendable launch vehicle fairings are being investigated. Compos- mechanisms of noise transmission for composite fairing
ites are already offering higher strength, less weight, and larger structures.
payload envelopes than current systems. Two examples of inno- The Sound, Systems, and Structures Laboratory at the Univer-
vative composite construction are the Advanced Grid-Stiffened sity of Pittsburgh is currently engaged in a more thorough char-
共AGS兲关1兴 and Chamber Core designs 关2兴, which are additional acterization of the vibroacoustic behavior of the mock-scale AGS
contenders to replace the conventional machined-aluminum iso- rocket fairing. Methods used combine experimental structural-
grid structures. Both are fabricated of filament-wound graphite acoustic modal analysis, finite element analysis 共FEA兲, direct
fiber that has been preimpregnated with an epoxy binder. The measurement of the noise transmission, and an investigation of the
improvements in weight 共40– 60% decrease兲 and strength 共1,000% effects of the ring frequency and internal and external coinci-
increase along the hoop direction兲 are accompanied by a worsen- dence. Thorough characterization of the TL behavior of rocket
ing of the sound transmission loss 共TL兲 behavior of the structure, fairings is necessary in order to understand the dominant physical
which is governed by the physical structural properties 共mass, mechanisms of noise transmission for these types of structures.
stiffness, and damping兲 and the geometry. The interior acoustic Such knowledge will aid future efforts to improve the noise trans-
levels experienced during launch are significant enough to damage mission loss through geometry or property tuning or through the
the sensitive cargo.
use of passive or active control techniques.
Initial studies of AGS structures have focused on static or buck-
This paper will present experimental and companion numerical
ling strength, the impact of defects, effects of damage or repairs,
and rudimentary dynamic behavior 关1兴. Modeling studies on im- 共FEA兲 study results for the mock-scale AGS rocket fairing system.
proving the transmission loss in fairings have been conducted for Measurements include modal analysis of the structure and acous-
Minotaur-sized launch vehicles where passive 关3兴 and feedback tic cavity and direct measurement of the noise transmission or
active structural-acoustic control 共ASAC兲关4,5兴 were numerically reduction for the structure. A novel multi-level finite element
evaluated. Two earlier experimental studies focused on specific analysis is developed 关10兴 to predict the decoupled resonant fre-
types of transmission control for the AGS mock-scale prototype, quencies and mode shapes for the in-vacuo structure and rigid
including evaluating the effectiveness of hybrid structural and wall acoustic modes. Coupling between the structure and acoustic
acoustic control inputs 关6兴 as well as positive position feedback cavity was found to be light, as expected, permitting the separate
共PPF兲 controllers 关7兴, which do not require overly complicated analyses. The modal analysis results are then related back to the
system models and can offer some stability guarantees. An active transmission behavior of the structure, which is governed by reso-
control study on a full-scale model has shown that actuator control nance and coincidence behaviors. Organization of the manuscript
authority and power limitations may prove difficult 关8兴. A more is as follows. The next sections contain a general description of
recent study 关9兴 briefly examined an isotropic, truncated, conical transmission loss and structural-acoustic behavior, comments on
shell and investigated the effects of cavity resonances and ring the fairing construction, the results of the experimental and nu-
merical modal analyses of the structure and the acoustic cavity,
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Feb. 2002; measurement of the noise-reduction, which is similar to transmis-
Revised Nov. 2002. Associate Editor: R. F. Keltie. sion loss, followed by a summary of the findings.
Fig. 1 Typical transmission loss spectrum for a flat panel
Acoustic Transmission Loss

The ‘‘transmission loss’’ 共TL兲 of a panel is defined as
␥ SA ⬀ 冕
S
␾ mn 共 rជ 兲 ␺ o pq 共 rជ 兲 dS (2)
where m, n, o, p, q are modal indices and rជ is a position vector.

TL⫽10 log10 冋册
W inc
W trans
⫽10 log10
1
␶冋册 (1)
Such coupling is typically complex and dominated by a few com-
binations of modes. In particular, internal coincidence between the
structural and acoustic axial wave numbers can occur, as will be
discussed further in the next section.
where W trans is the acoustic power transmitted through the panel
Note that transmission measurements based on power 共Eq. 共1兲兲
and W inc is the incident power, and the ratio: ␶ ⫽W trans /W inc is
would be very difficult to perform for this structure. For example,
referred to as the transmission coefficient. Intuitively, the trans-
if substitution techniques are performed by placing a known
mission of power will depend upon the reactive impedance mis-
power source inside the acoustic cavity, the acoustic environment
match of the panel with air 共stiffness and mass兲 as well as any
of the cavity itself would be changed. Also, internal sound inten-
internal dissipation mechanisms 共damping兲. These physical prop-
sity measurements are impractical. Therefore, an in-situ method of
erties of the system in conjunction with the geometry determine
measuring the noise transmission into the structure based on noise
the stiffness-controlled, resonance-controlled, mass-controlled,
reduction 共NR兲关13–17兴 or noise reduction spectrum 共NRS兲 was
and coincidence-controlled regions of the TL spectrum. Figure 1
used. The noise reduction spectrum 共NRS兲 was computed from
presents a typical TL curve for a flat panel with these regions
labeled. ‘‘Good’’ TL is often associated with materials having low the spatial average of the exterior, 具 p ext( ␻ ) 典 2 , and internal,
stiffness, high mass, and high damping 共e.g. lead兲. The panel stiff- 具 p int( ␻ ) 典 2 , pressure measurements as
ness dominates the TL behavior at very low frequencies where
there is little inertial effect. Above this region, is the resonance-
controlled region where the lowest resonance frequencie共s兲 of the
NRS⫽10 log10 冋具 p ext共 ␻ 兲典 2
具 p int共 ␻ 兲典 2
册 (3)
structure will create a peaked nature to the TL spectrum. At ap-
Although this measurement is more closely related to energy
proximately twice the frequency of the lowest resonance, the
than power, it has been deemed a good indicator of noise trans-
mass-controlled region begins and it ends at the critical coinci-
mission behavior and produces results that are similar to TL
dence frequency. Since modes radiate with greatest efficiency
measurements.
above the coincidence frequency, a dip is observed in the trans-
mission loss before resuming a mass-law type behavior.
The TL behavior of a fairing differs from that of a panel in two Mock-Scale AGS Fairing
distinct ways. First, the curvature of the fairing stiffens the struc- A mock-scale fairing was constructed to perform vibroacoustic
ture at low frequencies and produces coupling between the axial experimental analyses. The shell of the fairing forms a tapered
and circumferential modes, and second, an interior acoustic cavity cylinder that is 1.8 mm thick and having end diameters of 61 cm
is created. Therefore, the resonance-controlled region of the fair- and 38 cm and a height of 56 cm as shown in Fig. 2. Tridirectional
ing will not be governed solely by the structurally dominant reso- internal ribs with a cross section of 2.5⫻13 mm line the inner
nances, but also by the acoustically dominant cavity resonances. surface of the structure. Two sets crisscross at helix angles that
Previous studies have determined that the fairing transmission vary between 6 and 10 degrees as the diameter of the fairing
characteristics are most strongly influenced by the ‘‘acoustically decreases while the third set of 8 are vertically arranged around
dominant’’ cavity modes, while ‘‘structurally dominant’’ modes the inner circumference of the cone, as will be illustrated later. A
are seen to play a lesser role 关3–7,11兴. The terms ‘‘structurally cylindrical coordinate system (r, ␪ ,z) is applied as shown in
dominant’’ and ‘‘acoustically dominant’’ are used since the cou- Fig. 3 and will be used to describe various positions as well as
pling between the in-vacuo structural modes and the rigid-wall the structural and acoustic modes of the system. Other parameters
acoustic modes is found to be very light. As a result, separate denoted in Fig. 3 include the height, h, the minor and major
experimental and numerical modal analyses of the structure and end diameters, 2R 1 and 2R 2 , respectively, and the conical half
the cavity are performed in order to predict the mode shapes of the angle, ␣.
structure and cavity. Coupling between the structural modes,
␾ mn (rជ ), and acoustic cavity modes, ␺ o pq (rជ ), is proportional to Construction. The following subsection outlines the con-
the integral of their product over the shared surface, S, between struction of the fairing structure by the Air Force Research Labo-
them 关12兴: ratory. An automated process has been developed 关1兴 for the Ad-

Fig. 3 Cylindrical coordinate system and dimensions that de-
scribe the structure
phone measurement points were achieved with a boom that was

installed within the acoustic cavity and is capable of traversing
both axially (z苸兵 10,50其 cm), azimuthally 共␪苸兵0,360其 degrees兲,
and radially (r苸兵 0,28.5 其 cm). The average mass density of the
composite material was determined by the water displacement
method to be 1,500 kg/m3, which agrees well with calculations.
The resulting mass per unit area is 3.7 kg/m2.
Special Structural-Acoustic Frequencies. Certain
structural-acoustic frequencies are important in characterizing the
sound transmission behavior of cylindrical structures 关12,18,19兴,
including the ring frequency, f R , the critical coincidence fre-
quency, f C , and the internal cut-off frequencies, f CO . The ring
frequency occurs when the longitudinal wavelength equals the
circumference and it sets the threshold where the stiffness effects
from the curvature no longer impact the vibrational behavior of a
Fig. 2 Picture of a mock-scale AGS fairing test bed cylinder. For high frequencies ( f Ⰷ f R ), the vibration and associ-
ated acoustic radiation behave much like a flat plate. Transmission
is affected since the internal radiation efficiencies tend to be unity
above f R . Below the ring frequency, the circumferential modes of
vanced Grid-Stiffened 共AGS兲 fairings whereby the skin and ribs
the cylindrical structure will occur in pairs 共rotated 90° of each
are created using a 5-axis filament winding process that requires
other兲, which are identical for axisymmetric structures. A dip in
only a single cure cycle. The relative simplicity of the construc-
noise transmission loss is typically observed in the vicinity of f R
tion compared to sandwich style composite structures allows
due to the high number of internal coincidence phenomena that
lower costs and better quality results. Type IM7 graphite fiber
occur 关12兴. Computation of f R is accomplished with the following
preimpregnated with 977-2 ‘‘Toughened Epoxy’’ resin was used to
formula
create the shell and ribs of the fairing. A volume fraction of ap-
proximately 60– 65% resulted. The presence of voids resulted in cl
reduced stiffness and higher than expected damping in the struc- f R⫽ , (4)
2␲R
ture. Theoretical stiffnesses for the graphite-epoxy composite are:
E 11⫽170 GPa, E 22⫽10 GPa, G 12⫽5.5 GPa. where both end radii are used to compute two ring frequencies
Heavy 1.3 cm thick aluminum plates were installed at the top and c l ⫽ 冑E/ ␳ (1⫺ ␯ 2 ) is the longitudinal plate wave speed, where
and bottom of the fairing 共see Fig. 2兲 in order to approximate E is the Young’s modulus, ␳ is the density of the structural mate-
clamped, rigid end conditions. The larger plate was still found to rial, and ␯ ⫽ ␯ 12⫽ ␯ 13⫽0.245 is the Poisson’s ratio for the mate-
couple strongly to the interior acoustics so a heavy I-beam frame rial. Since the structure is anisotropic, average effective values for
was added to make it more ‘‘rigid.’’ Ultimately, the end plates the modulus and thickness were determined with FEA and used to
were modeled in the finite element analysis since they possessed compute the wave speeds. It was determined from the FEA model
resonances within the pertinent range of the fairing structural that an average isotropic modulus of E eff⫽47 GPa produces struc-
resonances and were found to mildly couple with the fairing. A tural natural frequencies that are commensurate with the experi-
Kenwood KFC-W2000 20.3 cm speaker was installed on the mentally observed resonance frequencies. Using the value for
small end of the structure as shown in Fig. 2. The speaker drives E eff , the two ring frequencies corresponding to the two end radii
the fairing cavity through a hole of 18 cm diameter and excites all (R 1 ⫽0.19 m, R 2 ⫽0.305 m) are estimated to be: f R1 ⬇4,840 Hz
acoustic modes well. A restrictor plate with a 2 cm hole allows and f R2 ⬇3,020 Hz. Data presented in a previous study 关9兴 sug-
approximate monopole excitation of the cavity with the speaker, gests that truncated conical shells do not show dips at or between
but was found to provide unsatisfactory results. Internal micro- f R1 and f R2 , as expected, but may occur above f R2 .

Table 1 Cut-off frequency bounds for various acoustic modes end radii and are given in Table 1. In systems with a significant
axial dimension, the lowest cut-off frequency also determines
Mode Shape f CO (Hz) f CO (Hz) whether higher-order duct modes can propagate down the cylin-
(pq)⫽( ␪ ,r) (R⫽0.19 m) (R⫽0.305 m)
der, which is not an issue for this structure.
共10兲 329 529
共20兲 546 876
共01兲 686 1,100
共30兲 752 1,207 Structural Modal Analysis
Many of the structural modes of the tapered fairing resemble
the modes of a uniform cylinder and thus the modes will be iden-
tified by indices (m,n) corresponding to the axial 共z兲 and trans-
The critical ‘‘coincidence frequency,’’ f C , is where external
verse 共␪兲 coordinates, respectively. The axial index represents the
coincidence between the lowest structural wavenumber k b and the
number of half wavelengths along the z dimension, and the trans-
acoustic wavenumber, k⫽ ␻ /c, occurs. Stated another way, the
verse coordinate represents the number of full waves or nodal
bending wave speed and the acoustic wave speed are equal. Struc-
diameters 关20兴 around the perimeter. Although the stiffeners
tural modes below f C are termed slow since their radiation effi-
mildly prevent true axi-symmetry, most of the circumferential
ciencies tend to be below unity 关18兴 and thus contribute less to the
modes were still predicted by the FEA to occur in pairs 共rotated
interior acoustic levels. Conversely ‘‘fast modes’’ ( f ⬎ f C ) have
by 90 degrees兲 and with proximal resonant frequencies when be-
unity radiation efficiencies and thus are more efficient at passing
low f R .
acoustic energy to the interior. As such, the mass-controlled region
of the TL spectrum typically ends around the coincidence fre- Finite Element Analysis. Multiple levels of FEA models
quency, which can be computed as were developed to aid in calculating the special structural-acoustic
c 20 冑3
frequencies and to predict the modal behavior, which is related to
f C⫽ , (5) sound transmission. The complexity of the structure is very high
t␲cl due to the taper along the cylinder axis, the anisotropic properties
where c 0 ⫽343 m/s is the speed of sound in air, and t⫽1.8 mm is of the graphite-epoxy material, and the inclusion of the internal
thickness of the shell. Thus, the approximate coincidence fre- rib structures. Further, there was also a lack of information about
quency for the fairing is f C ⬇6,230 Hz. the fiber orientation within the shell and some dynamic interaction
Internal coincidence will also occur between the axial wave- between the composite and ‘‘rigid’’ end caps was observed. A
numbers of the acoustic and structural modes. Below f R , an ex- modal homogenization of the composite structure permitted an
cellent estimate of these coincidence frequencies is provided by effective isotropic elastic modulus to be determined for computa-
the duct cut-off frequencies 关12兴, which can be computed from the tion of f C and f R in the previous section.
The most complex structural model includes the outer skin 共Fig.
acoustic radial wavenumber k rpq as
4共a兲兲, the rib stiffeners 共Fig. 4共b兲兲, and the aluminum boundary
k rpq c 0 conditions 共Figs. 4共c兲 & 共d兲兲, including four suspension points of
f C,int⬇ f CO ⫽ (6) the large plate 共see Fig. 2兲. The outer skin of the fairing was
2␲
modeled using SHELL93 共Shear deformation shell element with
where p is the number of diametral pressure nodes and q is the rotational degree of freedom 共dof兲, 8 nodes, 6 dof/node兲 and
number of concentric circular pressure nodes. Note that these cut- BEAM189 共first order shear deformation, 3-D, 3 nodes, 6 dof/
off frequencies correspond to the natural frequencies for acoustic node兲 type elements in ANSYS finite element analysis 共FEA兲 soft-
modes with no axial variation (k z ⫽0). The bounds for the first ware 关21兴. The heavy aluminum boundary plates were found to
four internal coincidence frequencies are computed using the two possess some modes in the same frequency range as the shell,
Fig. 4 Details of FEA model components

Fig. 5 First few structural mode shapes from FEA results
which is why they were included in the model. Isotropic homoge- stiffeners are actually stiffer than the isotropic model. A few
neous media was used to model the shell and rib composites, and modes occured in the end caps before the first structural mode at
a parametric study determined that an effective stiffness of 47 GPa 612 Hz, but did not cause appreciable normal displacement in the
produces similar modal behavior to the real system. Dynamic be- shell. The mode shapes and frequencies predicted by FEA are
havior from the model was evaluated by comparing the first eight given in the first two columns of Table 2 below.
natural frequencies with those measured in the experiments. The
novel beam-shell mixed FEA model proved to be computationally Experimental Tap Tests. Structural tap tests were conducted
effective and accurate. in order to experimentally determine the structural modal proper-
Figure 5 shows the first three mode shapes predicted by the ties. A PCB model 086C03 modally tuned hammer was used to
FEA model. Only the shell is shown in Fig. 5 since the end caps excite the structure while an array of 8 PCB 352B22 0.5-gram
would obstruct the cylinder modes. Both the end and side view of accelerometers measured the response at various locations. An
the lowest frequency mode 共0,3兲共repeated兲 is shown at the top imaginary measurement grid consisting of 208 points 共16 trans-
two panes of Fig. 5. The next two modes: 共1,0兲 and 共0,2兲 are verse angles 共␪兲 and 13 vertical planes 共z兲兲 was created. Three
shown in the bottom two panes of Fig. 5. Similarity is noted different force input points at z⫽ 兵 0.31,0.38,0.46其 m were used.
between the predicted fairing modes and those for a uniform cyl- Instead of measuring all 208 response points, representative arbi-
inder. Note that the lowest frequency bending mode for a cylin- trarily selected circumferences and vertical slices were measured
drical shell does not correspond to the lowest indices: (m,n) along the fairing in order to try to corroborate the FEA results. A
⫽(0,1). It is interesting to observe that the low-frequency modes curve-fit was performed on several of the frequency response
of the structure are predicted to respond globally as a shell, rather functions 共FRFs兲 between various inputs and response locations in
than a collection of smaller panels created by the rib stiffeners. order to extract global resonant frequency and damping character-
This may or may not be true, since the unidirectional composite istics to validate the FEA models. The complexity of the system
Table 2 Numerical and experimental structural modal properties.
Predicted Measured Measured

Predicted Mode Resonant Freq. Resonant Freq. %Error Between Damping
Shape (m,n)⫽(z, ␪ ) ␻ r,FEA 共Hz) ␻ r,Exp 共Hz) ␻ r,FEA & ␻ r,Exp ␨ 共%兲
共0,3兲 615 612 0.5 0.8
共0,3兲 618 627 ⫺1.2 0.7
共1,0兲 702 678 3.2 0.6
共0,2兲 761 699 8.1 0.8
共0,4兲 776 775 0.2 1.1
共0,2兲 818 787 3.8 1.0
共0,4兲 842 794 4.5 0.5

Fig. 6 Representative FRF of structure
precluded fitting experimental data for mode shapes. Columns 3 Finite Element Analysis. The rigid-wall acoustic cavity
and 5 in Table 2 contain the resonant frequencies and damping model of the fairing was created using a uniform tapered cylindri-
共viscous model兲, respectively, obtained from the curve-fit of the cal boundary that neglected the internal ribs, since they would
FRFs. Column 4 of this table shows the error between predicted insignificantly affect the acoustic modal properties of the system.
and measured resonant frequencies. Errors range from 0.2– 8.1%, FLUID30 elements 共ANSYS 5.7兲 with properties for air at stan-
which is considered good for a structure of this complexity. The dard conditions 共density of 1.21 kg/m3, sound speed of 343 m/s,
damping values 共0.8 –1.1%兲 are perhaps higher than expected for and static pressure of 100 kPa兲 were used to model the cavity.
this type of material due to the existence of voids in the matrix. Mesh convergence was performed and the total resulting number
Figure 6 depicts a representative FRF and curve-fit taken at mea- of elements and degrees of freedom were 1,244 and 4,976, respec-
surement point (z, ␪ ,r)⫽(39.5 cm, 56.25°, 24.4 cm) and drive tively. The four degrees of freedom for each node were three
point (z, ␪ ,r)⫽(18.9 cm, 56.25°, 24.4 cm). Some small resonant Cartesian particle velocity components (V x , V y , and V z ) and the
responses are observed around 280, 400, and 470 Hz. The FEA pressure, P. A second FEA model included the speaker that was
predicted these modes to be corner flapping modes of the large used for the experimental modal analysis. It was modeled as a
end cap and two weak accordion-like modes of the structure, re- rigid piston 共moving mass兲 on a spring-damper 共spider兲 support.
spectively. It is the modes with large normal displacement which Since both acoustic FEA models yielded similar results, the sim-
couple well to sound, and thus these modes were not of interest pler model without the speaker was adopted for the study. This
for this study. Note from Table 2 that the lowest structural fre- model is also consistent with the NRS measurements setup, which
quency is 612 Hz, meaning that the resonance-controlled region is did not contain the attached speaker. Given the high degree of
expected to end around 2⫻612⫽1,224 Hz. Thus, the mass con- ideal nature in the model, the predicted acoustic modes are given
trolled region is expected to extend from 1,224 Hz to the critical a high degree of confidence, particularly at low frequency where
frequency of f C ⫽6,230 Hz. models are known to generally perform well. Figure 7 shows the
first six predicted acoustic mode shapes. The axial-symmetric
mode shapes are repeated 共rotated by 90 deg兲, but the repeats are
Acoustic Modal Analysis omitted for brevity. The experimental modal analysis is presented
in the next section.
Similar to the fairing shell, the modes of the internal tapered
acoustic cavity conform well to those for a uniform cylindrical Experimental Results. The experimental acoustic modal
cavity with closed ends. These modes will also be referenced with analysis was performed by exciting the fairing with the attached
respect to the cylindrical coordinate system given by Fig. 3 and speaker at z⫽r⫽0 共shown in bottom of Fig. 2兲. A total of 260
the indices (o,p,q) or (opq). The first modal index, o, represents acoustic measurements were taken with the internal microphones
the number of half wavelengths along the z-axis, the second, p, positioned at combinations of the following coordinates: r
represents the number of nodal diameters occuring along the ⫽ 兵 3.81,7.62,11.4,15.2其 cm, ␪⫽兵15,30,45,60,75,90,105,120,135,
transverse ␪ coordinate, and q represents the number of radial 150,165,180其 degrees, and across vertical planes at z
nodes along the r direction in Fig. 3. The acoustic mode shapes ⫽ 兵 10,20,30,40,50其 cm. The FRF was computed between the input
are important for three reasons. First, the modal pressure distribu- to the speaker and each of the four array microphones and thus
tions could be examined for ‘‘problem modes,’’ when considering included the speaker dynamics, which are much lower frequency
the payload location. More importantly, the predicted mode than the first acoustic mode of the fairing cavity. Figure 8 shows a
shapes can be used to estimate the cut off frequencies f CO for the typical measured FRF between the speaker and an internal micro-
truncated cone, which will be bounded by the predictions given in phone located at (z, ␪ ,r)⫽(19.5 cm, 337.5 deg, 23 cm兲. Note
Table 1. Because internal coincidence occurs at these frequencies, also from the figure that the modes and speaker dynamics have
worse transmission loss would be expected, which will be verified been identified. The dominance by the fundamental 共100兲 mode at
in measurements section. Also, the structural-acoustic coupling 301 Hz is typical of most all of the measured FRFs. However, the
given by Eq. 共2兲 indicates that the relative shape of the in-vacuo axial zero modes 共010兲, 共020兲, and 共001兲 at 385, 613, and 788 Hz,
structural modes and rigid-wall acoustic modes will influence the respectively, will be shown in the next section to also be very
noise transmission. However, this relationship is of limited value important for sound transmission. These three modes are used to
due to the complexity of the coupling. estimate the internal coincidence frequencies, the bounds of which

Fig. 7 First few acoustic mode shapes from FEA results
were given in the first three rows of Table 1. Interestingly, al- are within 0.1–5.1% in magnitude. Repeated modes that have
though they are significant for sound transmission, many of these nearly identical natural frequencies are indicated 共‘‘⫻2’’兲 in col-
modes appeared insignificant in the acoustic FRFs. The difference umn 1 of Table 3.
results from the direct excitation by the speaker for the FRF ver- Finally, note that when comparing the measured resonant fre-
sus excitation via transmission for the NRS measurement. Previ- quencies of the structure and acoustic cavity in Tables 2 and 3, in
ous studies focused mainly on the 共100兲 mode as the primary general the same modal frequencies do not show up in both tables,
contributor to noise transmission 关6,7兴. indicating that indeed the structural acoustic coupling is light.
A curve-fit was performed on an arbitrary selection of 32 of the There are however, two frequencies in the vicinity of 612 and 787
260 FRFs so that modal properties could be extracted. Natural Hz that are given in the tables that appear for both the structure
frequencies and damping ratios for the first eight modes are given and the acoustic cavity. Although internal coincidence is also pre-
in columns 3 and 5, respectively, of Table 3. The predicted mode dicted to occur for the acoustic modes at these two frequencies, it
shape and natural frequencies from the FEA analysis are given in is not between the structural modes that have nearly identical
columns 1 and 2 of the same table, along with the error between frequency, since internal coincidence can only occur between
predicted and measured natural frequencies in column 4, which modes of the same transverse wavenumber, k ␪ , which is related
Fig. 8 Representative FRF between speaker and cavity microphone

Table 3 Numerical and experimental acoustic modal properties.
Predicted Measured Measured

Predicted Mode Resonant Freq. Resonant Freq. % Error Between Damping
Shape (opq)⫽(z ␪ r) ␻ r,FEA 共Hz) ␻ r,Exp 共Hz) ␻ r,FEA & ␻ r,Exp ␨ 共%兲
共100兲 313 301 4.0 0.29
共010兲⫻2 383 385 ⫺0.5 0.23
共110兲⫻2 550 547 0.5 0.18
共020兲⫻2 617 583 5.6 0.22
⬇共020兲 622 613 1.4 0.27
共210兲⫻2 765 735 3.8 0.14
共001兲 787 788 ⫺0.1 0.20
共120兲⫻2 815 840 ⫺3.1 0.33
to modal index n for the structure and p for the acoustic cavity. A convergence study on the spatial sampling was performed by
When comparing the modal indices n and p from the first column computing NRS using 1, 2, 4, 8, 16, and 20 sets of measurement
of Tables 2 and 3, one observes that (n⫽3⫽p⫽2) at 612 Hz and points across the surfaces of the shell. The NRS was mostly con-
(n⫽2⫽p⫽1) at 787 Hz. verged by 8 points, and the difference between 16 and 20 points
was indistinguishable. Thus, 20 was chosen as a suitable number
Measurement of Sound Transmission of points. Effects of angular and vertical position were also exam-
NRS measurements are based upon a spatial average of mean- ined. Little variation in NRS was noted between angles when
square pressure measurements from the inside and outside of the averaging the 5 vertical measurement heights together. Variation
structure as given by Eq. 共3兲. To simulate a diffuse field, four within the vertical measurement planes was more pronounced,
speakers were arranged around the structure and driven with in- with the sharpest dips in NRS noted near the end caps, which is
dependent white noise sources. Note that a previous work reported consistent with modal pressure maxima and edge radiation effects.
that directional effects in the radiation and reception patterns do Figure 10 shows the measured NRS for the structure using a
not occur in finite cylinders 关13兴. A wide bandwidth 共0–20 kHz兲 linear frequency axis from 0–1000 Hz. Note that vertical dashed
for the noise sources was desired due to the small scale of the and dotted lines have been added at the frequencies of the acoustic
structure. The speakers and the structure were suspended 0.7 m and structural resonances, respectively, which were obtained from
above the floor and the room volume was approximately 300 m3. the structural and acoustic modal analyses. In the low-frequency
An external microphone was traversed to 20 different measure- stiffness-dominated region of the NRS 共100–250 Hz兲, the reduc-
ment locations which were 1.5 cm from the surface of the shell. tion is seen to range from 34 –52 dB, with most frequencies above
The 20 locations spanned across five different vertical heights: h 40 dB. The dip at very low frequency 共40 Hz兲 is attributed to the
⫽ 兵 10,20,30,40,50其 cm, and four different transverse angles: Helmholtz frequency for the enclosure, which occurs due to leak-
␪⫽兵0,90,180,270其 deg. An internal microphone boom was in- age. Two dips likely occured because an access door to the struc-
stalled 1.5 cm from the edge of the ribs. This microphone re- ture had to be opened and closed 20 times to arrange the internal
corded 20 internal measurements which were at the same height microphone at each measurement location, which may have cre-
and angles as the exterior microphone. Figure 9 depicts a top view ated nonuniform leakage each time. Commensurate dips in the
of the measurement setup. Signals were generated and measured NRS are noted at most all of the acoustic resonant frequencies
using a Siglab MC20-84 dynamic signal analyzer. Bruel and Kjaer 共e.g. 301 Hz (opq)⫽(100), 385 Hz 共010兲, 547 Hz 共110兲, 613 Hz
4190 microphones with B&K 2669 preamplifiers were used for 共020兲, 735 Hz 共210兲, 788 Hz 共001兲, and 840 Hz 共120兲. A Dip in
the acoustic measurements. Marchand PS-24 power amplifiers the NRS is not noted for the acoustic mode at 583 Hz. Most of the
drove the four KLH-9912 speakers to generate the external noise structurally dominant resonances are seen not to influence the
field. Autospectrum measurements from the inside and outside NRS, presumably due to the relatively high level of damping for
were post processed and linearly averaged to determine the NRS. this composite structure. The exceptions are at 612, 787, 794, 845,
Fig. 9 Noise transmission measurement setup

Fig. 10 Measured noise reduction spectrum showing acoustic and struc-
tural modes
and 950 Hz. Each of these frequencies also coincides with an cidence as well. Again, as noted when examining Fig. 8, these
acoustic mode, so it is hard to decipher whether the acoustic, modes do not have a particularly strong response in this FRF, but
structural or both contribute to the NRS. There are three frequen- transmit energy well due to coincidence. The 共010兲 mode was also
cies 共612, 787–794, and 845 Hz兲 where the noise transmission is observed to be a weak mode in most all of the acoustic FRFs, but
actually negative, indicating an amplification of the noise by the is seen to have a significant dip in the NRS at that frequency 共385
structure. Modes at 612 and 787 Hz have already been associated Hz兲, again attesting to the importance of internal coincidence for
with internal coincidence frequencies. The mode at 845 Hz is NRS. The 794 Hz dip appears to be due to the structural mode at
assumed to be associated with the 815 Hz 共120兲 mode from the that frequency.
FEA, but is also near the 共030兲 mode with a predicted natural In Fig. 11, the NRS is replotted with a logarithmic frequency
frequency of 840 Hz. Note that since the accuracy of FEA models axis which extends from 100 to 20,000 Hz. The approximate
deteriorate at higher frequency, the measured 840 Hz frequency stiffness-, resonance-, and coincidence-controlled regions of the
may actually be the 共030兲 mode, which would have internal coin- NRS are indicated in the figure, as well as the probable contribu-
Fig. 11 Measured noise reduction spectrum „NRS… showing various

regions of transmission loss

tion by the ring frequency. Referring back to the section on trans- also impacted the sound transmission, as noted by the wide dip in
mission loss, the stiffness-controlled region will begin at 0 Hz and the NRS plot from 6 –12 kHz. However, this dip is found to be
end at the first resonance, which is an acoustic cavity resonance at inconsequential compared to the singularities found at some of the
300 Hz. The ‘‘resonance-controlled’’ region was predicted to con- acoustic mode frequencies. Future transmission control will pro-
tinue through approximately twice the first structural resonance vide a significant challenge due to the payload and weight require-
frequency or 共2⫻612⫽1,224 Hz兲. Judging from the figure, the ments for the launch vehicles. Since the NRS is dominated by the
dips in NRS from the resonances are noted to stop around 1,500 acoustic modes, control should focus on adding damping to the
Hz and a general upward trend begins until a wide dip begins at acoustic modes or focus on disrupting the internal coincidence
6,000 Hz. The slope in the mass-controlled region is 28 dB per phenomena. Once the effect of the acoustic modes is sufficiently
decade rather than the classical 20 dB/decade mass law. Since f C ameliorated, additional marginal improvements could also be ex-
was predicted to be 6,230 Hz, this dip is attributed to external pected by increasing the damping of certain structural modes.
coincidence. The width of the dip is rather large 共6,000–12,000
Hz兲, which may be attributed to effects of the stiffeners and/or Acknowledgments
ring frequency effects 关12兴. Regardless, the dip is inconsequential
when compared to the levels at the singularities in the NRS that The work was sponsored by the Air Force Research Laboratory
occur at internal coincidence frequencies and in the region around Space Vehicles Directorate 共AFRL/VS兲. The POC for this effort is
where the resonance- and mass- controlled regions meet. Dr. Steven A. Lane. The authors also acknowledge ANSYS, Inc.
Note that to improve the noise transmission behavior will for the John Swanson Fellowship. Thanks also to David J. Be-
largely require the acoustically dominant modes and/or internal lasco, Jr. for help with data collection and reduction.
coincidence phenomena to be controlled. Internal absorptive treat-
ments are limited in rocket fairings due to the volume require- References
ments of the payload. The NRS could perhaps also be improved at 关1兴 Huybrechts, Steven, and Tsai, Stephen W., 1996, ‘‘Analysis and Behavior of
612, 787, 794, 845, and 950 Hz by adding damping treatments or Grid Structures,’’ Comp. Sci. Technol. 56共9兲, pp. 1001–1015.
control that targets these structural modes. 关2兴 George, Tommy J., Herman Shen, M. H., Huybrechts, Steven M., Meink, Troy
E., and Wegner, Peter M., 2001, ‘‘Optimal Design of Composite Chambercore
Structures,’’ Compos. Struct., 52共3– 4兲, pp. 277–286, May–June.
Conclusions 关3兴 Griffin, Steven, Denoyer, Keith K., and Das, Alok, 1999, ‘‘Passive Vibroacous-
The noise transmission behavior for a novel, advanced grid- tic Isolation for Payload Containers,’’ J. Intell. Mater. Syst. Struct., 10共0083-
05兲, pp. 83– 87.
stiffened 共AGS兲 composite structure has been investigated. This 关4兴 Griffin, Steven, Hansen, Collin, and Cazzolato, Ben, 1999, ‘‘Feasibility of
material is being considered by the US Air Force for constructing Feedback Control of Transmitted Sound into a Launch Vehicle Fairing Using
expendable space launch vehicles. A combined numerical/ Structural Sensing and Proof Mass Actuators,’’ AIAA 99共1529兲, pp. 2592–
experimental approach was used to examine the physical mecha- 2602.
关5兴 Griffin, Steven, Hansen, Collin, and Cazzolato, Ben, 1999, ‘‘Feedback Control
nisms of noise transmission by the structure. Separate analyses of Structurally Radiated Sound into Enclosed Spaces Using Structural Sens-
were performed on the structure and the acoustic cavity since the ing,’’ JASA, 106共5兲, pp. 2621–2628.
coupling between the two was found to be very light. Finite ele- 关6兴 Griffin, Steven, Denoyer, Keith K., and Sciuli, Dino, 1998, ‘‘Hybrid Structural/
ment analysis permitted a smeared or average value for the com- Acoustic Control of a Sub-Scale Payload Fairing,’’ SPIE Conference on Smart
Structures and Integrated Systems, Vol. 3329 of 0277-786X, pp. 237–243.
posite elastic modulus to be determined, which in turn was used to 关7兴 Griffin, Steven, and Denoyer, Keith K., 1998, ‘‘Experimental Sensor and Ac-
estimate the ring and critical frequencies for the structure, since tuator Location Procedure for Control of Dynamically Complex Smart Struc-
they influence transmission loss. Mode shapes were also predicted tures,’’ SPIE Conference on Smart Structures and Integrated Systems, Vol.
by FEA, which allow the modal pressure distributions to be ex- 3329 of 0277-786X, pp. 717–723.
关8兴 Lane, Steven A., Kemp, J. D., Griffin, Steven, and Clark, R. L., 2001, ‘‘Active
amined and which also permit investigation of the internal coin- Acoustic Control of a Rocket Fairing Using Spatially Weighted Transducer
cidence phenomena. Experimental modal analyses on the structure Arrays,’’ AIAA J., 38共1兲, 112–119.
and cavity were also performed. Tap tests at various locations 关9兴 Neple, Pascale, el Hafidi, Ali, and Lesueur, Claude, 2001, ‘‘On Sound Trans-
around the structure provided frequency response functions mission Through a Truncated Isotropic Conical Shell: Experimental and Pre-
liminary Numerical Results,’’ Internoise 2001, The Hague, The Netherlands,
共FRFs兲 that were curve-fit to obtain resonant frequencies and August 27–30.
damping ratios. A speaker was used to drive the acoustic cavity 关10兴 Vipperman, Jeffrey S., Li, Deyu, Avdeev, Ilya, and Lane, Steven A., 2001,
with a white input while the pressure response was measured in- ‘‘Characterization and Control of Sound Radiation in a Complex Fairing Struc-
side the cavity. Again, the measured FRFs were fitted to obtain ture,’’ Internoise 2001, The Hague, The Netherlands, August 27–30.
关11兴 Denoyer, Keith K., Griffin, Steven, and Das, Alok, 1998, ‘‘Passive Vibroacous-
resonant frequencies and damping ratios for the acoustic modes. tic Isolation for Reusable Launch Vehicle Payload Containers,’’ AIAA/ASME/
Comparison of measured and predicted natural frequencies per- ASCE/AHS/ASC Structures, Structural Dynamics & Materials Conference,
mitted corroboration of the FEA models. The resonances for the AIAA-98-1977, pp. 2248 –2256, Longbeach, CA, April 20–23.
structure and especially the acoustic cavity were found to signifi- 关12兴 Fahy, F., 1985, Sound and Structural Vibration, Academic Press, Inc., New
York.
cantly affect the noise transmission into the structure. 关13兴 White, Pritchard H., 1966, ‘‘Sound Transmission Through a Finite, Closed
Finally, direct measurement of sound transmission was con- Cylindrical Shell,’’ JASA, 50共5兲, pp. 1124 –1130.
ducted based upon the noise reduction spectrum 共NRS兲, which is 关14兴 Manning, J. E., and Maidanik, G., 1964, ‘‘Radiation Properties of Cylindrical
computed from the ratio of the spatial averages of acoustic pres- Shells,’’ JASA, 36共9兲, pp. 1691–1698.
关15兴 Cheng, L., 1994, ‘‘Fluid-structural Coupling of a Plate-ended Cylindrical
sures measured along the internal and external surfaces of the Shell: Vibration and Internal Sound Field,’’ JSV, 174, pp. 641– 654.
shell. Classic stiffness, resonance, and coincidence controlled re- 关16兴 Koval, L. R., 1976, ‘‘On Sound Transmission into a Thin Cylindrical Shell
gions were identified, which agreed very well with the predic- Under Flight Conditions,’’ JSV, 48, pp. 265–275.
tions. The NRS was dominated by the sharp dips that were noted 关17兴 Koval, L. R., 1978, ‘‘Effects of Cavity Resonances on Sound Transmission
into a Thin Cylindrical Shell,’’ JSV, 59, pp. 23–33.
at most of the acoustic resonant frequencies. Internal coincidence 关18兴 Blake, William K., 1986, Mechanics of Flow-Induced Sound and Vibration,
was found to play a significant role as well, since negative NRS Vol. 1, General Concepts and Elementary Sources, Academic Press, Inc., Lon-
values 共amplification兲 was noted at these frequencies. The internal don.
cut-off frequencies are good estimates of the internal coincidence 关19兴 Blake, William K., 1986 Mechanics of Flow-Induced Sound and Vibration,
Vol. 2, Complex Flow-Structure Interactions. Academic Press, Inc., London.
frequencies below the ring frequency. These frequencies were de- 关20兴 Blevins, Robert D., 1979, Formulas for Natural Frequency and Mode Shape,
termined from the FEA results by identifying the axial-zero modes Van Nostrand Reinhold Co., New York.
(k z ⫽0). External coincidence and possibly ring frequency effects 关21兴 ANSYS, Inc., 2001, Canonsburg, PA 15317.

Optimal Stiffener Design
for Interior Sound Reduction
Using a Topology Optimization
Jianhui Luo
Based Approach
Hae Chang Gea
A topology optimization based approach is proposed to study the optimal configuration of
Department of Mechanical and Aerospace stiffeners for the interior sound reduction. Since our design target is aimed at reducing the
Engineering, low frequency noise, a coupled acoustic-structural conservative system without damping
Rutgers, The State University of New Jersey, effect is considered. Modal analysis method is used to evaluate the interior sound level for
Piscataway, NJ 08855 this coupled system. To formulate the topology optimization problem, a recently intro-
duced Microstructure-based Design Domain Method (MDDM) is employed. Using the
MDDM, the optimal stiffener configurations problem is treated as a material distribution
problem and sensitivity analysis of the coupled system is derived analytically. The norm of
acoustic excitation is used as the indicator of the interior sound level. The optimal stiff-
ener design is obtained by solving this topology optimization problem using a sequential
convex approximation method. Examples of acoustic box under single frequency excita-
tion and a band of low frequency excitations are presented and discussed.
关DOI: 10.1115/1.1569512兴
1 Introduction introduces the modal analysis approach to calculating the fre-

quency response for the coupled acoustic-structural system; Sec-
The reduction of low frequency noise is of great interest in
tion 3 discusses the sensitivity analysis of frequency response for
designing transportation vehicles because the 20-200 Hz low fre-
this system. In Section 4, topology optimization problem is for-
quency noise has an important influence on the product satisfac- mulated and the solution procedure of identifying the optimal con-
tory quality. In the automobile passenger compartment, the low figurations of stiffeners for the reduction of the interior sound
frequency noise can be generated by engine vibration or induced level is described. In Section 5, four numerical examples of an
by road roughness; in the aircraft cabin, the low frequency noise acoustic box under external excitations are presented. Conclusion
can be originated from engines, propellers or aerodynamic forces. remarks and future work are discussed in the final section.
Although active vibration control is a fast growing research field,
passive vibration controls from structural optimization techniques
are often used in practice due to their low implementation cost.
Studies of noise reduction by structural modification are found in 2 Frequency Response of Coupled Systems
the literature. Hagiwara et al. 关1兴 investigated the reduction of In this section, frequency response of coupled acoustic-
vehicle interior noise with shell thickness redistribution using the structural systems is derived from the modal analysis method.
sensitivity information semi-analytically. Huff, Jr. and Bernhard Consider that a coupled acoustic-structural system consists of an
关2兴 used a parametric shape optimization method in the reduction enclosure surrounded by an elastic body and the elastic body is
of the sound pressure level. Wodtke and Koopmann 关3兴, Constants subjected to both force and displacement boundary conditions as
and Belegundu 关4兴 introduced the structural modifications by plac- shown in Fig. 1. Using the finite element method to discretize the
ing optimally sized point masses in order to minimize the radiated coupled system yields the following equations 关8兴
冋册再冎冋册再冎再冎
sound power of vibrating structures.
In this paper, a topology optimization based approach is pro- M ss 0 ü e K ss K sa ue fs
⫹ ⫽ (1)
posed to study the optimal configuration of stiffeners for interior M as M aa p̈ e 0 K aa pe 0
sound reduction. Since our design target is aimed at reducing the
where the upper equation refers to the structural system and the
low frequency noise, a coupled acoustic-structural conservative
lower equation refers to the acoustic system. In these equations,
system without damping effect is considered. Modal analysis
u e is the vector representing displacement components at the grid
method is used to evaluate the interior sound level for this coupled
points of the structural finite element model, p e is the vector rep-
system 关5兴. To formulate the topology optimization problem a
resenting sound pressures at the grid points of the acoustic finite
recently introduced Microstructure-based Design Domain Method
element model. f s represents the external forces applied to the
共MDDM兲关6兴 is employed. Using the MDDM, the optimal stiff- structure, such as mechanical excitations. M ss and K ss are struc-
ener configurations problem is treated as a material distribution tural mass and stiffness matrices, M aa and K aa are acoustic mass
problem and sensitivity analysis of the coupled system is derived and stiffness matrices. M as and K sa are acoustic-structural cou-
analytically. The optimal stiffener design is obtained by solving
pling matrices, and they have the property of K sa ⫽⫺M as T
. The
this topology optimization problem using a sequential convex ap-
structural and acoustic equations of motion are coupled through
proximation method called Generalized Convex Approximation
the matrix M as which transforms the structural accelerations to
关7兴.
acoustic excitations of the interior cavity, and through the matrix
The remainder of the paper is organized as follows: Section 2 K sa which transforms the acoustic pressures to loads acting on the
structure.
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received June 1997; If the external force has a harmonic form of f s e i ␻ t , then the
Revised October 2002. Associate Editor: R. L. Clark. structural and acoustic responses can be expressed as u e e i ␻ t and
where ⌿⫽ 兵 ⌿ s ⌿ a 其 T denotes the right eigenvector matrix of the
coupled system, ␰ is the coefficient matrix.
Inserting Eq. 共7兲 into the governing equations of Eq. 共2兲 and
premultiplying both sides by left eigenvector matrix ⌿̄ T
⫽ 兵 ⌿̄ sT ⌿̄ Ta 其 , the use of M-orthonormal condition leads to
共关 ⌳ 兴 ⫺ ␻ 2 关 I 兴兲 ␰ ⫽⌿̄ T 再冎 fs
0
⫽⌿̄ sT f s (8)
Denote ⌿̄ sT f s ⫽p, the coefficient matrix ␰ can be computed in the

following form
pi
␰ i⫽ (9)
␭ i⫺ ␻ 2
where ␰ i , p i and ␭ i are the i th component of ␰, p and ⌳. Com-
bining Eq. 共9兲 with Eq. 共7兲, the response of the coupled acoustic-
structural system is recovered.
Fig. 1 A coupled acoustic-structural system
3 Sensitivity Analysis of Coupled Systems

p e e i ␻ t , where ␻ is the excitation frequency. Introducing these ex- When an optimization problem is solved by mathematical pro-
pressions to Eq. 共1兲 yields the frequency response equation as gramming algorithms, we must determine the effect resulting from
冋册再冎再冎
K ss ⫺ ␻ 2 M ss K sa a small perturbation in the current design on the objective and
ue fs constraint functions. This is known as the sensitivity analysis. In
⫽ (2)
⫺ ␻ 2 M as K aa ⫺ ␻ 2 M aa pe 0 this study, frequency response sensitivities are the derivatives of
There are two different approaches to computing frequency re- structural response and the sound pressure with respect to the
sponse in Eq. 共2兲: the direct method and the modal analysis structural modification design variables. With the correct fre-
method. The direct method is a rather straightforward approach quency response sensitivity information, one can use various op-
that solves Eq. 共2兲 directly but it is very computationally expen- timization methods to arrive at a modified structure with reduced
sive. On the other hand, the modal analysis method is extremely interior sound level. The frequency response sensitivity,
attractive because both structural and acoustic modal density are 兵 u e⬘ ,p ⬘e 其 T , can be derived by differentiating Eq. 共2兲 as
relatively low under the low excitation frequencies. Therefore, the
modal analysis method is adopted in this study.
Since the mass and stiffness matrices are unsymmetric in the
冋 K ss ⫺ ␻ 2 M ss
⫺ ␻ M as
2
K sa
K aa ⫺ ␻ M aa2 册再冎冋
u ⬘e
p ⬘e
⫽⫺
⬘ ⫺ ␻ 2 M ss
K ss
0
⬘ 0
0
册再冎
uc
pe
coupled acoustic-structural system, the left eigenvectors are not (10)
the same as the right ones. Using the modal analysis method, the ⬘ and M ss
where K ss ⬘ are the derivatives of the structural stiffness
right eigenvalue problem can be formulated as and mass matrices with respect to the design variable, respec-
冋 K ss
0
K sa
K aa
册再冎冋
⌿s
⌿a
⫽关⌳兴
M ss
M as
0
M aa
册再冎
⌿s
⌿a
(3)
tively.
can be rewritten as
⬘ ⫺ ␻ 2 M ss
If we define a pseudo load g s ⫽⫺ 关 K ss ⬘ 兴 u e , Eq. 共10兲
冋册再冎再冎
where ⌳ is the eigenvalue matrix; ⌿ s and ⌿ a are the right eigen-
vector matrices corresponding to the structural and acoustic fields K ss ⫺ ␻ 2 M ss K sa u e⬘ gs
⫽ (11)
respectively. Similarly, the left eigenvalue problem is represented ⫺ ␻ M as
2
K aa ⫺ ␻ 2 M aa p ⬘e 0
as
Eq. 共11兲 has the same form as the frequency response governing
兵 ⌿̄ sT ⌿̄ Ta 其冋 K ss
0
K sa
K aa
册 ⫽ 关 ⌳ 兴兵 ⌿̄ sT ⌿̄ Ta 其冋 M ss
M as
0
M aa
册 (4)
equations in Eq. 共2兲 except the excitation force has been changed
from f s to g s . Therefore, the solution of Eq. 共11兲 can be obtained
in the same way as shown in the previous section.
where ⌿̄ s and ⌿̄ a denote the left eigenvector matrices of the Assume response sensitivities can be decomposed as,
structural and acoustic fields. The M-orthonormal condition of the 兵 u ⬘e ,p ⬘e 其 T ⫽⌿ ␩ with ␩ being the undetermined coefficient matrix.
coupled system is stated as follows: Put this expansion into Eq. 共11兲 and premultiply both sides by
兵 ⌿̄ sT ⌿̄ Ta 其冋 M ss
M as
0
M aa
册再冎⌿s
⌿a
⫽关I兴 (5)
⌿̄ T , we have
共关 ⌳ 兴 ⫺ ␻ 2 关 I 兴兲 ␩ ⫽⌿̄ T 再冎 gs
0
⫽⌿̄ sT g s (12)
For large coupled acoustic-structural systems, it is impractical
to solve Eq. 共3兲 and Eq. 共4兲 directly because of the high compu- Denote ⌿̄ sT g s ⫽q, then coefficient matrix can be determined in the
tational cost. Luo and Gea 关5兴 proposed a symmetrization ap- indices form
proach to effectively extract eigenmodes with less computational
time and storage. They also proved that the right eigenvectors and qi
␩ i⫽ (13)
the left eigenvectors of a coupled system can be related as ␭ i⫺ ␻ 2
再冎再冎⌿̄ s
⌿̄ a
⫽
⌳⌿ s
⌿a
(6)
With Eq. 共13兲, the frequency response sensitivity, 兵 u ⬘c ,p ⬘e 其 , can be
calculated easily.
Suppose that the forced frequency responses are expanded by

the right eigenvectors of the coupled system as
4 Solution Procedures
再冎
In this section, the solution procedures of identifying the opti-
ue mal configuration of stiffeners placement for the reduction of the
⫽⌿ ␰ (7)
pe interior sound level are discussed.

The overall interior sound level for the acoustic field is evalu- ume fraction, c (1i ) , is treated as the design variables in the stiffener
ated as topology optimization formulation: as c 1(i ) ⫽0 no stiffener is re-
1
m quired and c (1i ) ⫽1 stiffener exists. Using these relation, K ss ⬘ and
S PL⫽
m 兺共 10 log
i⫽1
10关 p i /p 0 兴 2 兲 (14) ⬘ under the pseudo load can be directly calculated.
M ss
Generally, the total allowable weight of the structure is consid-
where p i denotes the sound pressure at the i th node, m is the total ered as a design constraint that cannot exceed a prescribed
number of nodal points in the acoustic field, p 0 is a reference amount. The optimization problem of minimizing acoustic excita-
pressure, normally, chosen as 2⫻10⫺5 Pa. tion is stated as,
Although SPL is a direct indicator of sound noise, it inherits Minimize 储 ⫺ ␻ 2 M as u e 储 (21)
strong localized effects from individual nodal pressure. Therefore,
N
if this function is used as the objective function in the optimiza-
tion process, search algorithm will exhibit unstable oscillations
and consequently have difficulty to converge. Consider the fact
Subject To: 兺 ␳ v 共 1⫹c
i⫽1
i
共i兲
1 兲 ⭐W̄ (22)
that the level of interior sound is fully determined by the acoustic where ␳ denotes the material density, v i is the volume of ith
excitation, that can be measured by the magnitude of ⫺ ␻ 2 M as u e element, c i1 is the design variable as described previously, N is the
from Eq. 共2兲 for a given excitation frequency. The minimization of number of elements, and W̄ is the upper limit of total weight.
the norm of ⫺ ␻ 2 M as u e can lead to the reduction of SPL. There- In order to search for the optimal stiffener topology, the number
fore, the minimization of the acoustic excitation is chosen as the of design variables is related to the number of finite elements used
objective function. In our numerical examples, iteration histories in the structural model. Therefore, the optimizer used here must
of SPL are also listed for the purpose of comparing the sound be very efficient in handling a large number of design variables. In
pressure level reductions from the optimization process. this paper, the Generalized Convex Approximation 共GCA兲 method
When a structure is subjected to a band of frequency excitations introduced by Chickermane and Gea 关7兴 is used to formulate and
关 ␻ 0 , ␻ 1 兴 , the objective function can be modified as solve the topology optimization problem. In the GCA, functions
Minimize
1
␻ 1⫺ ␻ 0 冕␻0
␻1
储 ⫺ ␻ 2 M as u e 共 ␻ 兲储 d ␻ (15)
are approximated as the sum of a series of separable functions of
the design variables as
n n
And, the overall interior sound level from Eq. 共14兲 can be defined
as
f 共 x 兲⬵ f 共 xk兲⫺ 兺 b 共 x ⫺d 兲 ⫹ 兺 b 共 x ⫺d 兲
i
i
k
i i
ri
i
i i i
ri
(23)
SPL⫽
1
␻ 1⫺ ␻ 0 冕 ␻1
␻0
冉 1
m 兺
m
i⫽1
冊
共 10 log10关 p i 共 ␻ 兲 /p 0 兴 2 兲 d ␻
where b i , d i and r i are a set of approximation parameters to be
determined, and f (x k ) is the value of the original function at the
kth design. Function values and first order sensitivity information
(16) from the current and previous design is utilized to determine the
Recently, the stiffener layout optimization problem is tackled values of the approximation parameters. The approximation prob-
by material distribution formulations, in which a given amount of lem is solved iteratively using mathematical programming to gen-
‘‘artificial’’ stiffener material is dispersed optimally to the base erate the next design. This process continues until a satisfactory
structure for the best structure performance. The material model design is reached.
for the combination of base material and ‘‘artificial’’ stiffener ma-
terial is often borrowed from various theories of composite mate- 5 Numerical Examples
rials. This method is called the topology optimization because the Optimal stiffener designs of an aluminum alloy 共1100-H14兲 box
topology of stiffener design is to be optimized. Diaz and Kikuchi using the proposed topology optimization based method are pre-
关9兴 used this technique on solving stiffener layout design optimi- sented in this section. The aluminum box of 30 cm⫻40 cm
zation under natural frequency and they applied the homogeniza- ⫻50 cm is fixed at four bottom corners and under various external
tion theory 关10兴 to evaluate the material properties of the compos- excitations. The aluminum plate has thickness 0.2 cm, Poisson’s
ite consisting of the base and stiffener material. In this paper, a ratio 0.33, Young’s modulus 6.9⫻1010 Pa, density 2.7
Microstructure-based Design Domain Method 共MDDM兲 intro- ⫻103 kg/m3 and the box without any stiffeners weights 5.076 Kg.
duced by Gea 关6兴 is applied to model the add-on stiffener. In the The ‘‘artificial’’ stiffen material has the same density as the base
MDDM, material is treated as a ‘‘composite’’ consisting of matrix material but with 100 times higher strength in order to simulate
and spherical inclusion materials. The effective material properties the strong rigidity effect produced by stiffeners. The upper weight
of the ith element is expressed in the form limit of the add-on ‘‘artificial’’ stiffen material is set to be 10% of
冉
␬ 共 i 兲 ⫽ ␬ 0 1⫹
c 共1i 兲共 ␬ 1 ⫺ ␬ 0 兲
共 1⫺c 共1i 兲兲 ␣ 0 共 ␬ 1 ⫺ ␬ 0 兲 ⫹ ␬ 0
冊 (17)
the total base weight. That gives the total allowable weight of the
whole structure be 5.5836 Kg. In the initial design, the add-on
‘‘artificial’’ stiffen material is uniformly distributed to the box in
冉
␮ 共 i 兲 ⫽ ␮ 0 1⫹
c 共1i 兲共 ␮ 1 ⫺ ␮ 0 兲
共 1⫺c 共1i 兲兲 ␤ 0 共 ␮ 1 ⫺ ␮ 0 兲 ⫹ ␮ 0
冊 (18)
order to produce an unbiased starting point. From the analyses of
the initial design, we found 20 structural eigenmodes and 3 acous-
tic eigenmodes including one Helmholtz mode for the uncoupled
with system in the range of 0 to 500 Hz. Then, a coupled acoustic-
structural analysis was performed. We found that the eigenfre-
1 1⫹ ␯ 0 quencies of the uncoupled systems were shifted only slightly after
␣ 0⫽ (19) being coupled in the same frequency range. In Table 1, eigenfre-
3 1⫺ ␯ 0
quencies in the range of 0 to 500 Hz for uncoupled and coupled
2 4⫺5 ␯ 0 systems are listed for comparison.
␤ 0⫽ (20) Since we are only interested in the low frequency sound reduc-
15 1⫺ ␯ 0
tion, damping effect for the coupled acoustic-structural system
where subscripts 0 and 1 represent matrix and inclusion materials; was neglected. In the structural field, 1536 4-node plate elements
␬ and ␮ denote the bulk modulus and the shear modulus. ␯ 0 is are used and each of them is also treated as design variable for the
Poisson’s ratio of the matrix, and c 1(i ) is the volume fraction of the stiffener design optimization formulation. In the acoustic field,
inclusion material in the ith element. Using this model, the vol- 4096 8-node solid elements are constructed. To reduce the interior

Table 1 Coupled and uncoupled eigenfrequencies of a 30 cm
Ã40 cmÃ50 cm aluminum box
Structural Acoustic Coupled

Mode frequency frequency frequency
number 共Hz兲共Hz兲共Hz兲
1 - 0.00 0.00
2 114.95 - 117.02
3 130.72 - 129.92
4 179.50 - 178.67
5 195.32 - 194.99
6 195.36 - 196.09
7 231.06 - 239.07
8 242.49 - 241.66
9 257.62 - 254.44
10 291.94 - 291.35
11 302.71 - 302.71
12 323.13 - 304.21
13 335.41 - 330.40
14 - 344.05 370.61
15 387.39 - 382.70
16 389.28 - 386.75
17 411.79 - 411.72
18 418.62 - 418.62
19 - 430.06 443.02
20 443.03 - 443.33
21 462.87 - 452.73
22 484.05 - 482.94 Fig. 3 Iteration history of the objective function in case 1
23 494.91 - 495.40
5.2 Case 2. In the second case, the unit harmonic excitation

with the same excitation frequency as in the case 1 is applied to
the center point of the flank plate as shown in Fig. 5.
sound level, the norm of acoustic excitation is chosen as design Iteration history of the objective function is shown in Fig. 6 and
objective as we previously discussed. Four optimal design ex- we found the acoustic excitation is decreased from 6.246733e-7 to
amples under different external excitations are presented below. 9.847646e-8 with the overall interior sound level reduction from
76.73 dB to 51.92 dB in ten iterations. The result of the optimal
5.1 Case 1. In the first case, a unit harmonic excitation with stiffeners placement is shown in Fig. 7 and we can see the stiff-
frequency of ␻ ⫽50 Hz is applied at the center of the top plate as
shown in Fig. 2. Using the proposed methodology, the norm of
acoustic excitation is reduced from 8.609643e-7 to 4.758937e-7 in
Table 2 Trend of interior SPL reduction in case 1
ten iterations and the overall interior sound decreased from 91.89
dB to 68.97 dB. The iteration history is showed in Fig. 3 and the Iteration Sound Pressure Level Weight
trend of SPL reduction is listed in Table 2. The final stiffener number 共dB兲共Kg兲
configurations is shown in Fig. 4. It was found that the stiffener is
initial 91.89 5.5836
mainly concentrated in the region where the external force ap- 1 89.62 5.5836
plies, which is reasonable because the excitation frequency is 2 80.87 5.5836
lower than resonance frequencies of the system. The uncoupled 3 77.05 5.5836
structural frequencies and the coupled system frequencies in the 4 75.64 5.5836
5 74.33 5.5836
range of 0 to 200 Hz are listed in Table 3. Comparing to the initial 6 72.08 5.5836
design with uniformly distributed stiffener material, we found that 7 70.88 5.5836
the final design also exhibits great changes of natural frequencies 8 70.52 5.5836
and mode shapes. 9 70.26 5.5836
10 68.97 5.5836
Fig. 2 A box under an excitation loading at the center of the Fig. 4 Optimal configuration of the stiffeners placement in
top plate case 1

Table 3 Natural frequencies of the final design in case 1
Mode Structural frequency Coupled frequency

number 共Hz兲共Hz兲
1 - 0.00
2 70.18 72.31
3 120.74 120.27
4 120.87 120.53
5 124.97 125.10
6 138.17 146.12
7 147.54 148.25
8 177.84 177.51
9 185.15 184.02
10 190.37 187.38
eners are also mostly located in the external force applied region.
Table 4 shows the trend of interior SPL reduction throughout the
iterations. The corresponding uncoupled structural frequencies Fig. 7 Optimal configuration of the stiffeners placement in
case 2
and coupled system frequencies in the range of 0 and 200 Hz are
listed in Table 5. The final design also experiences changes in
natural frequencies and mode shapes from the initial design.
5.3 Case 3. In the third case, a unit harmonic excitation optimal stiffener location. Unlike the first two cases, in this case
with higher frequency is applied at the center of the top plate. The the locations of stiffeners are spread out to different region of the
excitation frequency is chosen as ␻ ⫽150 Hz. We found the ob- box. This is due to the fact the excitation frequency is in the
jective function reduced from 2.811064e-5 to 1.548597e-5 and the middle of several resonance frequencies of the system. To mini-
overall interior sound decreased from 80.13 dB to 67.03 dB after mize the acoustic excitation, the add-on stiffeners try to reduce the
ten iterations. The iteration history is showed in Fig. 8 and the effects from all these resonance modes as much as possible. The
trend of SPL reduction is showed in Table 6. Fig. 9 shows the natural frequencies below 200 Hz of the final design are also
listed in Table 7 for reference.
5.4 Case 4. In the last example, the stiffener design under a
band of frequency excitation is studied. A unit harmonic excitation
is applied at the center of the top plate with a band of low exci-
tation frequencies ranged from 20 to 100 Hz. A modified objective
function, Eq. 共15兲, is used here. Since the excitation frequency
range falls below any resonance frequency, the optimal stiffener
location turns out to be very similar to that of the first case as Fig.
4. We found the objective function is reduced from 3.106249e-7 to
1.842361e-7 and the overall interior sound defined in Eq. 共16兲 is
decreased from 93.53 dB to 75.16 dB after ten iterations. Fig. 10
shows the interior SPL comparison between the initial design and
the final design within the exciting frequency range.
Table 4 Trend of SPL reduction in case 2

Fig. 5 A box under an excitation loading at the center of the
Iteration Sound Pressure Level Weight
flank plate number 共dB兲共Kg兲
initial 76.73 5.5836
1 72.20 5.5836
2 61.03 5.5836
3 58.29 5.5836
4 57.33 5.5836
5 56.62 5.5836
6 56.90 5.5836
7 55.69 5.5836
8 53.92 5.5836
9 52.12 5.5836
10 51.92 5.5836
Table 5 Natural frequencies of the final design in case 2
Mode Structural frequency Coupled frequency

number 共Hz兲共Hz兲
1 - 0.00
2 104.94 107.27
3 117.86 117.13
4 158.14 157.36
5 167.27 167.43
6 188.17 195.37
Fig. 6 Iteration history of the objective function in case 2

Fig. 8 Iteration history of the objective function in case 3
Table 6 Trend of interior SPL reduction in case 3 Fig. 10 Interior SPL comparison between the initial design
and the final design
Iteration Sound Pressure Level Weight
number 共dB兲共Kg兲
initial 80.13 5.5836
1 79.59 5.5836
2 77.39 5.5836
3 74.91 5.5836 6 Conclusion and Future Work
4 75.88 5.5836
5 73.40 5.5836 In this paper, optimal stiffener design for interior sound reduc-
6 71.20 5.5836 tion of coupled acoustic-structural system is studied. Forced fre-
7 70.46 5.5836 quency response and its sensitivity of this coupled system were
8 68.99 5.5836
9 67.63 5.5836 solved by the modal analysis method. Using a topology optimiza-
10 67.03 5.5836 tion based approach, the optimal stiffener placement problem was
converted into an optimal material distribution problem. Instead
of using the overall interior sound level directly, the norm of
acoustic excitation was used as the objective function and from
our numerical examples, it showed the acoustic excitation is a
good indicator for sound reduction. At low frequency excitations,
the optimal stiffeners might be applied to the region near the
external force source, however when the excitation frequency is
among several resonance frequencies, stiffener might be placed
throughout the structure in order to compensate the resonance
modes involved. In this work, a conservative system with no
damping effect is considered, therefore the proposed approach is
only valid for single frequency excitation or a band of low
frequency excitations without any resonance frequency included.
More general method to handle damping is currently under
investigation.
References
Fig. 9 Optimal configuration of the stiffeners placement in 关1兴 Hagiwara, W., Kozukue, W., and Ma, Z. D., 1993, ‘‘The Development of
case 3 Eigenmode Sensitivity Analysis Methods for Coupled Acoustic-Structural Sys-
tems and Their Application to Reduction of Vehicle Interior Noise,’’ Finite
Elem. Anal. Design, 14, pp. 235–248.
关2兴 Huff, J. E., Jr., and Bernhard, R. J., 1995, ‘‘Acoustic Shape Optimization
Using Parametric Finite Elements,’’ ASME 1995 Design Engineering Techni-
Table 7 Natural frequencies of the final design in case 3 cal Conferences, DE-Vol. 84-2, pp. 577–584.
关3兴 Wodtke, H. W., and Koopmann, G. H., 1995, ‘‘Quieting Plate Modes with
Mode Structural frequency Coupled frequency Optimally sized Point Masses—A Volume Velocity Approach,’’ ASME 1995
number 共Hz兲共Hz兲 Design Engineering Technical Conferences, DE-Vol. 84-2, pp. 647– 654.
关4兴 Constans, E., and Belegundu, A., 1996, ‘‘Minimizing Radiated Sound Power
1 - 0.00 from Vibrating Shells,’’ The 6th AIAA/NASA/ISSMO Symposium on Multidis-
2 88.77 90.39 ciplinary Analysis and Optimization, Bellevue, WA, Paper 96-4111, pp. 1106 –
3 115.09 115.62 1116.
4 161.26 160.88 关5兴 Luo, J. H., and Gea, H. C., 1997, ‘‘Modal Sensitivity Analysis of Coupled
5 166.55 166.00 Acoustic-Structural Systems,’’ J. Sound Vib., 119, pp. 545–550.
6 173.40 173.79 关6兴 Gea, H., 1996, ‘‘Topology Optimization: A New Micro-Structure Based De-
sign Domain Method,’’ Comput. Struct., 61„5…, pp. 781–788.

关7兴 Chickermane, H., and Gea, H. C., 1996, ‘‘A New Local Function Approxima- 关9兴 Diaz, A. R., and Kikuchi, N., 1992, ‘‘Solutions to Shape and Topology Eigen-
tion Method for Structural Optimization Problems,’’ Int. J. Numer. Methods value Optimization Problems Using a Homogenization Method,’’ Int. J. Nu-
Eng., 39, pp. 829– 846. mer. Methods Eng., 35, pp. 1487–1502.
关8兴 Craggs, A., 1971, ‘‘The Transient Response of a Coupled Plate-Acoustic Sys- 关10兴 Bendsøe, M. P., and Kikuchi, N., 1988, ‘‘Generating Optimal Topologies in
tem Using Plate and Acoustic Finite Elements,’’ J. Sound Vib., 15, pp. 509– Structural Design Using A Homogenization Method,’’ Comput. Methods Appl.
528. Mech. Eng., 71, pp. 197–224.

Yuji Ohue New Evaluation Method on Gear
Associate Professor,
Department of Intelligent Mechanical
Systems Engineering,
Dynamics Using Continuous and
Kagawa University,
Hayashi, Takamatsu, 761-0396, Japan
Discrete Wavelet Transforms
The aim of this study is to propose a new evaluation method of gear dynamics using the
Akira Yoshida continuous and discrete wavelet transforms. The wavelet transform (WT) is a method for
Professor the time-frequency analysis of signals. In order to evaluate the difference in the gear
Mem. ASME, dynamics due to the gear materials, which are sintered and steel ones, the dynamic
Department of Mechanical Engineering, characteristics of gears were measured using a power circulating gear testing machine.
Okayaya University, The gear dynamics were analyzed in a time-frequency domain by the continuous and
Tsushima-naka, Okayama, 700-8530, Japan discrete WTs. The new evaluation method using the WTs proposed in this paper was more
e-mail: [email protected] useful compared with the conventional one to investigate the damping characteristic and
the dynamic condition of the gear equipment. 关DOI: 10.1115/1.1547465兴
Introduction Test Gear and Testing Machine

Generally, the dynamic signals in the field of engineering prob- Table 1 and Fig. 1 show the specification and the shapes of gear
lems such as vibration, sound and so on, have been analyzed using pair, respectively. The module and the pressure angle of the gear
the fast Fourier transform 共FFT兲. The FFT has been the most pair are 5 mm and 20 deg., respectively. The contact ratio of the
common method to analyze the frequency properties of the sig- gear pair is 1.246. The gear has 16 teeth and a face width of 6
nals. In the FFT analysis, the signal as the function of time is mm. The materials of the test gears are sintered and steel 共JIS;
converted to the power spectra in a frequency domain. However, SCM440兲 ones. Those test gears were induction-hardened before
except for a special case, the frequency components of the most grinding of tooth surface. The material of the mating pinion is
steel 共JIS: SCM415兲. The chemical composition of the sintered
signals encountered in the engineering problem change with time.
material is almost the same as that of the steel SCM440. Many
Based on the FFT alone, it is hard to investigate whether the
pores existed in the sintered great. Generally, the sintered materi-
frequency components of the signals vary with time or not, even als have higher damping ratio due to the pores, compared with the
though the phase of the Fourier transform relates to time shifting. steel ones. The density of the sintered gear in this study is 6.8
The FFT analysis is no more adequate for those applications. g/cm3. The Young’s modulus and Poisson’s ratio of the sintered
Therefore, it is significant and important to adopt the time- gear are 152 GPa and 0.25, and those of the steel gear are 206
frequency analysis for those signals varying with time 关1–3兴.
It is possible by using the time-frequency analysis to investigate
how the frequency components of the signal vary with time. The Table 1 Specification of gear pair
time-frequency analysis can provide more beneficial information
about the frequency compared with the FFT. During the last two Pinion Gear
Module mm 5
decades, a new mathematical technique for the time-frequency Pressure angle deg. 20
analysis, which has been called the wavelet transform 共WT兲, Number of teeth 15 16
has been extensively developed 关1,2兴. The WT has some fea- Addendum modification 0.571 0.560
coefficient
tures different from those of the short-time fast Fourier trans- Tip circle diameter mm 90.71 94.60
form or the Wigner-Ville distribution. Applications of the WT Center distance mm 82.55
are actively studied in a variety of the fields of engineering Face width mm 18 6
science 关4 –7兴. Contact ratio 1.246
Accuracy* Class 1 Class 1
Gears have been employed in automobiles, motorcycles, indus- Tooth surface finishing Grinding
trial robots and so on. The gears are an important machine ele-
ment to transmit the power and to change the speed of the equip- *JIS B 1702
ment. The FFT is generally used for analyzing the gear dynamics.
However, it is important to evaluate the gear dynamic perfor-
mance more precisely using WT. Therefore, in order to evaluate
the difference in the gear dynamics due to the gear materials, the
dynamic performances of sintered and steel gears were measured
at gear rotational speeds of 1600 rpm to 10,000 rpm, using a
power circulating gear testing machine. The gear dynamics con-
cerned with tooth root strain, vibration acceleration of gear box
and sound near gear box, were analyzed in the time-frequency
domain by the continuous WT, and the signals of the gear dynam-
ics were decomposed and reconstructed by the discrete WT. The
validity of the new evaluation method by the WT is discussed.

in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002;
Revised Sept. 2002. Associate Editor: R. P. S. Han. Fig. 1 Shapes and dimensions of gear pair
Wavelet Transform
Continuous Wavelet Transform. The continuous wavelet
transform 共CWT兲 of a function f (t) is defined as follows.
共 W ␺ f 兲共 b,a 兲 ⫽ 兩 a 兩 ⫺1/2 冕
⫺⬁
⬁
f 共 t 兲␺ 冉冊
t⫺b
a
dt (1)
Where, the ␺ (t) indicates the conjugate of a mother wavelet func-

tion ␺ (t). The a and b indicate the parameters on frequency and
time. The Gabor function defined by Eq. 共2兲 was adopted as the
Fig. 2 Tooth profile errors of gear pair mother wavelet function ␺ (t) in this study
GPa and 0.3, respectively. Figure 2 shows the tooth profile errors
of test gear pairs. The tooth surfaces of both gears and pinion were
␺ 共 t 兲 ⫽ ␲ ⫺1/4 冉冊再冉冊
␻p
␥
1/2
exp ⫺
1 ␻p
2 ␥
2
⫹i ␻ p t 冎 (2)
finally ground. The tooth profile errors of both sintered and steel Where, ␻ p is a center of angular frequency, and ␥ is a constant
gears and almost the same. The accuracy of the test gear pairs are and was set to be ␲ (2/ln 2) 1/2⫽5.336. Figure 4 shows an example
in class 1 according to Japanese industrial standard 共JIS B 1702兲. of the Gabor function and its Fourier spectrum. The integral for
The test apparatus shown in Fig. 3 is a power circulating type Eq. 共1兲 was calculated using the trapezoidal rule and the calcula-
gear testing machine with a center distance of 82.55 mm. The test tion results are expressed using the following equation.
gear was driven at an interval of 200 rpm within a range of rota-
tional speeds n 2 of 1600 rpm to 10,000 rpm. The loading torque 兩共 W ␺ f 兲兩 ⫽ 冑兵 Re共 W ␺ f 共其 2 ⫹ 兵 Im共 W ␺ f 共其 2 (3)
was statically applied to the gear shaft by a lever loading with
dead weights and a torsional coupling. The torsional torque was Where, Re(W ␺ f ) and Im(W ␺ f ) are the real and the imaginary
taken as a scale for loading. In this experiment, the value of the parts in Eq. 共1兲. To express the wavelet map, that is the intensity
torque was 30 Nm, which corresponds to a load per unit face map in a time frequency domain, the intensity calculated by the
width of 142 N/mm in the normal direction to the tooth surface. CWT was normalized by the maximum intensity in the map.
The gear pair was lubricated with a gear oil by pressure feeding at
Discrete Wavelet Transform. When the coordinates 共b, a兲 of
a flow rate of 750 ml/min and an oil temperature of 313 K.
the CWT shown in Eq. 共1兲 are discretized to the coordinates
The dynamic tooth root strain was measured using a dynamic
strain amplifier and a wire strain gage with a gage length of 0.3 (2 ⫺ j k,2⫺ j ) using two integers j and k, the discrete wavelet trans-
mm, which was bonded on a compression side of the root fillet of form 共DWT兲 is defined as follows.
冕
the gear. The vibration acceleration of the gear box was detected ⬁
by a piezo-electric pickup through an amplifier. The pickup has a d 共kj 兲 ⫽2 j/2 f 共 t 兲 ␺ 共 2 j t⫺k 兲 dt (4)
maximum response frequency of 20 kHz. The sound pressure near ⫺⬁
the gear box was detected by a condenser type microphone having
a maximum response frequency of 12.5 kHz. The microphone was Where, d (kj ) is equal to (W ␺ f )(2 ⫺ j k, 2⫺ j )). j is called level. The
located at a distance of 300 mm from the side of the gear box. inverse discrete wavelet transform 共IDWT兲 is defined as
These signals were stored into a data recorder with a synchronous
signal obtained by a phototransistor and two disks with slits which
were fixed on both shifts of the gear and the pinion. These stored
f 共 t 兲⬇ 兺 g 共t兲
j
j (5)
signals were analyzed using a computer with an A/D converter
board. The function g j (t) on the wavelet component is given by
Fig. 3 Power circulating gear testing machine and measurement system of dynamic performance of
gear pair

Fig. 5 Scaling and wavelet functions based on cardinal
B-spline In order of m Ä4
Fig. 4 Gabor function f 共 j 兲共 t 兲 ⫽ f j⫺1 共 t 兲 ⫹g j⫺1 共 t 兲 ⫽g j⫺1 共 t 兲 ⫹g j⫺2 共 t 兲 ⫹g j⫺3 共 t 兲 ⫹•••

(12)
In this study, the scaling and the mother wavelet functions
based on the cardinal B-spline function in order of m⫽4 was
g j共 t 兲⫽ 兺d
k
共 j兲
k ␺ 共 2 t⫺k 兲
j
(6) adopted. The cardinal B-spline function N m (t) in order of m is
defined as follows 关1兴.
冉冊
m
Suppose that f j (t) is the function at a level j, f i (t) is satisfied with 1 m
the following relation. N m共 t 兲 ⫽
共 m⫺1 兲 ! 兺共 ⫺1 兲
k⫽0
k
k
共 t⫺k 兲 ⫹
m⫺1
(13)
f j共 t 兲⫽ 兺c
k
共 j兲
k ␾ 共 2 t⫺k 兲
j
(7) Figure 5 shows the scaling function ␾ (t) and the mother wavelet
function ␺ (t) based on the function N 4 (t), that is the cardinal
B-spline function in order of 4. The function ␾ (t) is equal to the
Where, 兵 c (kj ) 其 is a sequence at a level j, and ␾ (t) is a scaling function N 4 (t). From this figure, it is understood that the function
function. The scaling function ␾ (t) and the mother wavelet func- ␺ (t) is very smooth and is very similar to sinusoidal functions.
tion ␺ (t) are satisfied with the two-scale relations as follows. The sequences 兵 a k 其 , 兵 b k 其 , 兵 p k 其 and 兵 q k 其 are given in the refer-
ence 关1兴.
␾共 t 兲⫽ 兺 p ␾ 共 2t⫺k 兲
k
k (8)
Conventional Evaluation of Gear Dynamics
␺共 t 兲⫽ 兺 q ␾ 共 2t⫺k 兲
k
k (9) Figure 6 shows the examples of the dynamic and static tooth
root strains or stresses measured in this experiment. Generally,
Where, 兵 p k 其 and 兵 q k 其 are two-scale sequences. The algorithms of in order to evaluate the dynamic load of the gear pair, the dyna-
the decomposition and the reconstruction using the DWT and the mic load ratio ␴ dmax / ␴ smax , that is the ratio of the maximum
IDWT on the function f (t) are given as Eqs. 共10兲 and 共11兲. dynamic tooth root stress ␴ dmax to the maximum static tooth
冧
c 共kj⫺1 兲 ⫽ 兺a l
共 j兲
1⫺2k c l
(10)
d 共kj⫺1 兲 ⫽ 兺 l
b 1⫺2k c 共l j 兲
c 共kj 兲 ⫽ 兺关p
l
共 j⫺1 兲
k⫺2l c l ⫹q k⫺2l d 共l j⫺1 兲兴 (11)
Where, 兵 a k 其 and 兵 b k 其 are the sequences to decompose the func-

tion f j (t) at a level j. Therefore, the decomposition and the recon-
struction on the function f (t) are concluded into the calculations
using the sequences 兵 a k 其 , 兵 b k 其 , 兵 p k 其 and 兵 q k 其 . The functions
g j (t) and f j (t) at a level j are able to be found by using Eqs. 共6兲
and 共7兲. The function f j (t) decomposed into the function g j (t) on
the wavelet component is satisfied with the following relation. Fig. 6 Dynamic and static tooth stresses

Fig. 7 Relation between dynamic load ration and n 2
root stress ␴ smax , has been used. In this study, the tooth root
stress at a rotational speed n 2 of 6.6 rpm was adopted as the static
stress ␴ smax .
Figure 7 shows the dynamic load ratio ␴ dmax / ␴ smax plotted
against the rotational speed n 2 of gear. The frequency f z indicates
the tooth mesh frequency. The natural frequency of the gear pair
including the shafts, which support the gear pair, was calculated
using a program given in reference 关8兴. The equation of motion
for the calculation of the natural frequency of the gear pair is
given by
2
M ẍ⫹Dẋ⫹K 共 t,x 兲 x⫽W⫹ 兺 K 共 t,x 兲 e 共 t 兲

i
i i (14)
where, M is the equivalent inertia mass of gear pair along the line
of action, D is the damping coefficient between gear pair includ-
ing the shafts, K is the stiffness of gear pair, W is the static load,
e i is the tooth profile error and x is the relative displacement along
the line of action. Subscript i indicates the pinion (i⫽1) and the
gear (i⫽2). Then, the natural frequency f e is calculated using the
following equation. Fig. 8 Tooth root strain, WT map and FFT at n 2 Ä1800 rpm
f e⫽
1
2␲ 冑冕1
Tz
Tz K 共 t,x 兲
0 M
dt (15)
Where, T z is the tooth mesh period. In this experiment, the calcu- dynamic load and used only the maximum value of the gear dy-
lated value of f e was approximately 2.56 kHz. The gear box was namic load. Thus, the information of the gear dynamic load in the
hit by a hammer to measure its natural frequency in the horizontal time-frequency domain is important to evaluate the great perfor-
direction of the gear box using the FFT. As the result, the natural mance in detail. In this section, the dynamic tooth root strains are
frequency of the gear box was about 5.0 kHz. From Fig. 7, for analyzed in the time-frequency domain using the WT.
both the sintered and steel gears, the ratio ␴ dmax / ␴ smax increases Figure 8 shows the waveform of the tooth root strain 共top of the
with fluctuation as n 2 increases. The ratio ␴ dmax / ␴ smax has three figure兲, its WT map 共lower left of the figure兲 and the analyzed
peak values at n 2 ⫽4200 rpm ( f z ⫽1.12 kHz兲, 6000 rpm ( f z result using the FFT 共lower right of the figure兲 at a rotational
⫽1.60 kHz兲 and 9000 rpm ( f z ⫽2.40 kHz兲, since the tooth mesh speed n 2 of 1800 rpm. The WT maps of both the gears were
frequency and its harmonics coincide with both the natural fre- obtained using the CWT. The WT map shows the intensity of the
quencies of the gear pair and the gear box. Namely, the gear pair tooth root strain in the time-frequency domain. In the WT map,
for both the sintered and steel gears became the resonance condi- not only the frequency components but also the change of the
tions at their speeds. The ratio ␴ dmax / ␴ smax of the sintered gear is intensity of each frequency in the tooth root strain with the time
mostly smaller than that of the steel one against n 2 . Therefore, it can be provided more precisely, compared with the FFT analysis.
is obvious that the sintered gear is superior to the steel one from In the WT maps of the sintered and steel gears, each intensity in
the viewpoint of the damping characteristics. two regions above and below f z ⫽480 Hz indicates each peak. The
contact time of one tooth pair at n 2 ⫽1800 rpm is approximately
2.6 ms. Suppose that the wave of the tooth root strain is roughly a
sine wave with a period of 5.2 ms, the frequency of the sine wave
Analysis of Gear Dynamics Using WT is 192 Hz. It can be considered that the frequency components
In the previous section, the conventional evaluation method below f z depend on the contact time of the one tooth pair. On the
was employed for evaluating the gear dynamic load. The conven- other hand, it can be considered that the frequency components
tional evaluation method, however, has not taken into account the above f z are caused by the torsional vibration of the gear pair. The
information on the time and the frequency contained in the gear relative intensities at the region above f z of the sintered gear are

Where, ␧ a is the contact ratio of the gear pair, z 2 is the number
of the gear teeth, T z is the tooth mesh period. Substitution of
the values of ␧ a ⫽1.246 and z 2 ⫽16 shown in Table 1 into Eq.
共16兲 yields f low ⫽0.4 f z . The relation between f low and f z of
the sintered gear was the same as that of the steel one. It is obvi-
ous that the tooth root strain consists of two main frequencies
due to the tooth contact 共below f z ) and due to the torsional vibra-
tion of the gear pair 共above f z ), independent of the material of
the gear.
In order to evaluate the tooth root strain consisting of both the
frequencies due to the tooth contact 共below f z ) and due to the
torsional vibration of the gear pair 共above f z ), the tooth root strain
is decomposed by DWT, and is reconstructed by IDWT in the two
regions above and below f z . Figure 10 shows the results of the
decomposed tooth root strains at n 2 ⫽1800 rpm by DWT. The
waveform f j becomes smooth as the absolute value of j becomes
larger. The relation between the waveforms f j and g j is given by
Eq. 共12兲, that is, the waveform f j is decomposed into the wave-
Fig. 9 Two main frequencies contained in tooth root strain forms f j⫺1 and g j⫺1 . At levels j⫽⫺3 and ⫺4 corresponding to
the frequencies above f z , the amplitude of the waveforms g j of
the sintered gear is smaller than that of the steel one. On the other
hand, at a level j⫽⫺6 corresponding to the frequency below f z ,
lower than those of the steel one. It can be understood that two the maximum value of the waveform f ⫺6 of the sintered gear is
main frequency components above and below f z exist in the dy- larger than that of the steel one, since the Young’s modulus 共152
namic tooth root strain. GPa兲 of the sintered gear is smaller than that 共206 GPa兲 of the
Figure 9 shows the relationship between the two main frequen-
steel one. Figure 11 shows the reconstructed tooth root strains and
cies contained in the tooth root strain of the steel gear and the
rotational speed n 2 . The tooth root strain consists of two main the subtracted strain ␧ n2 ⫺␧ 6.6 at n 2 ⫽1800 rpm in the case of the
frequencies above and below f z at each rotational speed of gear. sintered gear. The reconstructed strain 兺 g j above f z is almost the
Especially, the relation between the frequency f low below f z and same as the subtracted strain ␧ n2 ⫺␧ 6.6 in shape. The strain ob-
n 2 is given by a following relation. tained by subtracting the static strain ␧ 6.6 from the dynamic strain
␧ n2 can be supposed to be the strain due to the dynamic load.
1 z 2n 2 fz 1 Therefore, the reconstructed strain 兺 g j above f z is dependent not
f low ⫽ ⫽ ⫽ (16)
1␧ a 60 2␧ a 2␧ a T z on the static load but on the dynamic one.
Fig. 10 Decomposed tooth root strain

Fig. 11 Reconstructed tooth root strains
New Evaluation of Gear Dynamics

Figure 12 shows the conventional dynamic load ratio 共top of the
figure兲, the maximum strain value of the decomposed strain f j
below f z 共second of the figure兲, the strain amplitude of the recon-
structed strain 兺 g j above f z 共third of the figure兲 and the new
dynamic load ratio 共bottom of the figure兲 plotted against the rota-
tional speed n 2 . The maximum values of the decomposed strains
f j 共second of the figure兲 at n 2 higher than 1600 rpm are almost Fig. 13 Example of decomposed vibration of gear box
constant and close to that of the static strain at n 2 ⫽6.6 rpm. The
fluctuation of the amplitude of the reconstructed strain 兺 g j 共third
of the figure兲 is similar to that of the conventional dynamic load modulus, respectively. The new dynamic load ratio is given by
ratio 共top of the figure兲 against n 2 . Therefore, it is clear that the dividing ␴ dmax by the stress, to which the maximum value of the
values of the reconstructed strains above f z and the decomposed reconstructed strain f j is converted. The fluctuation of the new
strains below f z depend on the damping ratio and the Young’s dynamic load ratio 共bottom of the figure兲 is similar to that of the
conventional one against n 2 . The new dynamic load ratio pro-
posed in this paper can be calculated using only the dynamic tooth
root stress without the static tooth root stress. In some equip-
ment with gear sets, it is occasionally difficult to operate the
equipment at very low speed and to measure the static stress.
Therefore, the proposed new dynamic load ratio, which can be
calculated using only the dynamic tooth root stress, seems to be
beneficial in engineering.
Figure 13 shows the decomposed vibration acceleration of the
gear box at n 2 ⫽1800 rpm during one revolution of gear in the
case of the sintered gear. Both the vibration acceleration and the
sound pressure are reconstructed in two regions above and below
f z . Figures 14 and 15 show the root-mean-squares of the vibra-
tion acceleration and the sound pressure plotted against n 2 . The
root-mean-square 共R.M.S.兲 of the vibration acceleration and the
sound pressure was given by the following equation.
R.M.S.⫽ 冑冕 1
Tr
Tr
0
v共 t 兲 2 dt (17)
Where, T r is one revolution period of the gear and v (t) is a signal.

The values of both the vibration and the sound reconstructed
in the region below f z is proportional to n 2 to the power two,
and is independent of the material of the gear. On the other hand,
the values of both the vibration and the sound reconstructed in
the region above f z for the sintered gear are smaller than those of
the steel one. This tendency is the same as that in the original
vibration and sound. The vibration due to the tooth mesh propa-
gates to the gear box through shafts and bearings supporting the
gear pair. The vibration of the gear box and the sound radiated
from the gear box are caused by the torsional vibration of the gear
Fig. 12 Conventional and new dynamic load ratios pair due to the tooth mesh. The torsional vibration depends on the

root-mean-square values of the vibration acceleration and the
sound pressure below f z were independent of the gear material
and were proportional to the rotational speed to the power two.
The behavior above f z was influenced by the difference of the
gear material.
3. The new evaluation method using the continuous and dis-
crete wavelet transforms proposed in this paper was very benefi-
cial compared with the conventional one, in order to investi-
gate the damping characteristic. The dynamic condition of the
gear equipment due to the unbalance of the gear sets, the tooth
profile error and so one could be well assessed by analyzing
the vibration and the sound of gear sets in two regions above and
below f z .
Acknowledgments
The authors would like to thank Sumitomo Metal Industries,
Ltd. and Japan Energy Co. Ltd. for providing the sintered material
Fig. 14 Root-mean-square of vibration acceleration above and and the lubricating oil, respectively. This research was supported
below f z financially in part by the scientific research fund of the Japanese
Ministry of Education, Culture, Sports, Science and Technology
and the Mazda Foundation to which the authors express their
frequency components above f z . In short, the fluctuations of both gratitude.
the vibration and the sound are caused by the frequency compo-
nents above f z .
Nomenclature
Conclusion
D ⫽ damping coefficient between gear pair
In order to evaluate the difference in the gear dynamics with K(t,x) ⫽ stiffness of gear pair
different gear materials, the gear dynamics concerning tooth root M ⫽ equivalent inertia mass of gear pair along line
strain, vibration and sound were measured using a power circulat- of action
ing gear testing machine in a range of rotational speeds of 1600 to N m (t) ⫽ cardinal B-spline function in order of m
10,000 rpm. The gear materials employed were steel and sintered Tr ⫽ one revolution period of gear
ones. The gear dynamics were analyzed in a time-frequency do- Tz ⫽ tooth mesh period
main by the continuous and discrete wavelet transforms. W ⫽ static load of gear pair
1. The dynamic tooth root strain indicated indirectly the con- (W ␺ f )(b,a) ⫽ wavelet transform
dition of the torsional vibration of the gear pair. The value of a ⫽ frequency parameter
the tooth root strain reconstructed at the region below the tooth a k ⫽ sequence for decomposition by discrete wave-
mesh frequency f z depended on the Young’s modulus of the gear let transform
material. On the other hand, the amplitude of the strain recon- b ⫽ time parameter
structed at the region above f z depended on the damping ratio of b k ⫽ sequence for decomposition by discrete wave-
the gear pair. let transform
2. The behavior of the vibration and the sound of the gear box c (kj ) ⫽ sequence at level j
against the rotational speed of the gear could be also divided d k ( j ) ⫽ sequence at level j⫽(W ␺ f )(2 ⫺ j k, 2⫺ j )
obviously to two different behaviors above and below f z . The e(t,x) ⫽ tooth profile error
f (t) ⫽ signal
f j (t) ⫽ signal at level j
g j (t) ⫽ wavelet component of signal f (t) at level j
f e ⫽ natural frequency of gear pair
f low ⫽ ⫽ f z /2␧ a
f z ⫽ tooth mesh frequency ⫽z 2 n 2 /60
j ⫽ level
k ⫽ integer for time parameter
m ⫽ order of cardinal B-spline function
n 2 ⫽ rotational speed of gear
p k , q k ⫽ two-scale sequences for reconstruction by dis-
crete wavelet transform
t ⫽ time
v (t) ⫽ signal of vibration acceleration or sound pres-
sure
x ⫽ relative displacement along line of action
z 2 ⫽ number of gear teeth
␧ a ⫽ contact ratio
␾ (t) ⫽ scaling function
␺ (t) ⫽ mother wavelet function
␴ dmax ⫽ maximum dynamic tooth root stress
␴ smax ⫽ maximum static tooth root stress
Fig. 15 Root-mean-square of sound pressure above and ␥ ⫽ constant for Gabot function
below f z ␻ p ⫽ center of angular frequency

References 关6兴 Mori, K., Kasashima, N., Yoshioka, T., and Ueno, Y., 1996, ‘‘Prediction of
Spalling on a Ball Bearing by Applying the Discrete Wavelet Transform to
关1兴 Chui, C. K., 1992, Introduction to Wavelet, Academic Press. Vibration Signals,’’ Wear 195, pp. 162–168.
关2兴 Mallat, S., 1998, A Wavelet Tour of Signal Processing, Academic Press.
关7兴 Yoshida, A., Ohue, Y., and Ishikawa, H., 2000, ‘‘Diagnosis of Tooth Surface
关3兴 Cohen, L., 1995, Time-Frequency Analysis, Prentice-Hall PTR.
关4兴 Newland, D. E., 1993, Random Vibrations, Spectral & Wavelet Analysis, Failure by Wavelet Transform of Dynamic Characteristics,’’ Tribol. Int., 33,
Longman Scientific & Technical. 273–279.
关5兴 Staszewski, W. J., and Tomlinson, G. R., 1994, ‘‘Application of the Wavelet 关8兴 JSME Research Group on Accuracy and Design in Gears, 1977, Rep. Fujita,
Transform to Fault Detection in a Spur Gear,’’ Mech. Syst. Signal Process., K., Research Report on Accuracy and Design in Gears, RC-SC38, JSME, in
8共3兲, pp. 289–307. Japanese.

A Stochastic Model for Simulation
and Diagnostics of Rolling
J. Antoni
Lecturer,
Roberval UMR CNRS 6066,
Element Bearings With Localized
University of Technology of Compiègne,
France Faults
R. B. Randall This paper addresses the stochastic modeling of the vibration signal produced by local-
Professor, ized faults in rolling element bearings and its use for diagnostic purposes. The aim is
School of Mechanical and Manufacturing essentially to provide a better understanding of the recognized ‘‘envelope analysis’’ tech-
Engineering, nique as classically used in the diagnostics of rolling element bearings, and incidentally
The University of New South Wales, give theoretical proofs for the specific features of envelope spectra as obtained from
Sydney 2052, Australia experimental data. The proposed model may also prove useful for simulation purposes.
First, the excitation force generated by a defect is modeled as a random point process and
its spectral signature is derived analytically. Then its transmission through the bearing is
investigated in detail in order to find the spectral characteristics of the resulting vibration
signal. The analysis finally gives sound justification for ‘‘squared’’ envelope analysis and
the type of spectral indicators that should be used with it. 关DOI: 10.1115/1.1569940兴
1 Introduction ing of the spectral content of the envelope of the resulting vibra-
tion signal. This model was later refined by Ho & Randall who
Most frequent faults in rolling element bearings include defects
pointed out that actual rolling element bearings experience some
such as cracks, pits and spalls on the inner race, outer race, or
random slip in their operation so that the train of impacts is
rolling elements. Such defects are usually very localized during
slightly random instead of periodic 关4兴. However small these ef-
their early stages which is precisely when they need to be de-
fects, Ho showed that the resonance characteristic is no longer
tected. As a result, intensive vibrations are produced by the repeti-
sampled in the Fourier domain but rather resembles a continuous
tive impacts of the moving parts of the bearing on incipient de-
spectral density where all the harmonics tend to smear over each
fects. Numerous techniques have been proposed over the past
other. Ho’s model resulted in a significantly better description of
three decades to diagnose rolling element bearing in the case of
bearing vibration spectra as observed in the real world, and was
localized from the vibration signals they produce. Inter alia, the
next used by Randall, Antoni & Chobsaard to show that bearing
so-called ‘‘envelope analysis’’ or ‘‘high frequency resonance’’
signals are quasi-cyclostationary—i.e. their statistics have quasi-
technique is probably one of the most valuable and is currently
periodicity 关5兴. Incidentally, this observation offered an elegant
well established in vibration based condition monitoring
way for justifying the envelope analysis method from the theory
关1,2,3,4,5兴. It is based on the idea that repetitive impacts on a
of cyclostationary processes. In a following paper, Antoni & Ran-
defect excite some resonance—usually in a high frequency range
dall refined their results after specifying that signals from local-
where the signal-to-noise ratio is high—which acts as a modula-
ized faults are not exactly quasi-cyclostationary since the random
tion carrier. Demodulation of the vibration signal around the car-
slips are non-stationary in their nature 关6兴. However they con-
rier then yields the envelope of the signal whose spectral content
cluded that the bearing signals could still be treated as pseudo-
has been shown to be very relevant in identifying the location of
cyclostationary as a first approximation.
the fault in the bearing, and the shaft on which the bearing is
The purpose of this paper is to attempt a complete treatment of
mounted. Since the early and heuristic foundations of the enve-
the stochastic modeling of bearing vibrations as produced by lo-
lope analysis technique, many papers have tried to explain its
calized faults, putting together a number of unpublished results
actual virtues and unbeaten successes when applied to rolling el-
and putting the heuristic considerations of 关6兴 on a firmer math-
ement bearings. These efforts have first focused on proposing a
ematical foundation. The aim is twofold. Firstly, it is to provide
proper model for the vibration signal generated by localized
the mechanical community with a model that has proven very
faults. It must be said that the objective of such a model is not to
satisfactory in describing actual vibration signals and, in particu-
explain the physics of bearing failures but to describe its conse-
lar, their spectra and envelope spectra—including some typical
quences as observed by the experimenter, i.e. it is phenomeno-
features that have never been explained elsewhere. Secondly, it is
logical.
to demonstrate how this model permits a proper formalization of
Most likely, the first valuable model for the vibration signal
the envelope analysis technique as classically used in the diagnos-
produced by a localized defect is due to McFadden & Smith 关2,3兴.
tics of rolling element bearings. The paper is organized as follows.
Therein the repetitive impacts generated by a defect were modeled
A first section addresses the accurate modeling of the nonstation-
as a periodic train of Dirac delta functions with period T. Conse-
ary impacting process as generated by a localized defect on the
quently the resonance characteristic in the Fourier domain was
inner race, the outer race, or on a rolling element. The spectral
sampled at regular intervals 1/T 共Fourier series兲. Moreover, Mc-
characteristics of this process are then derived from the theory of
Fadden’s model had the advantage of explicitly including different
regular point processes and important results are deduced con-
sources of amplitude modulations 共the radial load distribution, the
cerning the nature of spectral harmonics. In a second section these
moving location of impact forces兲 thus giving a good understand-
results are used to investigate the spectral properties of the result-
ing vibration signal after the impacts have propagated through the
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received April 2002; system, i.e. as measured on the housing. In particular, the general
revised January 2003. Associate Editor: M. I. Friswell. spectral signature due to a localized defect is found and its mani-
Fig. 1 The impacting process viewed as a point process
Fig. 2 Product density of degree one for ␴ ⌬ Õ T Ä1Õ30

festations in a number of spectral indicators 共the Fourier trans-
form, the power spectral density, the spectral correlation density,
the Fourier transform of the squared signal and the envelope spec- Note the degeneracy occurring at ␶ ⫽0, where f 2 (t,0)
trum兲 are investigated in detail. The relative effectiveness of these ⫽ f 1 (t) ␦ ( ␶ ). By using the terminology of stochastic point pro-
spectral indicators in diagnostics is finally discussed in the light of cesses, explicit solutions will now be found for the first two mo-
the new results. ments of the impacting process.
2.2 Analytical Forms of the Product Densities. The prod-
2 Modeling the Impacting Process uct density of degree one in Eq. 共3兲 can be expanded into
2.1 Regular Point Process. At the outset, consider the pro- ⬁ ⬁
cess generated by the repetition of impact forces when a defect in
one surface strikes a mating surface. We shall refer to it as the
f 1 共 t 兲 dt⫽ 兺 P 兵 t⭐T ⭐t⫹dt/T ⫽0 其 ⫽ 兺 ␾ 共 t 兲 dt
i⫽0
i 0
i⫽0
i (5)
impacting process F(t). For a localized defect, each impact may where ␾ 1 (t) is the probability density function of the ith impact
be well described by a Dirac delta function ␦ (t) provided the conditioned to the fact that the zeroth occurred at t⫽0 and
measured signal is sampled at a rate well below the impact spec- ␾ 0 (t)⫽ ␦ (t).
tral bandwidth 关2,3,4兴. At this stage it is assumed that all impacts Similarly, the product density of degree two is
have equal magnitudes; magnitudes and signs of impacts will be
⬁ ⬁
accounted for later in the text by modulating the impacting pro-
cess with a suitable time-varying function. f 2 共 t, ␶ 兲 dtd ␶ ⫽ 兺兺 P 兵 t⭐T ⭐t⫹dt,t⫹ ␶ ⭐T ⭐t⫹ ␶ ⫹d ␶ /T
i⫽0 j⬎i
i j 0
Without loss of generality, the reference time t⫽0 is chosen to
coincide with an arbitrary impact which defines the point from ⫽0 其 (6)
which the process is starting to be observed. Hence,
⬁
for ␶ ⬎0. Remembering that the arrival time 兵 T i 其 is a stationary
兺 ␦ 共 t⫺T 兲
Markov process,
F共 t 兲⫽ i where T 0 ⫽0 (1)
i⫽0 ⬁ ⬁
The stochastic process 兵 T i 其 governing the arrival of the impacts f 2 共 t, ␶ 兲 dtd ␶ ⫽ 兺

i⫽0
␾ i 共 t 兲 dt 兺 ␾ 共 ␶ 兲d␶,
k⫽1
k ␶ ⬎0 (7)
can be defined in a variety of ways. However it was argued in 关6兴
that an adequate assumption for rolling element bearings is where and finally, adding the degeneracy case f 2 (t,0)⫽ f 1 (t) ␦ ( ␶ ) arising
the inter-arrival times ⌬T i ⫽T i ⫺T i⫺1 are independent and identi- when ␶ ⫽0, one gets the simple expression,
cally distributed random variables 共see Fig. 1兲. In turn, this can be f 2 共 t, ␶ 兲 dtd ␶ ⫽ f 1 共 t 兲 f 1 共 ␶ 兲 dtd ␶ , ␶ ⭓0 (8)
shown to define a stationary Markov process 兵 T i 其 , that it is to say
in which each arrival is only influenced by its immediate prede- In short Eqs. 共5兲 and 共8兲 give the explicit solutions for first two
cessor and irrespectively of its index: moments of the impacting process from which those of the vibra-
tion signal will later be derived. For the physical process under
P 兵 T i ⭐t i /T i⫺ j ⫽t i⫺ j , j⫽1, . . . ,i 其 ⫽ P 兵 T i ⭐t i /T i⫺1 ⫽t i⫺1 其 consideration, it is noteworthy that the product density of degree
⫽ P 兵 T 1 ⭐t 1 /T 0 ⫽0 其 (2) one f 1 suffices to describe it at least up to the second order 共be-
cause f 2 factorizes into a product of f 1 terms兲, thus assigning to
It can easily be checked that under these conditions the arrival the instantaneous mean rate of impacts a major role in this paper.
time process 兵 T i 其 has a stationary mean E 兵 T i 其 ⫽i•T but a nonsta- As a matter of fact, the exact shape of f 1 (t) is worthy of further
tionary covariance function Co v 兵 T i ,T j 其 ⫽ ␴ ⌬2 • min(i,j) where investigation. In view of Eq. 共5兲, the first peak in f 1 (t) happens to
␴ ⌬2 ⫽Var 兵 ⌬T i 其 . It is specifically this non-stationarity that has not be the probability density function ␾ 1 (t) of the first time of oc-
properly been recognized before and actually explains distinctive currence T 1 , the second peak the probability density function of
features of the vibration signal ensuing from a faulty bearing 关6兴. T 2 and so on. Therefore the ith peak is the first one convolved
Now return to Eq. 共1兲 and define 兵 N(t) 其 as the number of with itself i times, i.e.
impacts that have occurred in the interval 关 0,t 兴 , so that 兵 dN(t) 其
denotes the number of impacts in the infinitesimal interval 关 t,t
⫹dt 兴 . Hence, Eq. 共1兲 simplifies to F(t)⫽dN(t)/dt which defines (9)
an ordinary point process 共ordinary means one in which the initial
impact occurs at zero time兲关7兴. For the physical case of interest, it Then, under mild conditions, the bandwidth of the ith peak as
is a sound assumption that the probability of occurrence of an measured by its standard deviation is 冑 i• ␴ ⌬ with ␴ ⌬ the standard
impact in dt is proportional to dt while the probability of more deviation of ␾ 1 (t). As the peaks slowly enlarge, their amplitudes
than one occurrence is negligibly smaller than 共dt兲. This property decrease accordingly so as to maintain a unit area. This is illus-
of regularity ensures the use of the product density technique to trated in Fig. 2. In the limit, the peaks completely vanish and f 1 (t)
obtain the moments of F(t) 关8兴. Specifically, we define f 1 (t) and tends to the constant value f 1 (⬁)⫽1/T, that is the mean overall
f 2 (t, ␶ ) as the product densities of degree one and two which may rate of occurrence.1 However the rate of convergence is extremely
be interpreted as the instantaneous mean rate of impacts respec- small: considering a percentage of random fluctuation of x/100
tively at time t and at time t plus ␶. Then, ⫽ ␴ ⌬ /T, then two peaks completely overlap when their band-
E 兵 dN 共 t 兲 n 其 ⫽ f 1 共 t 兲 dt, ᭙n⬎0 in N (3) 1

An heuristic proof to this result is that, in the limit, the area under each prob-
ability density function is still unity whereas the mean paving is one probability
E 兵 dN 共 t⫹ ␶ 兲 dN 共 t 兲其 ⫽ f 2 共 t, ␶ 兲 dtd ␶ , ␶ ⬎0 (4) density function per T units of time.

Fig. 3 Fourier transform „modulus… of the product density of
degree one for ␴ ⌬ Õ T Ä1Õ30
width is greater than their mutual spacing, that is when 冑 i• ␴ ⌬

⬎T or i⬎10000/x 2 . For random fluctuations of typically a few
percent, this means it takes a few thousands peaks for f 1 (t) to
reach its limit.
Similarly, the product density of degree two can be verified to
tend to f 2 (⬁, ␶ )⫽ f 1 ( ␶ )/T. This supports the assertion of refer-
ence 关6兴 where it was pointed out that the nonstationary impacting
Fig. 4 Double Fourier transform „modulus… of the product den-
process F(t) ultimately tends to stationarity, yet so slowly that the sity of degree two for ␴ ⌬ Õ T Ä1Õ30
phenomenon is hardly noticeable in practice over a finite time of
observation.
2.3 Spectral Characteristics. Since most processing in- (ii) Quality factor. The ‘‘apparent increase’’ of the bandwidth
volved in the diagnosis of rolling element bearings is performed in resulting from the fall off of the peaks may be quantified by the
the Fourier domain, it is now necessary to derive the formulas for quality factor Q i ⬇T 2 /(i• ␲ • ␴ ⌬ ) 2 共ratio of the maximum to the
the Fourier transforms of the product densities f 1 (t) and f 2 (t, ␶ ). minimum of the ith peak兲. This is naturally found to strongly
(a) Fourier Transform of the Product Density of Degree One depend on the percentage of random fluctuation ␴ ⌬ /T.
Combining Eq. 共5兲 and Eq. 共9兲, the Fourier transform of f 1 (t) is (b) Double Fourier Transform of the Product Density of De-
readily found to yield a continuous density, gree Two
冕
The formula for F 1 ( ␣ ) can now be used to compute the double
1
F 1共 ␣ 兲 ⫽ f 1 共 t 兲 e ⫺ j ␣ •t dt⫽ 关 1⫺⌽ * 共 ␣ 兲兴 ⫺1 (10) Fourier transform of f 2 (t, ␶ ):
2␲
冕冕
R
1
where ⌽共␣兲 is the characteristic function of the first time of arrival F 2共 ␣ , ␻ 兲 ⫽ f 2 共 t, ␶ 兲 e ⫺ j ␣ •t e ⫺ j ␻ • ␶ dtd ␶ (12)
4␲2 R2
T 1 —or equivalently of the independent and identically distributed
inter-arrival time process 兵 ⌬T i 其 . Equation 共10兲 is known as a Distinguishing three cases ␶ ⬍0, ␶ ⫽0 and ␶ ⬎0 and after some
‘‘renewal type’’ equation in the theory of stochastic processes, the algebra, one finds
study of which requires the exact knowledge of the probability
F 2 共 ␣ , ␻ 兲 ⫽F 1 共 ␣ 兲 • 关 F 1 共 ␣ 兲 ⫹F 1 共 ␣ ⫺ ␻ 兲 ⫺1 兴 (13)
law governing T 1 . The Gamma law would probably be a good
candidate here as it produces strictly positive inter-arrival times This defines a continuous spectral density with marked ridges run-
with a peaked probability around the mean value T. However ning along the ␻-variable and centered on all ␣ ⫽k/T, k苸Z. For
when its variance ␴ ⌬2 is small w.r.t. its mean T, the Gamma dis- large values of ␻, F 2 ( ␣ , ␻ ) ultimately tends to F 1 ( ␣ ) and thus
tribution is well approximated by the Normal distribution with the resembles a pattern of parallel and horizontal ridges along the
same mean and variance, thus making the calculations more trac- ␻-axis rapidly falling off on each side of ␣ ⫽0, as illustrated in
table. Under these assumptions, Fig. 4. The presence of these parallel ridges distinctively charac-
冉冊
terizes the 共second-order兲 spectral signature of a random train of
1 2 2 impact forces and consequently, that of a faulty rolling element
⌽ 共 ␣ 兲 ⬇exp ⫺ ␴ • ␣ ⫺ j ␣ •t (11)
2 ⌬ bearing. Moreover, the distance between the ridges indicates the
mean rate of occurrence of the fault, thus enabling its identifica-
from which Eq. 共10兲 is readily found to yield a pole at ␣ ⫽0 and tion in the mechanical system.
a series of finite-energy peaks equi-spaced by 1/T, with maxima
and minima respectively on ␣ ⫽k/T and ␣ ⫽k/T⫺1/2, k苸Z. Fig- 2.4 Discussion. At this stage, it is instructive to review
ure 3 depicts the behavior of the modulus 兩 F 1 ( ␣ ) 兩 where the some former models proposed in the literature in light of the de-
percentage of random fluctuation ␴ ⌬ /T has been set rather large rived results. Clearly, for the deterministic model f 1 (t) is a per-
for sake of demonstration. In contrast to the time domain, note fectly periodic train of Dirac deltas XT (t) and f 2 (t, ␶ ) a two-
firstly 共i兲 that the magnitude of the peaks falls off more rapidly in dimensional version of it, viz f 2 (t, ␶ )⫽XT (t)•XT ( ␶ ). The same
the Fourier domain and secondly, 共ii兲 that the bandwidth of the applies to their respective Fourier transforms, viz F 1 ( ␣ )⫽1/T
successive peaks of F 1 ( ␣ ) remain more or less constant since •X1/T ( ␣ ) and F 2 ( ␣ , ␻ )⫽1/T 2 •X1/T ( ␣ )•X1/T ( ␻ ). The limita-
they are bounded by 1/T. tion of these formulas arises from the experimental evidence that
(i) Fall off of the peaks. For ␴ ⌬2 ⰆT, the relative magnitude of actual data do not have line spectra especially in the vicinity of
the ith peak w.r.t. the first one decreases almost as fast as 1/i 2 , that the high frequency resonance where they are usually demodu-
is a slope of ⫺40 dB per decade. Since F 1 ( ␣ ) ultimately tends lated. On the other hand, the simplified stochastic model proposed
towards a constant amplitude density F 1 (⬁)⫽1, this means that by Randall & Antoni in 关5兴 leads to f 1 (t)⫽XT (t) * ␾ 1 (t), that is
there exists a cut-off radian frequency ␣ c ⬇ 冑 2/␴ ⌬ after which all a periodic train of Dirac delta functions low-pass filtered by the
the peaks have faded. Considering the percentage of random fluc- probability function ␾ 1 (t). Therein f 2 (t, ␶ ) turns out to be peri-
tuation x/100⫽ ␴ ⌬ /T, it is found that F 1 ( ␣ ) becomes almost con- odic and low-pass filtered in the t-variable while transient in the
stant after i c ⫽22.5/x. For example, for a random fluctuation of ␶-variable. These ‘‘low-pass filter’’ and ‘‘transient’’ effects give a
2%, this means as few as i c ⫽11 peaks. better explanation for the continuity of experimental spectra in the

vicinity of a high frequency resonance 关4兴. The refined stochastic P2: b(t) has an effective duration shorter than the mean inter-
model proposed herein obviously leads to almost identical prop- arrival time T, or equivalently its spectral bandwidth is larger
erties, yet on the basis of more accurate physical considerations. than the mean rate of impacts 1/T.
Of particular concern are the facts that f 1 (t) and f 2 (t, ␶ ) are no Therefore, the overall impulse response of the system is ob-
longer periodic functions neither in the t nor in the ␶ variables and tained from cascading the amplitude modulation function A(t)
that the ‘‘low-pass filter’’ effect is now replaced by a rapid ‘‘fall- with the time-varying impulse response r(t, ␶ ) and finally with the
off’’ effect. In other words, the Fourier transforms F 1 ( ␣ ) and band-pass filter b(t). This is illustrated in Fig. 5共b兲.
F 2 ( ␣ , ␻ ) are now purely continuous functions in both ␣ and ␻ In this procedure, A(t) and g(t, ␶ )⫽b(t⫺␭)r(␭, ␶ ) have some
共except at ␣ ⫽0) where all the former discrete lines 共harmonics兲 important properties which will make the computation of the
have been replaced by distributed peaks, gradually broadening input-output relationship tractable. Specifically, because A(t) en-
with increase in ␣. Actually, this fact is always observable in compasses all the periodic modulations with possible stochastic
envelope spectra, and was one of the main reasons for modifying effects, it has first and second-order statistics given by:
冉冊
the stochastic model first mentioned in reference 关5兴.
2␲
m A 共 t 兲 ⫽E 兵 A 共 t 兲其 ⫽m A t⫹
⍀
⫽
k苸Z
兺
a k e j⍀•t (14)
3 Spectral Statistics of the Vibration Signal
and
冉冊
In the preceding section, expressions were derived which de-
2␲
兺
scribe the spectral signature of a train of pulses as produced by a
R A 共 t, ␶ 兲 ⫽E 兵 A 共 t⫹ ␶ 兲 A * 共 t 兲其 ⫽R A t⫹ ,␶ ⫽ R Ak 共 ␶ 兲 e j⍀•t
localized defect. This section now discusses how this spectral sig- ⍀ k苸Z
nature is transformed after the impacts have propagated through (15)
the system, i.e. as it is likely to be measured on the bearing hous-
ing by an accelerometer. Equations 共14兲 and 共15兲 define a second-order cyclostationary
process, i.e. a stochastic process with periodic mean and autocor-
relation function of intrinsic period 2␲/⍀. Therein ⍀ is either
3.1 Response of a Rolling Element Bearing to a Random
equal to the speed of the inner race, that of the outer race or that
Train of Impacts. Following classical models, the vibration sig-
of the cage 共relative to the load vector兲 whether the fault is on the
nal produced by a faulty rolling element bearing may be viewed
inner race, the outer race, or on a rolling element.
as the response of a linear system driven by the impacting process
Similarly, g(t, ␶ ) being a periodic causal Green’s function de-
F(t) 关2,7兴. For this input-output relationship to be fully compre-
scribing the periodically varying transmission path, it expands
hensive, we now show that the impulse response of the system
into:
再
should be time-varying and should also accommodate some de-
gree of stochasticity.
At the outset, the impacting process should be modulated by a
periodic and positive function A(t) to account for the variations in g 共 t, ␶ 兲 ⫽
冉
g t⫹
2␲
⍀
,␶ ⫽ 冊
k苸Z
兺
g k 共 t⫺ ␶ 兲 e j⍀•t , ␶ ⭐t
(16)
the impact magnitudes as the defect enters and exits the load zone 0 ␶ ⬎t
关2兴. Some random modulation might be incorporated in A(t) due
to the dependence on the position and the number of the rolling From the above expansion, the mechanism relating the impacting
elements in the load zone at time t, but also due to rolling and slip process F(t) to the band-pass vibration signal Y (t) can finally be
on possibly rough surfaces especially after a defect has appeared obtained from the following Stieltjes stochastic integral
and spread to some extent.
Next, let us define r(t, ␶ ) the structural response at time t of the
system subjected to an impulse ␦共␶兲 at time ␶. In contrast to a
Y 共 t 兲⫽ 冕0
t
g 共 t, ␶ 兲 A 共 ␶ 兲 dN 共 ␶ 兲 ⫽ 兺e
k苸Z
j⍀•t
冕 R
g k 共 t⫺ ␶ 兲 A 共 ␶ 兲 dN 共 ␶ 兲
static structure, the impulse response r(t, ␶ ) of a rolling element (17)

bearing is time-varying for a variety of physical reasons, the most in which each Fourier coefficient g k (t) is to be interpreted as a
obvious of which being the variations in the transmission path as linear, causal and homogeneous impulse response.
the coordinates of the point of impact move w.r.t the location of
the sensor, and the variations in the relative angle between the
impact forces and the axis of the sensor. For a system operating at 3.2 Spectral Characteristics of the Vibration Response.
constant speed, these variations periodically affect the magnitude, From Eq. 共17兲, the spectral characteristics of the band-pass vibra-
the sign and the phase of the impulse response r(t, ␶ ) with a tion signal can now be derived and applied to a number of poten-
period depending on whether the defect lies on the inner race, the tial indicators for use in diagnostics, namely the Fourier transform
outer race or on a rolling element 关2兴. In addition, r(t, ␶ ) might of the expected signal, the power spectral density, the spectral
have some small random fluctuations to account for unpredictable correlation, the Fourier transform of the expected squared signal
effects such as contact non-linearities. Figure 5共a兲 gives a sche- and the power spectral density of the squared signal.
matic illustration of how the impacting process A(t)F(t) is trans- (a) Fourier Transform of the Expected Response
formed into a vibration X(t) after passing through the impulse From Eqs. 共1兲, 共14兲 and 共17兲, the expected value 共ensemble
response r(t, ␶ ). average兲 of the vibration signal is
冕
In practice, the vibration signal X(t) produced by a local fault
t
cannot be observed totally because it is contaminated by other
m Y 共 t 兲 ⫽E 兵 Y 共 t 兲其 ⫽ g 共 t, ␶ 兲 m A 共 ␶ 兲 f 1 共 ␶ 兲 d ␶ (18)
vibrations from a multitude of neighboring sources in the system. 0
Therefore, it is customary to filter it in a frequency band where the
signal-to-noise ratio is maximum so that virtually no other sources from which the Fourier transform is found to be:
冕
than that stemming from the faulty bearing are measured by the
冦
1
experimenter. This is usually done by designing a band-pass filter
b(t) around a high-frequency resonance of the structure 共or the
M Y共 ␣ 兲⫽
2␲ R
m Y 共 t 兲 e ⫺ j ␣ •t dt⫽ 兺M
k苸Z
Y k 共 ␣ ⫺k⍀ 兲
sensor兲 that is excited by the impacts 关1– 4兴. In order to retain the
diagnostic information, the band-pass filter b(t) must have the M Y k 共 ␣ 兲 ⫽G k 共 ␣ 兲 •F̃ 1 共 ␣ 兲 (19)
following properties:
P1: b(t) is a band-pass filter with central frequency ␻ 0 much
higher than the mean rate of impacts 1/T,
F̃ 1 共 ␣ 兲 ⫽M A 共 ␣ 兲 * F 1 共 ␣ 兲 ⫽ 兺 a •F 共 ␣ ⫺l⍀ 兲
l苸Z
l

Fig. 5 „a… Generation of the resulting vibration signal. A „ t …: magnitude of the
impacts; r „ t , ␶ …: time-varying „stochastic… structural impulse response at time t .
„b… Scheme of the overall impulse response. b „ ␶ … is a band-pass filter that ex-
tracts the bearing signal where its signal-to-noise ratio is the highest
where G k ( ␣ ) and M A ( ␣ ) are respectively the Fourier transforms to overlap with this support, its central radian frequency ␻ 0 共reso-
of g k (t) and m A (t). In view of Eq. 共19兲, M Y ( ␣ ) is a superposition nance frequency chosen for demodulation兲 should be such that
of shifted functions M Y k ( ␣ ); each of them being in turn con- ␻ 0 •T⬍100&/x with x/100⫽ ␴ ⌬ /T the percentage of fluctuation.
structed from shifted and scaled replicas of F 1 ( ␣ ) and then Or equivalently, with i c the number of peaks in F 1 ( ␣ ) before it
weighted by the frequency response G k ( ␣ ). The construction of dies to 1, ␻ 0 should be such that ␻ 0 •T⬍2 ␲ •i c . In most instances
M Y ( ␣ ) results in a mixed spectrum with a family of infinite- this condition would not be satisfied if a good signal-to-noise ratio
energy pseudo-harmonics around ␣ ⫽0, all equi-spaced by the ro- were to be maintained, thus justifying the poor performance an-
tation speed ⍀. The number of these pseudo-harmonics directly ticipated from the Fourier transform of the vibration signal.
depends on the number of Fourier coefficients in g(t, ␶ ) and (b) Power Spectral Density of the Response
m A (t). Note that this specific pattern repeats around all the peaks From Eqs. 共4兲, 共15兲 and 共17兲, the autocorrelation function of the
of F 1 ( ␣ ) at ␣ ⫽k/T, k苸Z, but with finite-energy peaks in place vibration response is:
of pseudo-harmonics for any k⫽0. Obviously, this makes the very
distinctive ‘‘spectral signature’’ of a localized defect as it is ex-
pected to appear in a faulty rolling element bearing 共see Fig. 6兲.
Most importantly its detection forms the main basis of diagnostics
since it contains the key characteristic frequencies 1/T and ⍀ that
enable the identification and the localization of faults in complex
systems.
However, the problem in the spectral indicator of Eq. 共19兲 is
that the frequency support of the band-filters G k ( ␣ ) is very likely
to be higher than the frequency support of the spectral signature
F 1 ( ␣ ) as shown schematically in Fig. 7. In fact, it was already
pointed out that F 1 ( ␣ ) falls off by ⫺40 dB per decade down to a Fig. 6 Typical spectral signature in the vibration signal for
cut-off radian frequency ␣ c of about &/ ␴ ⌬ . In order for G k ( ␣ ) ␴ ⌬ Õ T Ä1Õ30 and ⍀Ä T Õ10

Fig. 7 Illustration of the low-pass filter effect
R Y 共 t, ␶ 兲 ⫽E 兵 Y 共 t⫹ ␶ 兲 Y * 共 t 兲其 ⫽ 冕冕0
t
0
t⫹ ␶ ⫺ ␯
h 共 t⫹ ␶ , ␯ Fig. 8 Scheme of the spectral correlation density
⫹␭ 兲 h * 共 t, ␯ 兲 R A 共 ␯ ,␭ 兲 f 2 共 ␯ ,␭ 兲 d␭d ␯ (20)
This is a bivariate function since the vibration signal resulting where the diagnostic information is totally preserved. Specifically,
from the impacting process F(t) is nonstationary. In order to com- for small values of ␣ within twice the bandwidth of g(t, ␶ ) and
pute the power spectral density, let us first denote by R̄ y ( ␶ ) the large values of ␻ around the resonance frequency ␻ 0 , S Y ( ␣ , ␻ )
‘‘stationarized’’ autocorrelation function clearly displays the spectral signature of a defect because the
band-pass filters G k ( ␻ ) and G l ( ␻ ⫺ ␣ ) are band-passing in this
R̄ Y 共 ␶ 兲 ⫽ lim
W→⬁
1
W 冕0
W
R y 共 t, ␶ 兲 dt (21) area. This is a direct consequence of property P2, a schematic
illustration of it being shown in Fig. 8. Note that in the domain of
concern S Y ( ␣ , ␻ ) is markedly ridged in the horizontal ␻-direction
whose Fourier transform then yields the explicit expression for the just as F 2 ( ␣ , ␻ ) was in Fig. 4.
power spectral density: (d) Fourier Transform of the Expected Squared Response
冕 Because it is bivariate, the spectral correlation density may be
冦
1
SY共 ␻ 兲⫽
2␲ R
R̄ Y 共 ␶ 兲 e ⫺ j ␻ • ␶ d ␶ ⫽ 兺S
k苸Z
y k 共 ␻ ⫺k⍀ 兲 difficult to compute and therefore it has been suggested to replace
it by its integrated version over the ␻-variable 共while preserving
the diagnostic information兲. In reference 关5兴, this was shown to be
S Y k 共 ␻ 兲 ⫽ 兩 G k 共 ␻ 兲兩 2 •F̃ 2 共 ␻ 兲 (22)
equivalent to the Fourier transform of the expected squared signal,
2 i.e.
F̃ 2 共 ␻ 兲 ⫽ Re兵 F 1 共 ␻ 兲其 * S A 共 ␻ 兲
T
where S A ( ␻ ) is the Fourier transform of the stationarized version
M Y 2共 ␣ 兲 ⫽ 冕
R
SY共 ␣,␻ 兲d␻⫽
1
2␲ 冕 R
E 兩 Y 共 t 兲兩 2 e ⫺ j ␣ •t dt (25)
R̄ A ( ␶ ) of R A (t, ␶ ) in Eq. 共15兲. The set of Eq. 共22兲 indicate that the
principle of construction of S Y ( ␻ ) is similar to that outlined for This equation is easily found to be identical in structure to Eq.
the Fourier transform in Eq. 共19兲 because S A ( ␻ ) contains the 共19兲 where the coefficients a 1 are replaced by R Al (0) defined in
same discrete harmonics as F A ( ␣ ) and 兩 G k ( ␻ ) 兩 2 obviously acts in Eq. 共15兲, and where the band-pass filter G k ( ␣ ) is replaced by the
the same frequency band as G k ( ␣ ). Therefore, the same conclu- low-pass filter P k ( ␣ )⫽⌺ p G k ( ␣ )•G k⫺p ( ␣ ). The fact that P k ( ␣ )
sion holds in regard to the expected performance of the power is now necessarily a low-pass filter comes from the convolution of
spectral density as a diagnostic indicator. G k ( ␣ ) by itself and this is exactly the reason why M Y 2 ( ␣ ) can
(c) Spectral Correlation Density of the Response preserve the diagnostic information whereas M Y ( ␣ ) cannot. In-
We now demonstrate that the aforementioned shortcomings due deed, under property P2 the support of P k ( ␣ ) necessarily overlaps
to the non-intersection of the low-pass and band-pass filters 共see. with that of the spectral signature of the fault, contrary to the
Fig. 7兲 can be solved by considering the double Fourier transform scheme of Fig. 7.
of the autocorrelation function R Y (t, ␶ ). This yields a quantity A last point to consider is whether to take the square of the raw
called the spectral correlation density2 关5兴, very similar to the signal in Eq. 共25兲 or the squared magnitude of its analytic version.
‘‘generalized spectrum’’—within a simple change of variable— Strictly speaking, the analytic signal should be used so that the
used by Lin 关9兴. The spectral correlation density, expectation of its squared magnitude truly gives the squared en-
冕
velope. However, minor differences would be found when using
1
SY共 ␣,␻ 兲⫽ R Y 共 t, ␶ 兲 e ⫺ j ␻ • ␶ e ⫺ j ␣ •t d ␶ dt (23) the real signal provided it is properly band-pass filtered around a
4␲2 R resonance. This point was also addressed in a lot of detail in
reference 关4兴.
is found to have explicit expression (e) Power Spectral Density of the Squared Response
冦
In light of the previous demonstration, one can expect the
SY共 ␣,␻ 兲⫽ 兺 S y k y l 共 ␣ ⫺k⍀, ␻ ⫺l⍀ 兲 power spectral density S Y 2 ( ␻ ) of the squared signal to perform
k,l苸Z2 just as well as M Y 2 ( ␣ ). Indeed, S Y 2 ( ␻ ) is the exact definition of
the ‘‘spectrum of the squared envelope’’ as was proposed in 关4兴. In
S Y k Y l 共 ␣ , ␻ 兲 ⫽G k 共 ␻ 兲 G l* 共 ␻ ⫺ ␣ 兲 •F̃ 2 共 ␣ , ␻ 兲 (24)
order to prove this result, let us invoke property P1 in conjunction
with the assumption that the point process 兵 dN(t) 其 is regular 共see
F̃ 2 共 ␣ , ␻ 兲 ⫽ 兺 F 共 ␣ ⫺p⍀, ␻ 兲 * S 共 ␻ 兲
2
p
A section 2.1兲. Then, the following approximation holds
冕
p苸Z
t
where S Ap ( ␻ ) is the Fourier transform of R Ap ( ␶ ) in Eq. 共15兲. Al- 兩 Y 共 t 兲兩 2⬇ p 共 t, ␶ 兲兩 A 共 ␶ 兲兩 2 dN 共 ␶ 兲 (26)
though involving two frequency variables, the construction of 0
S Y ( ␣ , ␻ ) is again similar to that outlined in Eqs. 共19兲 and 共22兲.
Nevertheless, there is now a domain in the frequency plane 共␣, ␻兲 with p(t, ␶ )⫽ 兩 g(t, ␶ ) 兩 2 . After taking the Fourier transform of the
‘‘stationarized’’ autocorrelation function of 兩 Y (t) 兩 2 , the envelope
2
There is a simple relation between the spectral correlation density and the power spectrum is found to be identical to Eq. 共22兲 where S A ( ␻ ) is
spectral density, viz S Y (0,␻ )⫽S Y ( ␻ ) ␦ ( ␣ ) replaced by S A 2 ( ␻ )—the Fourier transform of the stationarized

Fig. 9 Power spectral density of a vibration signal in case of no fault „continuous line… and an
inner race fault „dotted line…
autocorrelation function R̄ A 2 ( ␶ ) of the squared process ture of the inner race fault shows up in good accordance with Fig.
兵兩 A(t) 兩 2 其 —and where 兩 G k ( ␻ ) 兩 2 is replaced by 兩 P k ( ␻ ) 兩 2 . The fact 4, with 1/T⫽71 Hz 共ball pass frequency on the fault兲 and ⍀
that 兩 P k ( ␣ ) 兩 2 acts as a low-pass filter demonstrates again that ⫽10 Hz 共shaft rotation兲.
S Y 2 ( ␻ ) is also a usable diagnostic indicator. Here again, the ana- From a theoretical point of view, it is worth mentioning that the
lytic version of the signal may be preferred in Eq. 共26兲 in order to Fourier transform of the squared signal preserves the diagnostic
estimate the power spectral density of the true squared envelope. information by exploiting the non-stationarity of the signal, while
the power spectral density exploits its non-Gaussianity. In fact, the
3.3 Discussion. It has been proven in some depth why the power spectral density of the squared signal is implicitly a fourth-
Fourier transform and the power spectral density generally are order ‘‘stationarized’’ statistic. Strictly speaking, these two indica-
poor indicators for diagnosing rolling element bearings in the case tors have different theoretical justifications and this is supported
of localized faults, a fact that the authors have regularly observed by recalling that the former only requires property P2, while the
on experimental data. Indeed, even though classical spectral latter requires the more stringent condition P2⫹P1. However,
analysis may perform very well in detecting a fault—e.g. through both are inclined to provide envelope analysis—or ‘‘squared’’ en-
monitoring the relative energy levels in some frequency bands—it velope analysis—with a strong formal justification.
rarely helps in recognizing its type nor its location—and this is Comparison of the five spectral indicators, which have been
exactly what diagnostics asks for. For example, Figure 9 compares assessed so forth, is summarized in Table 1.
the power spectral densities of a vibration signal measured on a
gearbox before and after one of the rolling element bearings 共12
balls, ⭋7.12 mm, pitch circle ⭋38.5 mm兲 was purposely damaged
by machining a small slot on its inner race. The frequency reso- 4 Conclusion
lution is 12 Hz. Note that the presence of the fault only shows up A comprehensive stochastic model has been proposed for de-
at high frequencies. The fact that there is no difference at low scribing and simulating the vibration produced by localized faults
frequencies is due to the extremely poor signal-to-noise ratio in in rolling element bearings. Sources of stochasticity were modeled
that band 共observe that most of the sources there relate to harmon- in both the impacting force process—by means of a regular point
ics from the gears兲. Of interest also is the fact that in spite of its process—and in the transmission path—by means of a cyclosta-
increase, the spectral density at higher frequencies is continuous tionary process, thus encompassing a large range of physical situ-
and therefore gives no indication of a fault producing repetitive ations. These refinements proved very valuable in explaining
impacts. some of the actual features observed on experimental data. The
In clear contrast with the Fourier transform and the power spec- spectral signature of a localized fault was derived analytically and
tral density, the same transformations applied on the squared sig- new results were deduced concerning the nature of spectral har-
nal 共or its analytic version兲 have been shown to solve the problem monics produced by the impacting process. These were shown to
in a surprisingly simple manner. For example, Fig. 10 displays the be distributed and equi-spaced 共by the mean rate of impacts兲
power spectrum of the squared magnitude of the analytic signal peaks with a rapid fall-off that could be quantified as a function of
after band-pass filtering in the frequency band 关1.8; 2.2兴 kHz with the percentage of stochastic fluctuations. Next, the spectral signa-
a frequency resolution of 2 Hz. Now the specific spectral signature of a defect was shown to duplicate when it propagates
through the structure 共with shifts equal to the rotation speed of the
defect兲, thus generating additional families of pseudo-harmonics.
These results finally helped in investigating the effectiveness of a
number of spectral indicators dedicated to the diagnostics of roll-
ing element bearings. From simple considerations on band-pass
and low-pass filtering operations, it was demonstrated that both
the Fourier transform and the power spectral density of the
squared signal are the most relevant indicators, thus bringing new
supports in favor of ‘‘squared’’ envelope analysis.
Fig. 10 Power spectral density of the squared envelope
Table 1 Comparison of five spectral indicators in terms of

their ability of detecting and identifying localized faults. References
关1兴 Darlow, M. S., and Badgley, R. H., 1975, ‘‘Applications for Early Detection of
Spectral indicators Diagnostic skills Rolling Element Bearing Failures Using the High-Frequency Resonance Tech-
nique,’’ ASME Paper 75-DET-46.
Fourier transform of expected signal ⫺ 关2兴 McFadden, P. D., and Smith, J. D., 1984, ‘‘Model for the Vibration Produced
Power spectrum of the signal ⫺ by a Single Point Defect in a Rolling Element Bearing,’’ J. Sound Vib., 91共1兲,
Spectral correlation density 共2-D兲 of the signal ⫹⫹ pp. 69– 82.
Fourier transform of the expected squared signal ⫹⫹ 关3兴 McFadden, P. D., and Smith, J. D., 1985, ‘‘The Vibration Produced by Mul-
Power spectrum of the squared signal ⫹⫹ tiple Point Defects in a Rolling Element Bearing,’’ J. Sound Vib., 98共2兲, pp.
69– 82.

关4兴 Ho, D., and Randall, R. B., 2000, ‘‘Optimization of Bearing Diagnostics Tech- 关6兴 Antoni, J., and Randall, R. B., 2002, ‘‘Differential Diagnosis of Gear and
niques Using Simulated and Actual Bearing Fault Signals,’’ Mech. Syst. Signal Bearing Faults,’’ ASME J. Vibr. Acoust., 127, pp. 1–7.
Process., 14共5兲, pp. 763–788. 关7兴 Roberts, J. B., 1966, ‘‘On the Response of a Simple Oscillator to Random
关5兴 Randall, R. B., Antoni, J., and Chobsaard, S., 2001, ‘‘The Relationship Be- Impulses,’’ J. Sound Vib., 4共1兲, pp. 51– 61.
关8兴 Srinivasan, S. K., et al., 1967, ‘‘Response of Linear Vibratory Systems to
tween Spectral Correlation and Envelope Analysis in the Diagnostics of Bear-
Non-Stationary Stochastic Impulses,’’ J. Sound Vib., 6共2兲, pp. 169–179.
ing Faults and other Cyclostationary Machine Signals,’’ Mech. Syst. Signal 关9兴 Lin, Y. K., 1965, ‘‘Nonstationary Excitation and Response in Linear Systems
Process., 15共5兲, pp. 945–962. Treated as Sequences of Random Pulses,’’ J. Acoust. Soc. Am., pp. 453– 460.

Nonlinear Dynamic Analysis
of a Rotor Shaft System With
Nabeel Shabaneh
Viscoelastically Supported
Graduate Student
Bearings
Jean W. Zu
Associate Professor This research investigates the dynamic analysis of a single-rotor shaft system with non-
linear elastic bearings at the ends mounted on viscoelastic suspension. Timoshenko shaft
Department of Mechanical & Industrial model is utilized to incorporate the flexibility of the shaft; the rotor is considered to be
Engineering, rigid and located at the mid-span of the shaft. A nonlinear bearing pedestal model is
University of Toronto, assumed which has a cubic nonlinear spring and linear damping characteristics. The
5 King’s College Road, viscoelastic supports are modeled using Kelvin-Voigt model. Free and forced vibration is
Toronto, Ontario, Canada, M5S 3G8 investigated based on the direct multiple scales method of one-to-one frequency-to-
amplitude relationship using third order perturbation expansion. The results of the non-
linear analysis show that a limiting value of the internal damping coefficient of the shaft
exists where the trend of the frequency-response curve switches. Also, the primary reso-
nance peak shifts to higher frequencies with the increase of the bearing nonlinear elastic
characteristics, but with a flattened curve and hence lower peak values. A jump phenom-
enon takes place for high values of the bearing nonlinear elastic characteristics.
关DOI: 10.1115/1.1547684兴
1 Introduction and the ring 关9兴. Ji and Zu 关10兴 performed free and forced vibra-
tion analysis to calculate natural frequencies for nonlinear rotor
An effective means for controlling and reducing vibrations in
bearing system with cubic nonlinearity. However, incorporating
rotating machinery is the use of external damping and elastic el- nonlinear bearing characteristics in rotor-shaft systems with vis-
ements often provided via flexible bearings and/or bearing sup- coelastic supports is very limited. A preliminary investigation was
ports. Viscoelastic materials have emerged recently as external reported by Bhattacharyya and Dutt 关11兴. They studied the unbal-
mechanical dampers and bearing supports, offering simplicity in anced response and stability of a rotor shaft system mounted on
design, operational enhancement and low cost in comparison with nonlinear rolling element bearings with viscoelastic supports. The
the traditionally used squeeze film dampers 关1兴. shaft was assumed to be massless with linear elasticity and inter-
Modeling of the dynamic behavior of vibrating systems incor- nal damping. In addition, the nonlinear restoring force of the bear-
porating viscoelastic elements has not been possible until recently ings was linearized by the method of effective linearization, en-
due to the elaborate work of identifying the characteristic behav- abling an approximate stability analysis using the Routh Hurwitz
ior of viscoelastic materials. Although there is much research on criterion. The effects of gravity on the system and nonlinearity
modeling systems for structural applications, the work on rotating were discussed. In the above work, the system was discretized and
shaft bearing systems mounted on viscoelastic supports is limited. the restoring force was approximated by a linear relation.
In the few works reported on linear analysis, Dutt and Nakra The aim of this research is to investigate the nonlinear dynamic
关2– 4兴 studied the stability and the unbalance vibration response of behavior of a continuous rotor shaft system with viscoelastically
a Jeffcott rotor system mounted on viscoelastic supports where a supported bearings. The nonlinearity occurs at the boundaries due
discrete model of the shaft and linear elastic bearings were as- to nonlinear characteristics of the bearings. Timoshenko shaft
sumed. Using the finite element method based on the Euler- model is used for the shaft, Kelvin Voigt model is utilized for the
Bernoulli beam model, Kulkarni et al. 关5兴 investigated the unbal- viscoelastic supports, and typical roller bearings with cubic non-
ance response and the stability of a rotating system with linearity are employed. In the analysis, free and forced vibration
viscoelastically supported bearings. In further investigations, Sha- are developed based on the direct multiple scales method of one-
baneh and Zu 关6 – 8兴 studied the dynamic behavior of a rotating to-one frequency-to-amplitude relationship of nonlinear system
disk-shaft system with linear elastic bearings at the ends mounted using the third order perturbation expansion. This approach does
on viscoelastic supports. Timsohenko model was assumed for the not require the selection of an orthogonal basis; however, it re-
shaft, and thus, the gyroscopic effect of the shaft was incorporated quires additional eigenfunction solutions at every level of approxi-
in the analysis. Free and forced vibration analysis were carried out mations. Moreover, the direct multiple scales method yields better
and the effects of the system characteristics were presented. results than the discretized one for finite mode truncations and for
Routh-Hurwitz criterion for polynomials with complex coeffi- systems having quadratic and cubic nonlinearities 关12兴. The re-
cients and the variational analysis were used to perform the sta- sults of the nonlinear analysis indicate that a limiting value of the
bility analysis of the system. Furthermore, experimental investi- internal damping coefficient of the shaft exists where the trend of
gations were performed to verify the obtained theoretical results. the frequency-response curve switches. Also, a jump phenomenon
Bearings in rotor shaft systems may possess nonlinear charac- takes place for high values of the nonlinear elastic coefficient of
teristics. For example, nonlinearity is inherent in ball bearings due the bearings.
to Coulomb friction and the angular clearance between the roller
2 Equations of Motion
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received February Consider a continuous shaft-rotor system, as shown in Fig. 1,
2002; Revised October 2002. Associate Editor: G. T. Flowers. where the frame oxyz is the inertial frame. A uniform shaft of
Fig. 1 Rotor shaft system with viscoelastically supported
bearings
length 2l is mounted on viscoelastically supported bearings. The

rotor is considered as a rigid disk. For convenience, the viscoelas-
tic element is represented by its equivalent complex stiffness k *
Q 共 1,t 兲 ⫽
1
2 冋
⫺M d
⳵ 2 u 共 1,t 兲
⳵t2
⫹M d e⍀ 2 exp共 i⍀t 兲册 (9)
and is located at both ends of the shaft rotor system underneath ␺ 共 1,t 兲 ⫽0 (10)
the bearings as support.
Adopting a continuous model of the shaft based on the Timosh- where M d and e are the mass of the disk and the eccentricity of
enko beam theory, and defining u x , u y , ␺ x , and ␺ y as the trans- the rotor, respectively. F l and F n are the linear and nonlinear
verse deflections along the ox and oy directions and the corre- terms exerted by the bearings on the shaft and the viscoelastic
sponding bending angles in the oxz and oyz planes, respectively, supports. Assuming that the bearings possess cubic nonlinear re-
the complex variables u and ␺ are assumed as storing forces in addition to linear damping forces, the associated
forces are
u⫽u x ⫹iu y , ␺ ⫽ ␺ x ⫹i ␺ y
冉冊
(1)
⳵u ⳵v
The general equations of motion of a uniform Timoshenko shaft F l ⫽F xl ⫹iF yl ⫽k b1 共 u⫺ v 兲 ⫹c b ⫺ (11)
system with internal damping are ⳵t ⳵t
␳ Aü⫹ 关 ␬ AG 共 ␺ ⫺u ⬘ 兲兴 ⬘ ⫹C i 共 u̇⫺i⍀u 兲 ⫽0 (2) F n ⫽F xn ⫹iF yn ⫽k b3 关共 u x ⫺ v x 兲 3 ⫹i 共 u y ⫺ v y 兲 3 兴 (12)

Equation 共14兲 can be rewritten in terms of the complex trans-
␳ I s ␺¨ ⫺i⍀J s ␺˙ ⫺ 共 EI s ␺ ⬘ 兲 ⬘ ⫹ ␬ AG 共 ␺ ⫺u ⬘ 兲 ⫽0 (3) verse deflections u and v , and their corresponding complex con-
where ( • )⫽D t ⫽ ⳵ / ⳵ t and ( ) ⬘ ⫽D z ⫽ ⳵ / ⳵ z. A, I, J s and ␳ are the jugates ū and v̄ , as
cross-sectional area, transverse moment of inertia, polar mass mo-
ment of inertia per unit length and mass density of the shaft. C i is F n ⫽k b3 关 43 共 u 2 ū⫺2uū v ⫹ v 2 ū⫺u 2 v̄ ⫹2u vv̄ ⫺ v 2 v̄ 兲
the internal viscous damping coefficient and ⍀ is the spin rate of 1
the shaft. E, G and ␬ are Young’s modulous, shear modulous and ⫹ 4 共 ū 3 ⫺3ū 2 v̄ ⫹3ū v̄ 2 ⫺ v̄ 3 兲兴 (13)
shear coefficient, respectively.
Introducing a set of nondimensional quantities given in Appen-
The viscoelastic support is assumed to have a mass of M 2 and
dix A, the nondimensional equations of motion with the associated
is modeled using Kelvin-Voigt model. Thus, the equivalent com-
boundary conditions can be written as
plex stiffness is defined as k * ⫽k v (1⫹i ␩ ). Denoting the displace-
ments of the viscoelastic mass M 2 along x and y directions by v x
and v y , respectively, the complex deflection of the viscoelastic
element is given by v ⫽ v x ⫹i v y .
⳵ 2u * ⳵ ␺ * ⳵ 2u *
⳵ t *2
⫹
⳵␨
⫺
⳵␨ 2
⫹C i
⳵u*
⳵t* 冉
⫺i⍀ * u * ⫽0 冊 (14)
Introducing the nondimensional space variable ␨ ⫽z/l, the re-

lationship between the stress resultants and displacements in a
complex form are
⳵ 2␺ *
⳵t* 2 ⫺i ␣ 1 ⍀ *
⳵␺*
⳵t*
⫺␣2
⳵ 2␺ *
⳵␨ 2
⫹ ␣ 3 ␺ *⫺
⳵u*
⳵␨
⫽0 冉冊 (15)
EI s ⳵ ␺ 共 ␨ ,t 兲 M * 共 0,t 兲 ⫽0 (16)
M 共 ␨ ,t 兲 ⫽M x 共 ␨ ,t 兲 ⫹iM y 共 ␨ ,t 兲 ⫽
冉冊冏
(4)
l ⳵␨ ⳵u* ⳵v*
Q * 共 0,t * 兲 ⫽k b1
* 共 u * ⫺ v * 兲 ⫹c * ⫺ ⫹k *
冉冊
b3 W *
1 ⳵ u 共 ␨ ,t 兲
b
⳵t* ⳵t* 共 0,t * 兲
Q 共 ␨ ,t 兲 ⫽Q x 共 ␨ ,t 兲 ⫹iQ y 共 ␨ ,t 兲 ⫽ ␬ AG ⫺ ␺ 共 ␨ ,t 兲 (5) (17)
l ⳵␨
where M ( ␨ ,t) and Q( ␨ ,t) are the transverse bending moment and ⳵ v * 共 0,t 兲
2
M 2* ⫹k v* 共 1⫹i ␩ 兲v * 共 0,t 兲
the shear force at each cross section along the shaft. ⳵ t *2
冉冊冏
Due to symmetry of the system, only the left half of the rotor-
shaft system, i.e., 0⭐z⭐l, is considered. Thus, the associated ⳵u* ⳵v*
⫽k *
b1 共 u * ⫺ v * 兲 ⫹c b* ⫺ ⫹k b3
* W*
boundary conditions can be defined as ⳵t* ⳵t* 共 0,t * 兲
M 共 0,t 兲 ⫽0 (6) (18)

Q 共 0,t 兲 ⫽F l 共 0,t 兲 ⫹F n 共 0,t 兲 (7) 1 ⳵ u * 共 1,t 兲 1
2
Q * 共 1,t 兲 ⫽⫺ M* ⫹ F * ⍀ * 2 exp共 i⍀ * t * 兲 (19)
2 d ⳵ t *2 2
⳵ 2v
M2 ⫹k v 共 1⫹i ␩ 兲v ⫽F l 共 0,t 兲 ⫹F n 共 0,t 兲 (8)
⳵t2 ␺ * 共 1,t 兲 ⫽0 (20)

3 Method of Multiple Scales
The direct multiple scales method is used in this section to
D 20 ␺ 3 ⫺i ␣ 1 ␻ n* D 0 ␺ 3 ⫺ ␣ 2
⳵ 2␺ 3
⳵␨ 2 冉
⫹ ␣ 3 ␺ 3⫺
⳵u3
⳵␨ 冊
solve for the free and forced vibration of the system. Expansions
of the displacements u * , v * and the shear deformation ␺ * are ⫽⫺2D 2 D 0 ␺ 1 ⫹i ␣ 1 共 ␻ *
n D 2␺ 1⫹ ␴ D 0␺ 1 兲 (30)
assumed as at ␨ ⫽0
u * ⫽␧u 1 共 ␨ ,T 0 ,T 2 兲 ⫹␧ u 3 共 ␨ ,T 0 ,T 2 兲 ⫹ . . .
3
v * ⫽␧ v 1 共 ␨ ,T 0 ,T 2 兲 ⫹␧ v 3 共 ␨ ,T 0 ,T 2 兲 ⫹ . . .
3
(21)
(22)
⳵␺3
⳵␨
⫽0, ␣ 3
⳵u3
⳵␨ 冉
⫺ ␺ 3 ⫺k b1
* 共 u 3⫺ v 3 兲冊
␺ * ⫽␧ ␺ 1 共 ␨ ,T 0 ,T 2 兲 ⫹␧ 3 ␺ 3 共 ␨ ,T 0 ,T 2 兲 ⫹ . . . (23) ⫽c 苸 D 0 共 u 1 ⫺ v 1 兲 ⫹k b3
* W 1 , and
v 共 1⫹i ␩ 兲v 3 ⫺k b1 共 u 3 ⫺ v 3 兲
2 D 0 v 3 ⫹k *
2
where u i , v i and ␺ i (i⫽1,3, . . . ) are of order O(1), ␧ is a small M* *
dimensionless parameter, and T n ⫽␧ n t * (n⫽0,2, . . . ) is the nth ⫽⫺2M 2* D 0 D 2 v 1 ⫹c 苸 D 0 共 u 1 ⫺ v 1 兲 ⫹k *
b3 W 1 (31)
order time scale. T 0 is a fast-time scale characterizing motions
occurring at the spin rate ⍀ * and natural frequency ␻ * at ␨ ⫽1
n of the
rotor bearing system; ␻ n* ⫽ ␻ n 冑␳ l 2 / ␬ G. Furthermore, T 2 is a ␺ 3 ⫽0,
冉冊
slow-time scale characterizing the modulation of the amplitude
and phase due to nonlinearity, damping and resonance. ⳵u3 M*
d
␣3 ⫺␺3 ⫹ d D 2D 0u 1⫹ 2 F 苸␻ *
D 20 u 3 ⫽⫺M * 1 2
Considering the one-to-one primary resonance of the system, ⳵␨ 2 n
the damping forces, the forcing term and the excitation frequency
are assumed as ⫻exp共 i 共 ␻ n* ⫹␧ 2 ␴ 兲 t * 兲 (32)
c b* ⫽␧ 2 c 苸 , C i* ⫽␧ 2 C i苸 , F * ⫽␧ 3 F 苸 , ⍀ * ⫽ ␻ n* ⫹␧ 2 ␴ where W 1 ⫽ 41 共 u 21 ū 1 ⫺2u 1 ū 1 v 1 ⫹ v 21 ū 1 ⫺u 21 v̄ 1 ⫹2u 1 v 1 v̄ 1 ⫺ v 21 v̄ 1 兲

(24)
⫹ 41 共 ū 31 ⫺3ū 21 v̄ 1 ⫹3ū 1 v̄ 21 ⫺ v̄ 31 兲 .
where ␴ ⫽O(1) is a detuning parameter. The assumptions in Eq.
共24兲 takes into account the interaction of damping terms with the 3.1 First Order Equations. Noting that the equations of
nonlinear forces at the same level of approximation, which is a motion and the associated boundary conditions are linear at order
necessary condition for a nontrivial solution of the governing ␧, the first order approximation of the deflections and deforma-
equations of motion. Thus, the effect of the nonlinearity of the tions can be assumed in the form
system can be balanced with the effect of the system damping
at the same level of approximation, i.e., the third order u 1 共 ␨ ,t * 兲 ⫽A 共 T 2 兲 e i ␻ *n T 0 Y u 共 ␨ 兲 ,
approximation 关13兴.
The small dimensionless parameter ␧ may take any value be- ␺ 1 共 ␨ ,t * 兲 ⫽A 共 T 2 兲 e i ␻ *n T 0 Y ␺ 共 ␨ 兲 ,
v 1 共 t * 兲 ⫽A 共 T 2 兲 e i ␻ *n T 0 Y v
tween 0 and 1.0. However, by assuming a value of zero, the so- (33)
lution will be trivial. Moreover, in order to obtain a formal solu- Substituting Eq. 共33兲 into Eqs. 共25兲 and 共26兲 leads to a set of
tion of the equations of motion 共14兲–共20兲, ␧ should be equated to ordinary differential equations given by
1.0. In such a case, it is stipulated that the nonlinear terms are
a 1 Y ⬙u ⫹a 2 Y u⬘ ⫺a 3 Y ␺⬘ ⫽0 (34)
themselves small.
Using the chain rule for the partial derivatives with respect to b 1 Y ␺⬙ ⫹b 2 Y ␺⬘ ⫺b 3 Y ⬘u ⫽0 (35)
both time scales T 0 and T 2 , substituting Eqs. 共21兲–共24兲 into Eqs.
共14兲–共20兲, and equating coefficients of like power of ␧, the fol- where ( ) ⬘ ⫽d/d ␨ . The coefficients in Eqs. 共34兲 and 共35兲 are de-
lowing approximations of the system are obtained: fined as
Order ␧: a 1 ⫽1, a 2⫽ ␻ * 2
n , a 3 ⫽1 (36)
D 20 u 1 ⫹ 冉 ⳵␺1 ⳵ u1
⳵␨
⫺
⳵␨ 2
2
冊
⫹C i苸共 D 0 u 1 ⫺i ␻ n* u 1 兲 ⫽0 (25)
b 1⫽ ␣ 2 , b 2 ⫽ 共 1⫺ ␣ 1 兲 ␻ *
n ⫺␣3 ,
Decoupling Eqs. 共34兲 and 共35兲 and assuming an exponential
2
b 3 ⫽⫺ ␣ 3 (37)
D 20 ␺ 1 ⫺i ␣ 1 ␻ *
n D 0␺ 1⫺ ␣ 2
⳵ 2␺ 1
⳵␨ 2 ⫹ ␣ 3 ␺ 1⫺
⳵u1
⳵␨
⫽0 冉冊 (26)
solution form for Y u and Y ␺ yield the general solutions
4 4
at ␨ ⫽0
Y u共 ␨ 兲 ⫽ 兺Ae
j⫽1
j
s j␨
, Y ␺共 ␨ 兲 ⫽ 兺Be
j⫽1
j
s j␨
(38)
⳵␺1
⳵␨
⫽0, ␣3 冉 ⳵u1
⳵␨
⫺ ␺ 1 ⫺k b1冊
* 共 u 1 ⫺ v 1 兲 ⫽0, and
where s j ( j⫽1, . . . ,4) are the roots of the quartic auxiliary equa-
tion 关7兴. Only four of the arbitrary complex constants A j and B j
( j⫽1, . . . ,4) are independent, since they are related by the
v 共 1⫹i ␩ 兲v 1 ⫺k b1 共 u 1 ⫺ v 1 兲 ⫽0
relation
2 D 0 v 1 ⫹k *
2
M* * (27)
at ␨ ⫽1 a 1 s 2j ⫹a 2
B j⫽␪ jA j , ␪ j⫽ 共 j⫽1, . . . ,4兲 (39)
冉冊
a 3s j
⳵u1 M*
d
␺ 1 ⫽0, ␣3 ⫺␺1 ⫹ D 20 u 1 ⫽0 (28) Substituting Eqs. 共33兲 and 共38兲 into the boundary condition
⳵␨ 2 Eqs. 共27兲 and 共28兲 yields the eigenvalue problem represented in
Order ␧ 3 : matrix form as
⳵ ␺ 3 ⳵ 2u 3 关 M共␻*
n 兲兴兵 Z 其 ⫽ 兵 0 其 (40)
D 20 u 3 ⫹ ⫺ ⫹C i苸共 D 0 u 3 ⫺i ␻ n* u 3 兲
⳵␨ ⳵␨ 2 in which the eigenconstants are defined in vector notation as
兵 Z 其 ⫽ 关 A 1 A 2 A 3 A 4 Y v 兴 T , and the 5⫻5 coefficient matrix 关M兴 is
⫽⫺2D 2 D 0 u 1 ⫺C i苸共 D 2 u 1 ⫺i ␴ u 1 兲 (29) defined as

m 1k ⫽ ␣ 3 共 s j ⫺ ␣ j 兲 ⫺k b1
* m 15⫽k b1
*
m 2k ⫽ ␪ j s j m 25⫽0
m 3k ⫽⫺k b1 2 ␻*
m 35⫽⫺M * v 共 1⫹i ␩ 兲 ⫹k b1
n ⫹k *
2
* *
m 4k ⫽e s j 共 ␣ 3 共 s j ⫺ ␪ j 兲 ⫺M d* ␻ n* 2 /2兲 m 45⫽0
m 5k ⫽ ␪ j e sj
m 55⫽0 共 k⫽1, . . . ,4兲 (41)
To obtain the natural frequency ␻ * n , the determinant of the where ( ) ⬘ ⫽ ⳵ / ⳵ T 2 , Ā(T 2 ) is the complex conjugate of the coef-
coefficient matrix 关M兴 must be zero. Considering that the coeffi- ficient A(T 2 ), and W 0 ⫽Y 3u (0)⫺3Y 2u (0)Y v ⫹3Y u (0)Y 2v ⫺Y 3v .
cient function A(T 2 ) in Eq. 共33兲 is arbitrary and assuming A 1 Note that the above terms in Eqs. 共45兲 and 共46兲 are collected only
⫽1, the normal modes Y u ( ␨ ), Y ␺ ( ␨ ) and Y v are uniquely defined for the primary resonance, i.e., exp(i␻n*T0).
by equation 共40兲 for the eigenvalue ␻ * n .
3.2 Third Order Equations. To achieve a third order solu-

tion of the system, assume the complex displacements u 3 , ␺ 3 and 4 Solvability Condition
v 3 to be
In order to obtain a non-trivial solution for the homogeneous
u 3 共 ␨ ,t * 兲 ⫽ ␾ u 共 ␨ ,T 2 兲 e i ␻ *n T 0 , part of Eqs. 共43兲 and 共44兲, the nonhomogeneous Eqs. 共43兲 and
共44兲 must satisfy a solvability or compatibility condition. As dis-
cussed in 关13兴, the solvability condition demands that the right
␺ 3 共 ␨ ,t * 兲 ⫽ ␾ ␺ 共 ␨ ,T 2 兲 e i ␻ *n T 0 , hand side of the nonhomogeneous equations be orthogonal to ev-
ery solution of the corresponding adjoint homogeneous problem.
v 3 共 t * 兲 ⫽ ␾ v 共 T 2 兲 e i ␻ n* T 0 (42) Since the corresponding homogeneous part of Eqs. 共43兲 and
Substituting Eq. 共42兲 into the set of Eqs. 共29兲–共32兲 yields 共44兲 is self-adjoint, the solvability condition can be derived as
⫺
⳵ 2␾ u
⳵␨ n ␾ u⫹
2 ⫺␻*
2
⳵␾␺
⳵␨
冕 1
0
兵 ũ ␣ 3 关 ⫺ 共 2i ␻ n* ⫹C i苸兲 A ⬘ 共 T 2 兲 Y u 共 ␨ 兲 ⫹iC i苸 ␴ A 共 T 2 兲 Y u 共 ␨ 兲兴
⫽⫺ 共 2i ␻ n* ⫹C i苸兲 A ⬘ 共 T 2 兲 Y u 共 ␨ 兲 ⫹iC i苸 ␴ A 共 T 2 兲 Y u 共 ␨ 兲 n 共 2⫺ ␣ 1 兲 A ⬘ 共 T 2 兲 Y ␺ 共 ␨ 兲 ⫺ ␣ 1 ␻ *
⫹ ˜␺ 关 ⫺i ␻ * n ␴ A 共 T 2 兲 Y ␺ 共 ␨ 兲兴其 d ␨
(43) ⫽ 兵 ũ 共 0,t * 兲关 c 苸 i ␻ n* A 共 T 2 兲共 Y u 共 0 兲 ⫺Y v 兲
⫺␣2
⳵ ␾␺
⳵␨ 2
2
⫹ 共 ␣ 1 ⫺1 兲 ␻ *
n ␾ ␺⫹ ␣ 3 ␾ ␺⫺
2
⳵␾u
⳵␨ 冉冊 ⫹ 43 k b3
* A 2 共 T 2 兲 Ā 共 T 2 兲 W 0 兴 ⫺ũ 共 1,t * 兲
⫻ 关 ⫺i ␻ * d A ⬘共 T 2 兲 Y u共 1 兲 ⫹ 2 F 苸␻ *
n exp共 i ␴ T 2 兲兴其 (47)
1 2
n M*
⫽⫺i ␻ n* 共 2⫺ ␣ 1 兲 A ⬘ 共 T 2 兲 Y ␺ 共 ␨ 兲 ⫺ ␣ 1 ␻ n* ␴ A 共 T 2 兲 Y ␺ 共 ␨ 兲
(44) where ũ and ˜␺ are the solutions of the adjoint homogeneous sys-
tem of Eqs. 共43兲 and 共44兲 and the boundary condition Eqs. 共31兲
at ␨ ⫽0 and 共32兲 with a zero right hand side.
It can be seen that the homogeneous adjoint system has an
⳵␾␺ identical form to the first order approximation given by Eqs. 共25兲–
⫽0,
⳵␨ 共28兲; hence, the solutions for ũ, ˜␺ and ṽ should have the same
form as u 1 , ␺ 1 and v 1 .
␣3 冉 ⳵␾u
⳵␨
⫺ ␾ ␺ ⫺k b1 冊
* 共 ␾ u⫺ ␾ v 兲
Substituting the solutions of ũ and ˜␺ into Eq. 共47兲, the solv-
ability condition takes the compact form
3 n ⫹C i苸兲 A ⬘ 共 T 2 兲 ⫹b 1 iC i苸 ␴ A 共 T 2 兲
b 1 共 ⫺2i ␻ *
⫽c 苸 i ␻ *
n A 共 T 2 兲共 Y u 共 0 兲 ⫺Y v 兲 ⫹ k * A 2 共 T 2 兲 Ā 共 T 2 兲 W 0 ,
4 b3 ⫺b 2 i ␻ n* 共 2⫺ ␣ 1 兲 A ⬘ 共 T 2 兲 ⫺b 2 ␣ 1 ␻ *
n ␴A共 T2兲
and
⫽c 苸 i ␻ *
3
n A 共 T 2 兲共 Y u 共 0 兲 ⫺Y v 兲 ⫹ 4 k b3
* A 2 共 T 2 兲 Ā 共 T 2 兲 W 0 Y u 共 0 兲
2 2
⫺ ␻ n* M * 2
* 共 1⫹i ␩ 兲 ␾ v ⫺k b1
2 ␾ v ⫹k v * 共 ␾ u⫺ ␾ v 兲
⫹i ␻ * d A ⬘共 T 2 兲 Y u共 1 兲 ⫺ 2 F 苸␻ *
n exp共 i ␴ T 2 兲 Y u 共 1 兲 (48)
2 1 2
n M*
⫽⫺2i ␻ n* M 2* A ⬘ 共 T 2 兲 Y v ⫹i ␻ n* c 苸 A 共 T 2 兲共 Y u 共 0 兲 ⫺Y v 兲
where b 1 ⫽ ␣ 3 兰 10 Y 2u ( ␨ )d ␨ , b 2 ⫽ ␣ 3 兰 10 Y 2␺ ( ␨ )d ␨ , Ā is the complex
3
⫹ k b3
* A 共 T 2 兲 Ā 共 T 2 兲 W 0
4
2
(45) conjugate of the coefficient A, and A ⬘ is the derivative of A with
respect to the slow time scale T 2 .
at ␨ ⫽1 The coefficient A can be expressed in a polar form such that
␾ ␺ ⫽0, A 共 T 2 兲 ⫽ 21 a 共 T 2 兲 e ⫺i ␪ 共 T 2 兲 (49)
␣3 冉 ⳵␾u
⳵␨
⫺␾␺ ⫺
2 n ␾u
␻* 2
冊 M*
d where a(T 2 ) and ␪ (T 2 ) represent the amplitude and phase angle
of the response, respectively. To eliminate the explicit dependence
on T 2 in the solvability condition and thus, transforming Eq. 共48兲
1 from a nonautonomous system to an autonomous system, the new
⫽⫺i ␻ n* M d* A ⬘ 共 T 2 兲 Y u 共 1 兲 ⫹ F ␻ * 2 exp共 i ␴ T 2 兲 dependent variable ␥ is introduced as
2 苸 n
(46) ␥ ⫽ ␴ T 2⫺ ␪ (50)

Substituting Eqs. 共49兲 and 共50兲 into Eq. 共48兲 and separating the terms and all the damping terms are eliminated. Such frequencies
real and imaginary parts, the modulation is given by are called free frequencies of the nonlinear system, and they are
dependent among others on the initial conditions and strictly ap-
b 3 a ⬘ ⫹b 4 a ␥ ⬘ ⫽b 5 a ␴ ⫹b 6 a 3 ⫹b 7 a⫺F 苸 ␻ n* 2 Y u 共 1 兲 cos ␥ plicable at steady state. However, these frequencies should not be
(51) confused with the linear system natural frequencies.
⫺b 4 a ⬘ ⫹b 3 a ␥ ⬘ ⫽b 8 a ␴ ⫹b 6 a ⫹b 9 a⫺F 苸 ␻ *
3
n Y u 共 1 兲 sin ␥
2
(52) 6 Forced Vibration Response

where the real coefficients b 3 , . . . ,b 9 are defined in Appendix B. The forced vibration of the system can be obtained by including
the damping terms, C i苸 , c 苸 and the forcing term F 苸 in Eqs. 共51兲
and 共52兲. To obtain the steady state response with periodic motion,
5 Free Vibration Response the rate of change of the amplitude and the transformed phase
The free undamped frequencies of the nonlinear system can be shift should be zero, i.e., a ⬘ ⫽ ␥ ⬘ ⫽0.
obtained by letting the quantities C i苸 , c 苸 and F 苸 equal to zero in Eliminating the transformed phase shift ␥ from both of the
Eqs. 共51兲 and 共52兲. Separating the variables and noting that b 7 modulation Eqs. 共51兲 and 共52兲, the detuning parameter ␴ can be
⫽b 9 ⫽0, Eqs. 共51兲 and 共52兲 become found as a function of the amplitude a
c 1 a ⬘ ⫹c 2 a⫹c 3 a 3 ⫽0 (53) ⫺b 14 1
␴⫽ ⫾ 冑b 2 ⫺4b 13b 15
冉冊
(62)
b3 a⬘ b6 2 b5 2b 13 2b 13 14
␪ ⬘⫽ ⫺ a ⫹ 1⫺ ␴ (54) where the coefficients of the detuning parameter b 13 , b 14 and b 15
b4 a b4 b4
are defined in Appendix C.
where c 2 ⫽(b 8 b 4 ⫺b 5 b 3 ) ␴ and c 3 ⫽b 6 b 4 ⫺b 6 b 3 .
c 1 ⫽b 24 ⫹b 23 , Equation 共62兲 defines the frequency-response curve of the non-
Equations 共53兲 and 共54兲 can be solved in conjunction with the linear system, and each point on the curve corresponds to a spe-
initial condition a⫽a 0 and ␪ ⫽ ␪ 0 at T 2 ⫽0 to yield either of the cific singular point on a state plane.
following three cases:
Case 1: c 2 /c 3 ⬍0 7 Numerical Results and Discussions
a⫽ 冑⫺c 2 共 a 21 ⫹e ⫺2c 2 T 2 /c 1 兲 / 共 c 3 a 21 ⫹c 3 e ⫺2c 2 T 2 /c 1 ⫺c 3 兲 (55) Numerical simulations are performed for a typical rotor shaft
再冋册冎冋册
system with nonlinear bearings. The physical parameters of this
b3 1 a 21 ⫹e ⫺2c 2 T 2 /c 1 1 b3 a 21 ⫹1 system are given as
␪共 T 2 兲⫽ ␪ 0⫹ ln ⫺ ln
b4 2 a 1 ⫹e ⫺2c 2 T 2 /c 1 ⫺1
2
2 b4 a 21 E⫽2.04⫻1011 Pa v ⫽0.3 ␳ ⫽7750 kg/m3 l⫽0.15 m
⫹
b6 c1
b 4 2c 3 共 a 21 ⫺1 兲
冋
ln共 a 21 ⫹e ⫺2c 2 T 2 /c 1 ⫺1 兲
C i ⫽200 kg/m.s
k b1 ⫽1⫻10 N/m 9
d s ⫽0.0286 m
c b ⫽400 kg/s
␬ ⫽0.68 ␩ ⫽0.01
k b3 ⫽1⫻107 N/m3
⫹
2a 21 c 2
c1
T 2 ⫺ln a 21 ⫹ 1⫺册冉冊 b5
b4
␴T2 (56)
M 2 ⫽0.5 kg
k v ⫽1⫻109 N/m
d d ⫽0.13 m M d ⫽55 kg e⫽0.01 m
where a 21 ⫽a 20 /(a 20 ⫹c 2 /c 3 ).
Case 2: c 2 /c 3 ⬍0 Assume that the shaft has a circular cross section, hence ␣ 1
⫽2 and J s ⫽2 ␳ I s . In addition, the shaft rotates at a typical speed
a⫽ 冑⫺c 2 / 共 c 3 a 22 ⫹c 3 e ⫺2c 2 T 2 /c 1 ⫺c 3 兲 (57) ⍀⫽300 rad/s. Free and forced nonlinear vibration analysis are
冋
carried out to this system to investigate the influence of nonlin-
c1 b3 earity in the bearings on the system characteristic frequencies and
␪共 T 2 兲⫽ ␪ 0⫺ ln共 a 22 ⫹e ⫺2c 2 T 2 /c 1 ⫺1 兲
2c 3 b 4 共 a 22 ⫺1 兲 frequency response.
册
Figure 2 shows the free nonlinear oscillations of this system for
2c 2 b6 different values of the elastic coefficient k b3 of the nonlinear bear-
⫹ T ⫺ln a 22 ⫹ 关 ln共 a 22 ⫹e ⫺2c 2 T 2 /c 1 ⫺1 兲 ⫺ln a 22 兴
c1 2 2b 4 ing. The initial transverse displacement u(0) is assumed as 10
mm and the initial phase angle ␪ 0 is zero. The fundamental natu-
⫹ 1⫺冉冊 b5
b4
␴T2 (58)
ral frequency of the above system is 458.71 rad/s. It can be seen
from the waveforms presented that increasing the nonlinearity of
the system, i.e., increasing k b3 , shifts the frequency of the non-
where a 22 ⫽(a 20 ⫺c 2 /c 3 )/a 20 . linear system to a higher value and increases the steady state
Case 3: c 2 ⫽0 amplitude of the oscillation.
It is noted that the detuning parameter ␴ in the nonlinear free
a⫽ 冑a 20 / 共 1⫹2c 3 a 20 T 2 /c 1 兲 (59)
vibration analysis is related to the nondimensional spinning fre-
␪共 T 2 兲⫽ ␪ 0⫺
b3
2b 4
ln 1⫹ 冉
2c 3 2
a T
c1 0 2 冊 quency ⍀ * through the relation ␴ ⫽(⍀ * ⫺ ␻ n* )/␧ 2 . Also, the non-
dimensional amplitude a can be related to the transverse deflec-
tion of the shaft u at the first order approximation as
⫺
c1 b6
2c 3 b 4
ln 1⫹ 冉
2c 3 2
c1
a 0 T 2 ⫹ 1⫺ 冊冉冊
b5
b4
␴T2 (60)
u(T 0 ,T 2 , ␨ )⫽␧l/2a(T 2 )e i ( ␻ *n T 0 ⫺ ␪ ( T 2 )) Y u ( ␨ ). This relation can be
simplified for a zero initial phase angle ␪ 0 to become a
⫽2u/(l␧).
The frequency of the nonlinear system ␻ * nn is related to the Assuming that mass eccentricity of the disk has no contribution
frequency of the linear system ␻ *
n through the relation to the nonlinear forcing terms, the forcing term F * defined in Eq.
* ⫽ ␻ n* ⫹␧ 2 ␪ ⬘
␻ nn (61) 共24兲 will be equal to zero, and subsequently F 苸 will be equal to
zero. The effect of the bearing nonlinear elastic coefficient k b3 on
where ␪ ⬘ is the phase angle derivative with respect to the slow the free frequency of the nonlinear system versus the amplitude a
time scale T 2 . of the transverse deflection is illustrated in Fig. 3. It is shown that
In a quasi-harmonic nonlinear system, the frequencies of the as the amplitude increases, the corresponding shift in the nonlin-
nonlinear system can be assumed periodic if the external forcing ear natural frequency remains minimal for low values of the elas-

Fig. 2 Waveform of the transverse deflection for various values of bearing
nonlinear elastic coefficient k b 3
tic coefficient k b3 of the nonlinear bearing; i.e., for weak nonlin- ing coefficient, C i ⬇270 kg/m.s, indicates the increased efficiency
ear system. As k b3 increases, the change in the natural frequency of the system to dissipate energy. This may be attributed to the
of the nonlinear system increases significantly. The increase of the optimum interaction between the nonlinear restoring forces, exter-
frequencies of the nonlinear system and the bend of the curve nal damping provided by the bearing supports and the viscoelastic
towards the higher frequencies are typical trends of hardening elements, and the internal damping of the shaft.
nonlinearities. It should be pointed out that since a zero forcing term corre-
Figures 4 and 5 show the effect of the internal viscous damping sponding to mass eccentricity of the disk is assumed for the re-
coefficient C i of the shaft on the free frequency of the nonlinear sults of both Figs. 4 and 5, the free frequencies of the nonlinear
system versus amplitudes. Figure 4 depicts the trend for low C i system can be obtained. Also, the internal damping force is as-
values of 100 kg/m.s to 200 kg/m.s, whereas Fig. 5 shows the sumed to be linear; hence, proportional to the strain velocity,
trend for higher C i values of 270 kg/m.s to 300 kg/m.s. Figure 4 which is determined by the difference in whirling angular velocity
shows that increasing the internal viscous damping of the shaft and the rotational speed of the shaft. In general, the internal damp-
shifts the free frequency of the nonlinear system to higher values ing force 共hysteretic damping兲 is nonlinear and roughly indepen-
while almost maintaining the curve characteristic. However, this dent of the frequency. In addition, it is proportional to the square
trend has a limiting value of C i ⬇270 kg/m.s beyond which any of the vibration amplitude, to the shaft stiffness, and to the number
increase in C i will result in a shift of the curve to lower frequen- of reversals of the sign of the strain velocity 关14兴.
cies, as shown in Fig. 5. The limiting value of the internal damp- Including the effect of the mass eccentricity and hence the ex-
Fig. 3 Free frequency of the nonlinear system versus amplitudes for vari-
ous values of bearing nonlinear elastic coefficient k b 3

Fig. 4 Free frequency of the nonlinear system versus amplitudes for vari-
ous values of internal viscous damping coefficient C i of the shaft
citation frequency ⍀, the steady-state frequency-response curve is the multiple scales method starts to deviate from the basic as-
shown near the primary resonance in Fig. 6. The effect of the sumption of weak nonlinearity. Thus, the error in predicting the
elastic coefficient k b3 of the nonlinear bearing on the response frequency response curves increases.
curve is illustrated. At k b3 equals zero, the response curve is iden-
tical to the linear response curve. As k b3 increases, the response
curve bends to the right side with lower and flattened peak. This 8 Summary and Conclusions
trend continues until a jump phenomenon takes place due to the The multiple scales method is adopted to analyze the free and
multi-valued phase points, indicating the bifurcation in the sys- forced vibration of rotor shaft system with viscoelastically sup-
tem. Also, for low values of the elastic coefficient k b3 (⬍1 ported nonlinear bearings. Timoshenko shaft model is assumed for
⫻107 N/m3 ) of the nonlinear bearing, and thus weak nonlinearity, the shaft, two-element viscoelastic model is utilized for the sup-
the frequency response curve deviates slightly from the frequency ports, and typical roller bearings with cubic nonlinearity are em-
response curve of the corresponding linear system. ployed. The free and forced vibration is developed based on the
It is worth noting that for large values of the nonlinear elastic direct multiple scales method. Third order perturbation expansion
coefficient k b3 (⬎5⫻108 N/m3 ), the perturbation analysis using with a one-to-one frequency-to-amplitude relationship of the non-
Fig. 5 Free frequency of the nonlinear system versus amplitudes for higher
values of internal viscous damping coefficient C i of the shaft

Fig. 6 Frequency response curves for various values of the bearing non-
linear elastic coefficient k b 3
linear system is used. The natural frequencies of the nonlinear Appendix B

system and the steady state response are obtained.
The coefficients of the modulation equations
Numerical simulations of nonlinear free and forced vibration on
b 3 ⫽2b 1m ␻ * n 共 ␣ 1 ⫺2 兲 ⫹b 12m M *
n ⫹b 1r C i苸 ⫺b 2m ␻ * d ␻*
a typical system are performed. The results show that a limiting n ,
value of the internal viscous damping coefficient of the shaft ex-
ists where the trend of the frequency-amplitude curve switches. b 4 ⫽2b 1r ␻ n* ⫺b 1m C i苸 ⫺b 2r ␻ n* 共 ␣ 1 ⫺2 兲 ⫹b 12r M d* ␻ n*
Also, the primary resonance peak shifts to higher frequencies with
the increase of the bearing nonlinear elastic characteristics, but b 5 ⫽2b 1r ␻ n* ⫹2b 2r ␻ n* ⫹b 12r M d* ␻ n* , b 6 ⫽ 共 3/16兲 k *
b3 b 10r ,
with a flattened curve and hence lower peak values. A Jump phe-
b 7 ⫽⫺b 11m c 苸 ␻ n* , b 8 ⫽2b 1m ␻ n* ⫹2b 2m ␻ n* ⫹b 12m M d* ␻ n* ,
nomenon takes place for high values of the bearing nonlinear
elastic characteristics. b 9 ⫽b 11r c 苸 ␻ n* ,
Acknowledgment b 10⫽W 0 Y u 共 0 兲 ⫽Y 4u 共 0 兲 ⫺3Y 3u 共 0 兲 Y v ⫹3Y 2u 共 0 兲 Y 2v ⫺Y u 共 0 兲 Y 3v ,

This research is financially supported by a research grant from b 11⫽Y 2u 共 0 兲 ⫺Y u 共 0 兲 Y v
the Natural Sciences and Engineering Research Council of
Canada. b 12⫽Y 2u 共 1 兲 , and b j ⫽b jr ⫹ib jm 共 j⫽1,2,10,11,12兲 .
Appendix A Appendix C
First set of nondimensional quantities The coefficients of the detuning parameter ␴ are defined as
t *⫽ 冑 ␬G
␳l2
t,
u
u *⫽ ,
l
␺ *⫽
␺
l
, ⍀ *⫽ 冑 ␳l2
␬G
⍀,
b 13⫽b 25 ⫹b 28 , b 14⫽2b 6 共 b 5 ⫹b 8 兲 a 2 ⫹2 共 b 5 b 7 ⫹b 8 b 9 兲 ,
1 2
b 15⫽ 共 b 6 a 2 ⫹b 7 兲 2 ⫹ 共 b 6 a 2 ⫹b 9 兲 2 ⫺ F ␻ *4Y 2共 1 兲
l v a2 苸 n u
C i* ⫽ Ci , v *⫽ ,
A 冑␬ G ␳ l
References
Js E Al 2 l ⳵␺* 关1兴 Darlow, M., and Zorzi, E., 1981, Mechanical Design Handbook of Elastomers,
␣ 1⫽ , ␣ 2⫽ , ␣ 3⫽ , M *⫽ M ⫽␣2 ,
␳Is ␬G Is ␬ GI s ⳵␨ NASA CR3423.
关2兴 Dutt, J. K., and Nakra, B. C., 1992, ‘‘Stability of Rotor Systems with Vis-
Q *⫽
l2
␬ GI s
Q⫽ ␣ 3
⳵u*
⳵␨ 冉
⫺␺* , 冊 *⫽
k b1
l3
k ,
␬ GI s b1
coelastic Supports,’’ J. Sound Vib., 153共1兲, pp. 89–96.
关3兴 Dutt, J. K., and Nakra, B. C., 1993, ‘‘Vibration Response Reduction of a Rotor
Shaft System Using Viscoelastic Polymeric Supports,’’ ASME J. Vibr. Acoust.,
115, pp. 221–223.
l3 l5 l2 关4兴 Dutt, J. K., and Nakra, B. C., 1995, ‘‘Dynamics of Rotor Shaft System on
k v* ⫽ k , *⫽
k b3 k , c b* ⫽ cb , Flexible Supports with Gyroscopic Effects,’’ Mech. Res. Commun., 22共6兲, pp.
␬ GI s v ␬ GI s b3 I s 冑␬ G ␳ 541–545.
关5兴 Kulkarni, P., Pannu, S., and Nakra, B. C., 1993, ‘‘Unbalance Response and
l l eM d Stability of a Rotating System with Viscoelastically Supported Bearings,’’
d⫽
M* 2⫽
M* F *⫽
M , M , , Mech. Mach. Theory, 28共3兲, pp. 427– 436.
␳Is d ␳Is 2 ␳Is 关6兴 Shabaneh, N. H., and Zu, Jean W., 1999, ‘‘Vibration Analysis of Viscoelasti-
cally Supported Rotor-Bearing Systems,’’ Asia-Pacific Vibration Conference
W * ⫽ 43 共 u * 2 ū * ⫺2u * ū * v * ⫹ v * 2 ū * ⫺u * 2 v̄ * ⫹2u * v * v̄ * 1999 共A-PVC’99兲, Singapore, December.
关7兴 Shabaneh, N. H., and Zu, Jean W., 2000, ‘‘Dynamic Analysis of Rotor-Shaft
Systems with Viscoelastically Supported Bearings,’’ Mech. Mach. Theory,
⫺ v * 2 v̄ * 兲 ⫹ 41 共 ū * 3 ⫺3ū * 2 v̄ * ⫹3ū * v̄ * 2 ⫺ v̄ * 3 兲 35共9兲, pp. 1313–1330.

关8兴 Shabaneh, N. H., and Zu, Jean W., 2000, ‘‘Dynamic and Stability Analysis of Bearings with Viscoelastic Supports,’’ ASME J. Vibr. Acoust., 119, pp. 539–
Rotor-Shaft Systems with Viscoelastically Supported Bearings,’’ Trans. Can. 544.
Soc. Mech. Eng., 24共1B兲, pp. 179–189. 关12兴 Nayfeh, A. H., Nayfeh, J. F., and Mook, D. T., 1992, ‘‘On Methods for Con-
关9兴 Yamamoto, T., Yasuda, K., and Nagasaka, I., 1976, ‘‘Ultra-Subharmonic Os- tinuous Systems with Quadratic and Cubic Nonlinearities,’’ Nonlinear Dyn. 3,
cillations in a Nonlinear Vibratory System,’’ Bull. JSME, 19共138兲, pp. 1442–
pp. 145–162.
1447.
关13兴 Nayfeh, A. H., 1981, Introduction to Perturbation Techniques, Wiley, New
关10兴 Ji, Z., and Zu, J. W., 1998, ‘‘Method of Multiple Scales for Vibration Analysis
of Rotor-Shaft Systems with Non-Linear Bearing Pedestal Model,’’ J. Sound York.
Vib., 218共2兲, pp. 293–305. 关14兴 Yamamoto, Toshio, and Ishida, Yukio, 2001, Linear and Nonlinear Rotordy-
关11兴 Bhattacharyya, K., and Dutt, J. K., 1997, ‘‘Unbalance Response and Stability namics: A Modern Treatment with Applications, Wiley Series in Nonlinear
Analysis of Horizontal Rotor Systems Mounted on Nonlinear Rolling Element Science.

Nonlinear Parameter Estimation
Animesh Chatterjee
Assistant Professor,
in Rotor-Bearing System Using
Department of Mechanical Engineering,
Visvesvaraya National Institute of Technology,
Nagpur, India-440011
Volterra Series and Method of
Harmonic Probing
Nalinaksh S. Vyas
Professor, Volterra series provides a structured analytical platform for modeling and identification of
Department of Mechanical Engineering, nonlinear systems. The series has been widely used in nonparametric identification
Indian Institute of Technology, through higher order frequency response functions or FRFs. A parametric identification
Kanpur, India-208016 procedure based on recursive evaluation of response harmonic amplitude series is pre-
e-mail: [email protected] sented here. The procedure is experimentally investigated for a rotor-bearing system
supported in rolling element bearings. The estimates of nonlinear bearing stiffness ob-
tained from experimentation have been compared with analytical values and experimental
results of previous works. 关DOI: 10.1115/1.1547486兴
1 Introduction transforms. Chatterjee and Vyas 关7兴 have suggested a recursive

iteration technique, which computes the first and higher order ker-
System identification is an inverse problem of determination of
nel transforms recursively from the measured response harmonic
the mathematical structure of physical system from input-output
amplitudes. The response series is considered with an optimum
measurement. Volterra series provides a structured analytical plat- number of terms governed by the convergence criterion.
form for modeling and identification of nonlinear systems. The In rotor-bearing systems, stiffness and damping nonlinearity
series represents nonlinear response through a set of multi- play a significant role in determining system behavior. Rolling
dimensional kernels, known as Volterra kernels. Fourier trans- element bearing analyses presented by Harris 关8兴 and Ragulski
forms of these kernels provide the definition of higher order ker- et al. 关9兴 point out that the stiffness force follows a symmetric
nel transforms, or higher order frequency response functions nonlinearity relationship with deflection. Hydrodynamic bearings
共FRFs兲关1兴. For harmonic excitations, the response can be conve- operating on fluid film lubrication also exhibit nonlinear relation-
niently expressed in terms of the first and higher order FRFs. ship in the stiffness and damping forces. Investigations on rotor
Extensive research work has been done in the area of non- vibration, using nonlinear modeling of bearing coefficients, have
parametric system identification through measurement of first and been carried out by Bannister 关10兴 and Choy et al. 关11兴. Garibaldi
higher order kernel transforms or FRFs. Boyd, Tang and Chua 关2兴 and Tomlinson 关12兴 have presented a method for identifying the
developed a second order kernel transform measurement proce- nonlinear bearing coefficients in a rigid rotor using Hilbert trans-
dure using multi-tone harmonic probing. Chua and Liao 关3兴 ex- form. Khan and Vyas 关13兴 have developed parameter estimation
tended the procedure for third and higher order kernel transforms. procedures for both rigid rotor and flexible rotor model. The pro-
Gifford and Tomlinson 关4兴 developed a multi-degree-of-freedom cedure employs Gaussian random excitation and Wiener kernel
curve fitting procedure for estimating the higher order FRFs. transforms are extracted from the input-output data. Volterra ker-
However, non-parametric system representation, in terms of nel transforms are then obtained from the measured Wiener kernel
higher order kernels, often requires extensive computational work transforms and used for nonlinear parameter estimation.
and needs large number of experiments either with harmonic or The parameter estimation procedure described here employs
stochastic excitation. harmonic probing technique in conjunction with Volterra series
Parametric system identification requires a-priori information response representation. Investigation is carried out on a rotor
about the topology of the mathematical structure of the system. supported in rolling element bearings. The system is analyzed as a
Chatterjee and Vyas 关5兴 have suggested an identification proce- single-degree-of-freedom system and linear and nonlinear stiff-
dure for classification of systems between polynomial and non- ness parameters are estimated from the response measured at the
polynomial form of nonlinearity. Further classification is made bearing housing. The estimates are compared with theoretical val-
between symmetric and asymmetric forms of nonlinearity. For a ues obtained from the analytical work of Ragulski and also with
polynomial form nonlinearity, the series structure is further some earlier experimental results.
identified through a peak ratio comparison test under multi-tone
excitation.
Higher order FRFs, in case of polynomial form nonlinearity,
can be synthesized with first order FRF and nonlinear parameters. 2 Volterra Series Response Representation
This provides a generalized basis for nonlinear parameter estima- Volterra series response for a general physical system with f (t)
tion through measurement of first and higher order kernel trans- as input excitation and x(t) as output response is represented by
forms. Lee 关6兴 extracted the response components of first har-
monic through component separation technique. First and higher x 共 t 兲 ⫽x 1 共 t 兲 ⫹x 2 共 t 兲 ⫹x 3 共 t 兲 ⫹ . . . ⫹x n 共 t 兲 ⫹ . . . (1)
order kernel transforms were then computed from the separated with
response components and nonlinear parameters were estimated
using the relationship between higher order and first order kernel

x n共 t 兲 ⫽ 冕⫺⬁
⬁
... 冕
⫺⬁
⬁
h n 共 ␶ 1 , . . . , ␶ n 兲 f 共 t⫺ ␶ 1 兲 . . . f 共 t
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002;

revised October 2002. Associate Editor: M. I. Friswell. ⫺␶n兲d␶1 . . . d␶n (2)
h n ( ␶ 1 , . . . , ␶ n ) is the nth order Volterra kernel and its Fourier where
兺冉 2冊
transform provides the nth order frequency response functions ⬁ 2n
共FRFs兲 or Volterra kernel transforms as A
X 0⫽ 2n 共 ␻ 兲
C n H n,n
2n
H n共 ␻ 1 , . . . , ␻ n 兲 ⫽ 冕⫺⬁
⬁
... 冕 ⬁
⫺⬁
h n共 ␶ 1 , . . . , ␶ n 兲
n⫽1
n
X共 n␻ 兲⫽ 兺 ␴ 共n␻兲
i⫽1
i and ␾ n ⫽⬔X 共 n ␻ 兲 (8)
⫻ 兿e
i⫽1
⫺ j ␻i␶i
d␶1 . . . d␶n (3)
with
For a single-tone harmonic excitation
A A
␴ i 共 n ␻ 兲 ⫽2 冉冊 A
2
n⫹2i⫺2
n⫹2i⫺2 n⫹i⫺1,i⫺1
C i⫺1 H n⫹2i⫺2 共␻兲 (9)
f 共 t 兲 ⫽A cos ␻ t⫽ e j ␻ t ⫹ e ⫺ j ␻ t (4)
2 2
3 Parameter Estimation
the expression for the nth order response component, following For a system with polynomial form of nonlinearity under har-
Eq. 共2兲, can be obtained as monic excitation given by
x n共 t 兲 ⫽ 冉冊
A
2
n
兺
p⫹q⫽n
n
C q H np,q 共 ␻ 兲 e j ␻ p,q t (5) mẍ 共 t 兲 ⫹cẋ 共 t 兲 ⫹k 1 x 共 t 兲 ⫹k 2 x 2 共 t 兲 ⫹k 3 x 3 共 t 兲 ⫽A cos ␻ t (10)
first three response harmonic series, after re-arranging and trun-
where the following brief notations have been used cating Eq. 共8兲 to a finite number of terms, k, can be expressed as
(5)
H 1共 ␻ 兲 ⬇
1
A
X共 ␻ 兲⫺
i⫽2
冋␴ i共 ␻ 兲兺
k
册 (11a)
The total response of the system, then becomes
x共 t 兲⫽ 兺冉
⬁
A
冊 n
兺 n
C q H np,q 共 ␻ 兲 e j ␻ p,q t (6)
H 2共 ␻ , ␻ 兲 ⬇
2
A2 冋
X共 2␻ 兲⫺
i⫽2
␴ i共 2 ␻ 兲兺
k
册 (11b)
冋册
n⫽1 2 p⫹q⫽n k
4
The response series given in Eq. 共6兲 can be written in terms of its H 3共 ␻ , ␻ , ␻ , 兲 ⬇
A3
X共 3␻ 兲⫺
i⫽2
␴ i共 3 ␻ 兲兺 (11c)
harmonics as
x 共 t 兲 ⫽X 0 ⫹ 兩 X 共 ␻ 兲兩 cos共 ␻ t⫹ ␾ 1 兲 ⫹ 兩 X 共 2 ␻ 兲兩 cos共 2 ␻ t⫹ ␾ 2 兲 where the higher order kernel transforms are related to the lower
order kernel transforms through nonlinear parameters 共Chatterjee
⫹ 兩 X 共 3 ␻ 兲兩 cos共 3 ␻ t⫹ ␾ 3 兲 ⫹.. (7) and Vyas, 关14兴兲 as
冋兺
册
p ,q 1 q ,q 2
k2 兵 n1C q1H n 1 共 ␻ 兲其 *兵 n2C q2H n2 共 ␻ 兲其⫹
p i ⫹q i ⫽n i 1 2
H 1 共 ␻ p,q 兲 n 1 ⫹n 2 ⫽n
H np,q 共 ␻ 兲 ⫽⫺ n for n⬎1 (12)
Cq
k3 兺
p i ⫹q i ⫽n i
兵 n1 p ,q
1
p ,q
2
p ,q
C q 1 H n 1 1共 ␻ 兲其 * 兵 n 2 C q 2 H n 2 2共 ␻ 兲其 * 兵 n 3 C q 3 H n 3 3共 ␻ 兲其
3
n 1 ⫹n 2 ⫹n 3 ⫽n
Higher order series terms are initially neglected in Eqs. 共11a–c兲 Step-II: System is excited at frequencies close to one-third of
and kernel transforms H 1 ( ␻ ), H 2 ( ␻ , ␻ ) and H 3 ( ␻ , ␻ , ␻ ) are ex- natural frequency, for distinct measurability of third harmonic
tracted from measured harmonic amplitudes. Equation 共12兲 is em- X(3 ␻ ) 共Ref. 关14兴兲. Employing Eq. 共12兲, preliminary estimate of
ployed to estimate the second and third order nonlinear param- nonlinear parameter, k 3 , is obtained through regression between
eters k 2 and k 3 . Higher order series term contribution is then the estimated third order kernel transform and its synthesized ker-
considered and the procedure is iterated till the estimates converge nel factor, ⌫ 3 ( ␻ ), using the following relationship
within a specified limit. For a typical Duffing oscillator, the esti-
mation algorithm can be structured through following steps. H 3 共 ␻ , ␻ , ␻ 兲 ⫽k 3 ⌫ 3 共 ␻ 兲 (14)
Step-I: System is excited at frequencies, ␻ i , with ␻ i varying
over a frequency range including the natural frequency, ␻ n , of the where
system. Response x(t) is measured and harmonic amplitude,
X( ␻ i ), is filtered to give preliminary estimate of first order kernel ⌫ 3 共 ␻ 兲 ⫽⫺H 31 共 ␻ 兲 H 1 共 3 ␻ 兲 (15)
transform as
Step-III: The series 兺 i⫽2
k
␴ i ( ␻ ) is computed with the H 1 ( ␻ )
H 1 共 ␻ i 兲 ⫽X 共 ␻ i 兲 /A i , i⫽1, . . . .N (13) values taken from the best fit curve estimated in Step-I and the
nonlinear parameter, k 3 , estimated in Step-II and substituted in
The excitation level A i is varied for constant response amplitude equation 共11a兲 to obtain new estimates of linear parameters.
X( ␻ i ) for keeping the series approximation error low and nearly Step-IV: The series 兺 i⫽2
k
␴ i (3 ␻ ) is computed and substituted in
uniform over the frequency range 关14兴. Standard curve fitting pro- Eq. 共11c兲 to refine the estimate of the nonlinear parameter k 3 .
cedure 共Ewins 关15兴兲 is employed to obtain the best fit FRF curve Iteration is continued till the estimate of nonlinear parameter, k 3 ,
and preliminary estimation of linear parameters is made. converges within a specified limit.

Fig. 3 Response acceleration spectrum from rap test
The sign of the nonlinear parameter, k 3 , is identified through

observations of change of sign of real part of measured third har-
monic amplitudes during transition of excitation frequency across
one-third of the natural frequency. From Eqs. 共14兲, 共15兲, one can
note that for ␻ ⬍ ␻ n /3, both H 1 ( ␻ ) and H 1 (3 ␻ ) have positive real
parts, which means that the kernel factor ⌫ 3 ( ␻ ) will have a nega-
tive real part. Consequently, the third order kernel transform
H 3 ( ␻ , ␻ , ␻ ) will also have a negative real part for positive k 3 .
For ␻ n ⬎ ␻ ⬎ ␻ n /3, H 1 ( ␻ ) will have a positive real part while the
real part of H 1 (3 ␻ ) will bear negative sign. The kernel factor,
⌫ 3 ( ␻ ), and H 3 ( ␻ , ␻ , ␻ ) will have a positive real part for positive
value of the nonlinear parameter k 3 .
4 Experimental Investigation
Experimental studies have been carried out on a test rig 共Fig. 1兲
consisting of a 10 mm diameter shaft supported in ball bearings
共SKF 6200兲. A single-degree-of-freedom model is considered and
cross-coupling stiffness parameters are neglected. Harmonic exci-
tation force is applied on the bearing housing 共in vertical direc-
Fig. 1 „a… Experimental set up along with instrumentation „b… tion兲 through an electro-dynamic shaker. Measurement for excita-
Close up view of exciter mounting arrangement and impedance tion force and resultant vibration are made at the bearing housing
head through an impedance head attached between the shaker and bear
Fig. 2 Schematic diagram of rotor bearing test rig and instrumentation

Fig. 4 Response component spectra for ␻ Ä330 Hz
Fig. 5 Excitation level variation, response amplitude, X „ ␻ …,
and preliminary estimate of H 1 „ ␻ …
ing cap. A schematic diagram of the instrumentation is shown in

Fig. 2. Ragulski 关9兴, however, observed that for a small deformation 兩 x 兩
⭐g, where g is the pre-load in the bearing, the stiffness function
4.1 Nonlinearity Structure Identification. An initial mea-
surement of the natural frequency of the system is obtained is given by the form k(x)⫽a⫺bx 2 . This enables one to model the
through a rap test, using an impact hammer. The rap test response bearing forces through cubic terms and represent the rotor-bearing
spectrum 共Fig. 3兲 shows that the natural frequency is in the vicin- system as
ity of 1010 Hz. An initial test is done to find the form of stiffness mẍ 共 t 兲 ⫹cẋ 共 t 兲 ⫹k 1 x 共 t 兲 ⫹k 3 x 3 共 t 兲 ⫽A cos ␻ t (16)
nonlinearity through ordered component separation method pre-
sented in 关5兴. The system is harmonically excited at 330 Hz 共near 4.2 Estimation of Stiffness Parameters. For estimation of
one-third of natural frequency兲 at excitation levels 4N, 3N and first order kernel transforms, H 1 ( ␻ ), an excitation frequency
2N and the first three response components x 1 (t), x 2 (t) and x 3 (t) range 925–1035 Hz is selected. Keeping in view, the single-
are separated. The response components are filtered to investigate degree-of-freedom treatment of the rotor-bearing set-up, a wider
presence of various ordered harmonics. For an asymmetric non- excitation frequency range is avoided in order to stay clear of
linearity structure both even and odd harmonics will be present in some other structural modes, which may influence measured vi-
the response spectrum, whereas for symmetric form, only odd bration data. In order to keep the series approximation error low,
harmonics will be present. Also for polynomial form nonlinearity, excitation level is varied over the frequency set to obtain response
spectra of response components will have ordered presence of amplitude at a constant level, which in this case was selected as
harmonic, i.e., odd response components will have odd harmonics 1.0⫻10⫺8 m. The variation of excitation level is plotted in Fig.
and even response components will have even harmonics 共refer 5共a兲 and corresponding response harmonic amplitude, X( ␻ ), is
关5兴兲. Fig. 4共a–c兲 show the spectra of first three response compo- shown in Fig. 5共b兲. Preliminary estimate of the first order kernel
nents separated from the measured response. The figures indicate transform, H 1 ( ␻ ), is shown in Fig. 5共c兲. Curve fitting of the
that while the stiffness nonlinearity is symmetric, its form is non- H 1 ( ␻ ) graph gives following estimates of natural frequency and
polynomial as the response component spectra do not exhibit or- damping
dered harmonic characteristics. Theoretical formulations based on
␻ n ⫽1012.2 Hz, ␵⫽0.01156.
Hertzian contact theory, 共Harris 关8兴, Ragulski 关9兴兲 also indicate
that the nonlinear stiffness associated with rolling elements fol- Estimation of nonlinear stiffness parameter k 3 is done through
lows a fractional power relationship and not a polynomial form. measurement of third response harmonic amplitude X(3 ␻ ). The

⫻1019 N/m3 . Assuming that both the bearings are identical and
act in parallel, the nonlinear stiffness parameter of each bearing
then becomes 0.875⫻1019 N/m3 . Figure 8共b兲 shows the final es-
timate of the first order kernel transform along with its prelimi-
nary estimate. It is seen that the successive iterations have im-
proved the FRF values significantly near and around the natural
frequency. Final estimates of natural frequency and damping be-
come
␻ n ⫽1011.47 Hz, ␵⫽0.01078.
The linear stiffness parameter, k 1 , of the bearing system is com-
puted from the natural frequency and the equivalent rotor mass
per bearing and is found to be 1.585⫻107 N/m.
Similar exercises are carried out with excitation level of 3 N
共Case II兲 and 2 N 共Case III兲 corresponding to 5% and 3% mea-
surability respectively. The convergence trend in the estimate k 3 is
shown respectively in Fig. 9共a, b兲. The estimates of nonlinear
stiffness parameter of each bearing is found to be 1.4285
⫻1019 N/m3 共Case II兲 and 1.740⫻1019 N/m3 共Case III兲 respec-
tively. Damping and linear stiffness parameter are found as
Case II: ␵⫽0.01015 k 1 ⫽1.535⫻107 N/m.
Case III ␵⫽0.01002 k 1 ⫽1.586⫻107 N/m
The estimates for different cases are summarized in Table 1 given
below.
It can be seen from the estimates that, while the values of the
estimated linear parameters are very accurate, the estimate of the
nonlinear parameter k 3 , varies with the excitation amplitude. This
is due to the fact that the system nonlinearity does not adhere to
polynomial form and therefore the cubic coefficient of equivalent
polynomial form would be amplitude dependent.
Figure 10 shows the sign variation of real part of X(3 ␻ ), over
the excitation frequency range 330–345 Hz for excitation levels
4N, 3N and 2N. The sign of real part of X(3 ␻ ) is seen to change
Fig. 6 „a… Typical response spectrum with excitation at ␻ from positive values to negative values during frequency transi-
Ä330 Hz „b… Measurability of third response harmonic at differ- tion through one-third natural frequency. This indicates that the
ent excitation levels nonlinearity is negative in sign.
displacement amplitude of the third response harmonic is low and 5 Validation of Estimates
therefore measurements were made for acceleration amplitudes.
Figure 6共a兲 shows the acceleration amplitudes of the first and third For validation of the experimentally estimated nonlinear stiff-
order harmonics, for an excitation frequency of 330 Hz 共near one- ness parameter, analytical formulations of Harris 关8兴 and Ragulski
third natural frequency兲. The ratio between the amplitudes at third et al. 关9兴 are employed. These formulations are based on Hertz’s
harmonic and first harmonic is defined as measurability index, theory of elastic contacts and treat the bearings in isolation of the
which provides the basis of selecting the excitation level and fre- shaft. Figure 11 shows a typical isolated ball bearing configura-
quency range for measurement of third response harmonic ampli- tion, in which external forces act along x-axis. ␩ i is the angle
tude X(3 ␻ ). For measurement of third harmonic one can select between load axis 共i.e., x-axis兲 and the radial direction of a typical
excitation level corresponding to a certain measurability index. ith ball element. For a displacement 共x, y兲 of the moving ring
Figure 6共b兲 shows measurability indices for three different excita- along the respective axes, total elastic force acting in radial direc-
tion amplitudes of 2N, 3N and 4N over a frequency range of 305 tion at the point of contact of ith ball is given by
Hz–355 Hz. Corresponding peak measurability indices can be F i ⫽k n 共 g⫹x cos ␩ i ⫹y sin ␩ i 兲 3/2 (17)
seen to be approximately of the order of 3%, 5% and 10%. Exci-
tation frequencies are selected at 330 Hz, 335 Hz, 340 Hz and 345 and its projections along x and y axes respectively are
Hz, as measurability is relatively higher in this range. Third re-
sponse harmonic amplitude X(3 ␻ ) is then measured at these se- F xi ⫽k n 共 g⫹x cos ␩ i ⫹y sin ␩ i 兲 3/2 cos ␩ i (18)
lected frequencies for three different excitation levels as
F yi ⫽k n 共 g⫹x cos ␩ i ⫹y sin ␩ i 兲 3/2 sin ␩ i (19)
Case I: Excitation level⫽4 N
Case II: Excitation level⫽3 N where g is the radial pre-load between the ball and the races, k n is
Case III: Excitation level⫽2 N a coefficient of proportionality depending on the geometric and
For case I corresponding to 10% measurability, the measured re- material properties of the bearing. Total restoring force of the
sponse acceleration spectra at the four selected frequencies are bearing is equal to sum of elastic forces of all the elements,
shown in Fig. 7共a–d兲. A preliminary estimate of nonlinear param- n
兺F
eter k 3 is made and the iteration procedure is repeated till esti-
mated value of the nonlinear parameter converges within a limit, i.e., F⫽ xi (20)
i⫽1
which in this case is specified as 0.1%. Figure 8共a兲 shows the
estimated values of nonlinear parameter k 3 over the stages of where n is the number of rolling elements in the bearing in the
successive iteration. Final estimate of k 3 is found to be 1.75 loading zone.

Fig. 7 Acceleration response spectra for Case I: Excitation amplitudeÄ4 N
Using the condition of zero elastic force along y-axis 共since no n n

external force acts along this axis兲 and following Eq. 共19兲, defor-
mation y is obtained as
A⫽ 兺关 g⫹x cos ␩ 兴
i⫽1
i
3/2
sin ␩ i ; B⫽ 兺关 g⫹x cos ␩ 兴
i⫽1
i
3/2
sin2 ␩ i
n n
兺
i⫽1
关 g⫹x cos ␩ i 兴 3/2 sin ␩ i C⫽ 兺关 g⫹x cos ␩ 兴
i⫽1
i
3/2
sin ␩ i cos ␩ i
y⫽ (21) (24)
n
n
兺关 g⫹x cos ␩ 兴
i⫽1
i
1/2
sin2 ␩ i D⫽ 兺关 g⫹x cos ␩ 兴 i
3/2
sin2 ␩ i cos ␩ i
i⫽1
Equations 共18兲 and 共21兲 are used in Eq. 共20兲 and the bearing It can be seen that the bearing stiffness is critically dependent on
stiffness is determined as a function of deformation x as the pre-load. Table 2 summarizes the theoretical bearing stiffness
k 共 x 兲 ⫽ ⳵ F/ ⳵ x (22) parameters for different pre-load along with experimentally ob-
tained stiffness parameters. The stiffness variations are also shown
Substituting Eq. 共20兲 in Eq. 共22兲, taking into account Eq. 共21兲 the in Fig. 12 along with experimental results of earlier researchers.
bearing stiffness is expressed as a function of deformation as
n
k 共 x 兲 ⫽K n 兺关 g⫹x cos ␩ ⫺ 共 A/Bn 兲 sin ␩ 兴

i⫽1
i i 关 cos ␩ i ⫺ 兵 CBn
1/2 6 Conclusion
The stiffness parameter estimates from the experiment show
reasonably good agreement with those from available analytical
⫺AD 共 n⫺1 兲其 / 共 Bn 兲 2 sin ␩ i 兴 cos ␩ i (23)
formulations for isolated ball bearings. The analytical formulation
where is dependent on the amount of preload and hence exact compari-

Fig. 8 „a… Iterative estimates of k 3 , „Case I: Excitation
amplitudeÄ4 N … „b… Final estimate of first order kernel trans- Fig. 9 „a… Iterative estimates of k 3 , „Case II: Excitation
form, H 1 „ ␻ … „Case I: Excitation amplitudeÄ4 N … amplitudeÄ3 N … „b… Iterative estimates of k 3 , „Case III: Excita-
tion amplitudeÄ2 N …
Table 1 Linear and nonlinear estimates at different measur-

ability cases
Nonlinear parameter Natural Linear stiffness

k 3 (⫻1019 N/m3 ) frequency k 1 (⫻107 N/m) Damping
Case I: 4N 0.8750 1011.47 Hz 1.585 0.01078
Case II: 3N 1.4285 1012.47 Hz 1.535 0.01015
Case III: 2N 1.7400 1011.90 Hz 1.586 0.01002
Fig. 10 Variation in sign of real part of X „3 ␻ … around ␻ n Õ3 Fig. 11 Schematic diagram of a loaded ball bearing

Table 2 Estimated and theoretical „Ragulskis et al. †9‡; Harris, †8‡… bearing stiffness
parameters
Theoretical Stiffness 共N/m兲 Estimated Stiffness 共N/m兲

Pre-load 共␮m兲 k(x) from equation 共22兲 (k 1 ⫺3k 3 x 2 )
0.2 1.20⫻107 ⫺4.01⫻1019x 2 Case i兲: 1.585⫻107 ⫺2.625⫻1019x 2
0.3 1.47⫻107 ⫺2.18⫻1019x 2 Case ii兲: 1.535⫻107 ⫺4.285⫻1019x 2
0.4 1.69⫻107 ⫺1.42⫻1019x 2 Case iii兲: 1.586⫻107 ⫺5.240⫻1019x 2
0.5 1.89⫻107 ⫺1.02⫻1019x 2
0.6 2.08⫻107 ⫺0.61⫻1019x 2
k1 ⫽ linear stiffness parameter

k2 ⫽ square non-linear stiffness parameter
k3 ⫽ cubic non-linear stiffness parameter
m ⫽ mass of the system
x n (t) ⫽ nth order response component
⌫ 3( ␻ ) ⫽ Third order synthesized kernel factor
␵ ⫽ damping factor
␴ i (n ␻ ) ⫽ ith term in the response harmonic series,
X(n ␻ )
References
关1兴 Bedrosian, E., and Rice, S. O., 1971, ‘‘The Output Properties of Volterra Sys-
tems 共Nonlinear System with Memory兲 Driven by Harmonic and Gaussian
Fig. 12 Comparison of estimates of stiffness parameters 1–5: Input,’’ Proc. IEEE, 59共12兲, pp. 1688 –1707.
关2兴 Boyd, S., Tang, Y. S., and Chua, L. O., 1983, ‘‘Measuring Volterra Kernels,’’
Theoretical values with pre-load 0.2, 0.3, 0.4, 0.5 and 0.6 ␮m IEEE Trans. Circuits Syst., CAS-30共8兲, pp. 571–577.
respectively. †Harris †8‡ and Ragulski et al. †9‡ 6,7,8: Present 关3兴 Chua, L. O., and Liao, Y., 1989, ‘‘Measuring Volterra Kernels 共II兲,’’ Int. J. of
experimental estimates for cases I, II and III respectively. 9: Ex- Circuit Theory and Applications, 17, pp. 151–190.
perimental estimate of Tiwari †16‡ 10: Experimental estimate of 关4兴 Gifford, S. J., and Tomlinson, G. R., 1989, ‘‘Recent Advances in the Applica-
Khan †13‡ tion of Functional Series to Nonlinear Structures,’’ J. Sound Vib., 135共2兲, pp.
289–317.
关5兴 Chatterjee, A., and Vyas, N. S., 2001, ‘‘Stiffness Nonlinearity Classification
through Structured Response Component Analysis using Volterra Series,’’
son of theoretical values with experimental estimates is difficult. Mech. Syst. Signal Process., 15共2兲, pp. 323–336.
While, the manufacturer, at times may provide the preload range, 关6兴 Lee, G. M., 1997, ‘‘Estimation of Nonlinear System Parameters using Higher
the exact value of preloading of the bearing balls in the shaft- Order Frequency Response Functions,’’ Mech. Syst. Signal Process., 11共2兲, pp.
219–228.
casing assembly, especially during operations which have in- 关7兴 Chatterjee, A., and Vyas, N. S., 2002, ‘‘Nonlinear Parameter Estimation
volved wear and tear, would be difficult to determine. The sug- through Volterra Series using Method of Recursive Iteration,’’ accepted for
gested procedure gives good estimates of nonlinear parameter as publication in J. Sound Vib.
well as damping. The method is based on the assumption of linear 关8兴 Harris, T. A., 1984, Rolling Bearing Analysis, Wiley, New York.
关9兴 Ragulskis, K. M., Jurkauskas A. Y., Atstupenas, V. V., Vitkute, A. Y., and
viscous damping model. It can however be extended for identifi- Kulvec, A. P., 1974, Vibration in Bearings, Mintis Publishers, Vilnius.
cation and estimation of damping nonlinearity. 关10兴 Bannister, R. H., 1976, ‘‘A Theoretical And Experimental Investigation Illus-
trating the Influence of Nonlinearity and Misalignment on the Eight Film Co-
efficients,’’ Proc. Inst. Mech. Eng., 190, pp. 271–278.
Acknowledgment 关11兴 Choi, F. K., Braun, M. J., and Hu, Y., 1992, ‘‘Nonlinear Transient and Fre-
quency Response Analysis of a Hydrodynamic Bearing,’’ ASME J. Tribol.,
The authors wish to express their thanks to the financial aid 114, pp. 448 – 454.
being provided by the Propulsion Panel of Aeronautical Research 关12兴 Garibaldi, L., and Tomlinson, G. R., 1988, ‘‘A Procedure for Identifying Non-
and Development Board, Ministry of Defense, Government of In- linearity in Rigid Rotors Supported in Hydrodynamic and Ball/Roller Bearing
dia, in carrying out the study. System,’’ I. Mech. Proc. on Vibrations in Rotating Machinery, 4, pp. 229–234.
关13兴 Khan, A. A., and Vyas, N. S., 2001, ‘‘Application of Volterra and Wiener
Theories for Nonlinear Parameter Estimation in a Rotor-Bearing System,’’
Nomenclature Nonlinear Dyn., 24共3兲, pp. 285–304.
关14兴 Chatterjee, A., and Vyas, N. S., 2000, ‘‘Convergence Analysis of Volterra
H n( ␻ 1 , . . . , ␻ n) ⫽ nth order Volterra kernel transform Series Response of Nonlinear Systems Subjected to Harmonic Excitations,’’ J.
X(n ␻ ) ⫽ response amplitude of nth harmonic Sound Vib., 236共2兲, pp. 339–358.
c ⫽ damping coefficient 关15兴 Ewins, D. J., 1984, Modal Testing: Theory and Practice, Research Studies
Press, England.
f (t) ⫽ excitation force 关16兴 Tiwari, R., and Vyas, N. S., 1995, ‘‘Estimation of Nonlinear Stiffness Param-
g ⫽ bearing pre-load eters of Rolling Element Bearings from Random Response of Rotor Bearing
h n( ␶ 1 , . . . , ␶ n) ⫽ nth order Volterra kernel Systems,’’ Journal of Sound Vib. 187 共2兲, pp. 229–239.

Effect of Thrust Magnetic Bearing
on Stability and Bifurcation of a
Flexible Rotor Active Magnetic
Y. S. Ho Bearing System
H. Liu This paper is concerned with the effect of a thrust active magnetic bearing (TAMB) on the
stability and bifurcation of an active magnetic bearing rotor system (AMBRS). The shaft
Department of Mechanical Engineering,
is flexible and modeled by using the finite element method that can take the effects of
The Hong Kong Polytechnic University,
Hong Kong SAR, P.R. China
inertia and shear into consideration. The model is reduced by a component mode synthe-
sis method, which can conveniently account for nonlinear magnetic forces and moments
of the bearing. Then the system equations are obtained by combining the equations of the
reduced mechanical system and the equations of the decentralized PID controllers. This
L. Yu study focuses on the influence of nonlinearities on the stability and bifurcation of T
Theory of Lubrication and Bearing Institute,
periodic motion of the AMBRS subjected to the influences of both journal and thrust
Xi’an Jiaotong University,
active magnetic bearings and mass eccentricity simultaneously. In the stability analysis,
Xi’an, Shaanxi, P.R. China
only periodic motion is investigated. The periodic motions and their stability margins are
obtained by using shooting method and path-following technique. The local stability and
bifurcation behaviors of periodic motions are obtained by using Floquet theory. The
results indicate that the TAMB and mass eccentricity have great influence on nonlinear
stability and bifurcation of the T periodic motion of system, cause the spillover of system
nonlinear dynamics and degradation of stability and bifurcation of T periodic motion.
Therefore, sufficient attention should be paid to these factors in the analysis and design of
a flexible rotor system equipped with both journal and thrust magnetic bearings in order
to ensure system reliability. 关DOI: 10.1115/1.1570448兴
1 Introduction vestigators deal with the effect of thrust bearing upon the nonlin-
ear stability of T period motion of unbalanced rotor systems.
Active magnetic bearing 共AMB兲 has found wide applications in
This paper is concerned with the effect of a TAMB on the
fields such as aerospace, petrochemical and power generation instability and bifurcation of an AMBRS supported by both journal
dustries due to its advantages over traditional rolling or sliding and thrust active magnetic bearings. The rotor is flexible, and
bearing. Like conventional mechanical bearings, the AMBs are system equations are formulated by combining the equations of
nonlinear in nature. These nonlinear characteristics cause nonlin- motion of the rotor and the equations of the decentralized PID
ear motion of the rotor. Various nonlinear studies have been car- controllers. This is typically a mechanical-electrical system con-
ried out on the AMBRS 关1–3兴. However, the effect of thrust active sisting of a large number of linear components 共such as flexible
magnetic bearing 共TAMB兲 on the stability and bifurcation of the shaft elements without disk and bearing兲 and a few of nonlinear
AMBRS has received little attention in system modeling and components 共nonlinear active magnetic bearings兲. Finite elements
analysis. method 关14,15兴, and order reduction techniques are used to reduce
The TAMBs are used to balance axial loads in system and usu- the order of the model 关10,11,13兴. Floquet theory 关16 –18兴, shoot-
ally their effects are often neglected when in the process of the ing method and path-following technique 关12,19兴 are used to
controller adjustment and the dynamic analysis of system. Using analyses the stability and bifurcation of T periodic motion of the
linear analysis method, researchers have revealed that hydrody- AMBRS equipped with both journal AMBs and TAMBs. The
namic thrust bearings have great effects upon the linear stability effects of the TAMB and the mass eccentricity are discussed
of rotor systems 关4 –5兴. Because of constraints of the linear sta- especially.
bility theory, only the effects of thrust bearings on the linear sta-
bility near the equilibrium point of a balanced rotor system has 2 Formulations of Model and Reduction
been studied. Because a rotor always has mass eccentricity that
affects the stability of system and this effect cannot be taken into 2.1 Finite Element Equations of the Flexible Shaft. An
account in linear analysis, a nonlinear analysis method has to be AMBRS is depicted in Fig. 1 when the rotor is flexible. It can be
used. Due to the complexity of nonlinear analysis, nonlinear mod- divided into linear components 共the flexible shaft elements with-
els of rotor system are often simplified using assumptions such as out disk and bearing兲 and nonlinear components 共nonlinear active
a symmetrical rigid rotor or a Jeffcott rotor 关6 –9兴. However, such magnetic bearings兲. The finite element method is used to form the
equations of the motion of linear components. Here a 2-node Ti-
simplified models do not truly represent the complex system ac-
moshenko shaft element model with 8 degrees-of-freedom
curately. Therefore several investigators introduced the order re-
关14,15兴, as shown in Fig. 2, which can account for the effects of
duction methods of system to study the periodic solutions of a
inertia and shear, is adopted. Equations of the lateral motions of a
high order unbalanced rotor systems 关10–13兴. Bearings in these
flexible shaft can then be written as
studies are traditional hydrodynamic bearings, although few in-
MS ẍS ⫹GS ẋS ⫹KS xS ⫽QS ⫹f S (1)
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received November where M , G , K 苸R
S S S n⫻n
and Q , f 苸R are the mass matrices,
S S n
2002; Revised October 2002. Associate Editor: J. Cusumano. gyroscope matrices, stiffness matrices, external force vector 共in-
Here, the columns of the (n,n k ) matrix ⌽i with these elastic
eigenmodes kept are the mass normalized solutions ( ␸ Tj MS ␸ j
⫽1) of the undamped eigenproblem for ␻ k 苸(0,␻ cut )( j
⫽1, . . . ,n k ):
共 ⫺ ␻ 2j MS ⫹KS 兲 ␸ j ⫽0 (8)
The columns of the (n,n b ) matrix ⌽b with the residual flexible
modes is defined as follows:
⫺1
⌽b ⫽ 关 KS ⫺⌽i ⍀⫺2
kk ⌽i 兴
T
冋册 Ibb
0ib
(9)
where ⍀ kk is a (n k ,n k ) diagonal matrix with the kept angular

Fig. 1 Structure of a flexible rotor-active magnetic bearing eigenfrequencies lower than or equal to ␻ c 关20兴. Then
冋册冋册冋册
system
xSb ⌽bb ⌽bk pSb
xS ⫽T1 p⬘ ⇒ S ⫽ S (10)
xi ⌽ib ⌽ik pk
cluding weight force and unbalance force兲 and nonlinear force Starting from Eq. 共10兲, the degrees-of-freedom of pSb are replaced
vector of bearings respectively. For a shaft with p nodal points, by the interface degrees-of-freedom of xSb using a coordinate
the displacement vector can be written as transformation matrix T 2 to permit simple coupling of the reduced
component equations:
xS ⫽ 兵 x 1 y 1 ␸ 1 ␺ 1 ¯ x p y p ␸ p ␺ p 其 T (2)
where x j , y j , ␸ j and ␺ j ( j⫽1,2, . . . ,p) are the lateral translations
and tilting angles of the jth nodal point along the horizontal and 冋册冋 pSb
pSk
⫽
⌽⫺1
bb
0kb
⫺⌽⫺1
bb ⌽bk xb
Ikk
S
p Sk
⇒p⬘ ⫽T2 p册冋册 (11)
vertical directions, respectively. The nonlinear force vector can be
written as This results in the following total transformation
f s ⫽ 兵 0 ¯ f x j f y j 0 0 ¯ 0 0 M x k M y k ¯ O其 T (3) xS ⫽Tp T⫽T1 T2 (12)
where f x i and f y j are the horizontal and vertical of magnetic forces Applying the transformation 共12兲, the reduced component equa-
tions become:
of the journal active magnetic bearing acting on the jth point; M x k
and M y k are the magnetic moments of the thrust active magnetic TT MS Tp̈⫹TT GS Tṗ⫹TT KS Tp⫽TT QS ⫹TT f S (13)
bearing acting on the kth point. To simplify notations, the order of After reduction, the local nonlinear forces of bearing and the
the vector components is rearranged and Eq. 共1兲 can be partitioned unbalance force of disk can be easily added to the reduced linear
as equation since the interface degrees-of-freedom is available in the
冋册再冎冋册再冎冋册再冎
MSbb S
GSbb S
KSbb S reduced equation. The equation of motion of the shaft in the axial
Mbi ẍSb Gbi ẋSb Kbi xSb direction is given by
⫹ ⫹
S
Mib MiiS ẍiS S
Gib GiiS ẋiS S
Kib KiiS xiS
mz̈⫽F z (14)
QSb
再冎再
⫽ S ⫹
Qi
f Sb 共 xSb ,ẋSb 兲
0 冎 (4) From Eqs. 共10兲 and 共11兲, the equations of motion of the reduced
mechanical system is given by
If the rotor has m points on which journal and thrust active Mq̈⫹Gq̇⫹Kq⫽Q (15)
magnetic bearings act, xSb and fSb 苸R n⫻n (n b ⫽4m) can be written
as with
T T
xSb ⫽ 兵 x 1 y 1 ␸ 1 ␺ 1 ¯ x m y m ␸ m ␺ m 其 T (5) q⫽ 兵 xSb pSk z其T
f Sb ⫽ 兵 f x j f y j 0 0 ¯ 0 0 M x k M y k 其 T
For reducing the order of the degree-of-freedom of linear com-
(6)
M⫽ 冋 TT MS T⫹Md
0
0
m
册 T
G⫽ 冋 TT GS T⫹Gd
0
0
0
册 T
冋册
ponents, xS can be written as a linear combination of n c columns:
TT KS T⫹Kd 0 T
x ⫽T1 p⬘
S
(7) K⫽
0 0
where T1 ⫽ 关 ⌽b ⌽i 兴
Q⫽ 兵 TT QS 0 其 T ⫹ 兵 TT f S 0 其 T ⫹ 兵 FTdex 0 F z 其 T
where Md , Gd and Kd are the mass, damping and stiffness matri-
ces of disks; m is the mass of the rotor; Fdex is unbalance forces
caused by mass eccentricity of disks. The behavior of this nonlin-
ear system depends on the rotating frequency ␻ of the shaft and
the eccentricities e explicitly.
The eigenfrequencies ␻ ⬍ ␻ cut of Eq. 共13兲 are accurate for the
case QS ⫽0 and GS ⫽0, with ␻ cut the highest cut-off eigenfre-
quency in the reduction, and higher eigenfrequencies will be in-
accurate. Because nonlinear systems can generate frequencies
higher than their excitation frequency, ␻ cut has to be chosen
higher than the maximum excitation frequency.
2.2 Electromagnetic Force of Journal Bearing. An eight-
Fig. 2 Shaft finite element model pole journal AMB shown in Fig. 3 is used for analysis. To sim-

Fig. 4 A thrust magnetic bearing in operation
Fig. 3 End view of an eight-pole journal bearing

h 共i 1 兲 ⫽c z ⫹z⫺ ␸ r i sin ␪ ⫺ ␺ r i cos ␪ ,
h 共o1 兲 ⫽c z ⫹z⫺ ␸ r o sin ␪ ⫺ ␺ r o cos ␪ ,
plify the notations, subscripts indicating bearing numbers are ig- (18)
nored. The forces in this journal magnetic bearing can be h 共i 2 兲 ⫽c z ⫺z⫹ ␸ r i sin ␪ ⫹ ␺ r i cos ␪ ,
expressed as
h 共o2 兲 ⫽c z ⫺z⫹ ␸ r o sin ␪ ⫹ ␺ r o cos ␪ ,
f x ⫽ f r ⫺ f l ⫹ ␣ xy 共 x/c r 兲共 f t ⫹ f b 兲 It is assumed that the magnetic flux from the point (r i , ␪ ) on the
(16) inner annulus runs entirely back to the point (r o , ␪ ) on the outer
f y ⫽ f t ⫺ f b ⫹ ␣ xy 共 y/c r 兲共 f r ⫹ f l 兲 annulus as shown in Fig. 5. Therefore the relationship between
radii r i and r o is
with
␮ 0 N r2 A r 共 I 0x ⫹i x 兲 2 ␮ 0 N r2 A r 共 I 0x ⫺i x 兲 2 R 4 ⫺R 3
f r⫽ ⫻ f 1⫽ ⫻ r o ⫽R 4 ⫹ ␣ 共 r i ⫺R 1 兲 with ␣ ⫽⫺ (19)
4 共 c r ⫹x 兲 2 4 共 c r ⫺x 兲 2 R 2 ⫺R 1
(17)
The forces and moments provided by the thrust magnetic bear-
␮ 0 N r2 A r 共 I 0y ⫹i y 兲 2
␮ 0 N r2 A r 共 I 0y ⫺i y 兲 2 ing can be expressed as 关22兴
f t⫽ ⫻ f b⫽ ⫻
4 共 c r ⫹y 兲 2 4 共 c r ⫺y 兲 2
F z ⫽F 共z1 兲 ⫹F 共z2 兲
where ␣ xy is the geometric coupling coefficient with an average
value of 0.16 关21兴. In this study, ␣ xy is taken to be 0.16.
M x ⫽M 共x1 兲 ⫹M 共x2 兲 (20)
2.3 Magnetic Forces and Moments of Thrust Bearing. A
two-pole TAMB in operation is shown in Fig. 4. When tilting of
M y ⫽M 共y1 兲 ⫹M 共y2 兲
runner occurs, the right and left air gaps at those points on the
inner and outer annuli are different and can be expressed as where
F 共z1 兲 ⫽⫺ 冕冕 0
2␲
R1
R2 关 ␣␮ 0 N 共 I 0z ⫹i z 兲兴 2
2␮0 共1兲冋共 r o /r i 兲 2
共 1 兲 r i dr i d ␪ ⫹
h o ⫺ ␣ 共 r o /r i 兲 h i 册冕冕 0
2␲
R4
R3 关 ␮ o N 共 I 0z ⫹i z 兲兴 2
2␮0
1
关 h 共o1 兲 ⫺ ␣ 共 r o /r i 兲 h 共i 1 兲兴 2
r o dr o d ␪
F 共z2 兲 ⫽ 冕冕 2␲
0 R1
R2 关 ␣␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 冋共 r o /r i 兲
h 共o2 兲 ⫺ ␣ 共 r o /r i 兲 h 共i 2 兲
2
册
r i dr i d ␪ ⫺ 冕冕 2␲
0 R4
R3 关 ␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 共2兲
1
关 h o ⫺ ␣ 共 r o /r i 兲 h 共i 2 兲兴 2
r o dr o d ␪
M 共x1 兲 ⫽⫺ 冕冕 0
2␲
R1
R2 关 ␣␮ 0 N 共 I 0z ⫹i z 兲兴 2
2␮0 冋 ro
h 共o1 兲 ⫺ ␣ 共 r o /r i 兲 h 共i 1 兲册 2
cos ␪ dr i d ␪ ⫹ 冕冕 0
2␲
R4
R3 关 ␮ 0 N 共 I 0z ⫹i z 兲兴 2
2␮0 冋 ro
cos ␪ dr o d ␪
M 共x2 兲 ⫽ 冕冕
0
2␲ R2
R1
关 ␣␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 冋 ro
h o ⫺ ␣ 共 r o /r i 兲 h 共i 2 兲
共2兲册 2
cos ␪ dr i d ␪ ⫺ 冕冕 0
2␲ R3
R4
关 ␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 冋 ro
共2兲册 2
cos ␪ dr o d ␪
M 共y1 兲 ⫽⫺ 冕冕 0
2␲
R1
R2 关 ␣␮ 0 N 共 I 0z ⫹i z 兲兴 2
2␮0 冋 ro
sin ␪ dr i d ␪ ⫹ 冕冕 0
2␲
R4
R3 关 ␮ 0 N 共 I 0z ⫹i z 兲兴 2
2␮0 冋 ro
sin ␪ dr o d ␪
M 共y2 兲 ⫽ 冕冕0
2␲ R2
R1
关 ␣␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 冋 ro
共2兲册 2
sin ␪ dr i d ␪ ⫺ 冕冕 0
2␲ R3
R4
关 ␮ 0 N 共 I 0z ⫺i z 兲兴 2
2␮0 冋 ro
共2兲册 2
sin ␪ dr o d ␪
(21)

When state variables X⫽(q,q̇,Ibb ) T are introduced, the corre-
sponding system equations in state space are
Ẋ⫽ 再 q̇
M⫺1 共 Q⫺Gq̇⫺Kq兲
Aa As 共 Kp ẋbb ⫹Ki xbb ⫹Kd ẍbb 兲
冎 (28)
The stability and bifurcation problem of the reduced system can

be solved by using shooting method and path-following tech-
nique.
3 Method of Solution
It is assumed that the external load acting on the system is
Fig. 5 Magnetic flux path in a thrust magnetic bearing periodic with period T:
Qex 共 t 兲 ⫽Qex 共 t⫹T 兲 (29)
2.4 Equations of Sensors, Amplifiers and PID Controllers. The steady-state behavior of the system, i.e., the attractor that is
The signal flow in a rotor-magnetic bearing system is shown in reached after the transient has damped out, might be periodic,
Fig. 6. The power amplifier can be modeled as a first-order sys- quasi-periodic or chaotic. The periodic solutions of the system can
tem, whose transfer function is become unstable in certain intervals of system parameters: such as
the angular speed ␻ of the rotor, the mass eccentricity e, and the
Aa axial force F z , etc. Such instability is due to the nonlinear char-
Ga 共 s 兲 ⫽ (22)
1⫹Ta s acteristics of the AMBs. Generally, in these intervals, the maxi-
mum response of the nonlinear system will increase and also ad-
where Aa is gain, and Ta is time constant. ditional vibration frequencies occur compared to the linear
The displacement sensor can also be represented by a first-order system. This may lead to shaft rubbing which is undesirable in
system. practice. So for practical application, it is important to determine
As the unstable intervals and the kinds of dynamic behaviors of the
Gs 共 s 兲 ⫽ (23) rotor that result in these intervals: periodic, quasi-periodic or
1⫹Ts s
chaotic.
where As is gain, and Ts is time constant.
In this investigation, the controller is taken to be a PID control- 3.1 Periodic Solutions. Periodic solutions are calculated by
ler as it is widely used. The transfer function of a PID controller is solving a two-point boundary value problem, which is defined by
Eq. 共28兲 supplemented with the boundary condition X(t)⫽X(t
Ki Kd s ⫹T). It can be written as:
Gc 共 s 兲 ⫽Kp ⫹ ⫹
再
(24)
s 1⫹Td s
Ẋ⫽f共 X,t, ␮ 兲
(30)
where Kp is proportional gain, Ki is integral gain, Kd is derivative X共 t 兲 ⫽X共 t⫹T 兲
再冎
gain and Td is time constant, For an ideal system, time constant
Ta ⫽Ts ⫽Td ⫽0. The transfer function of the whole system is thus q̇
given by with f共 X,t, ␮ 兲 ⫽ M⫺1 共 Q⫺Gq̇⫺Kq兲
冉 Ki
G共 s 兲 ⫽Ga 共 s 兲 "Gs 共 s 兲 "Gc 共 s 兲 ⫽Aa As Kp ⫹ ⫹Kd s
s 冊 (25)
Aa As 共 Kp ẋbb ⫹Ki xbb ⫹Kd ẍbb 兲
where ␮ is a system parameter. It can be the angular speed ␻, or
the mass eccentricity e, and or the axial force F z , etc.
and the corresponding differential equation can be written as In the shooting method, Eq. 共30兲 is integrated by means of a
numerical time integration method over the period T and a T pe-
İbb ⫽Aa As 共 Kp ẋbb ⫹Ki xbb ⫹Kd ẍbb 兲 (26)
riodic solution can be found if the following criterion is satisfied:
H共 Xs , ␮ 兲 ⫽0 (31)
with H共 Xs , ␮ 兲 ⫽Xs 共 t 0 ⫹T 兲 ⫺Xs 共 t 0 兲
For a given ␮ ⫽ ␮ s , the corresponding solution Xs can be ob-
and N J is the number of journal AMBs. tained from Eq. 共31兲 using the iterative Newton process. The Jaco-
2.5 System Equations. Combining the equations for the ro- bian matrix can be obtained as
tor, the sensors, the amplifiers and the PID controllers, the system
equations are obtained Table 1 Parameters of journal and thrust AMBs
再 Mq̈⫹Gq̇⫹Kq⫽Q
İbb ⫽Aa As 共 Kp ẋbb ⫹Ki xbb ⫹Kd ẍbb 兲
(27) Bearing type Parameter
radial clearance
Value
0.0004 m
bias current 4A
Journal bearings winding number 57
width 0.08 m
diameter 0.16 m
total axial clearance 0.0006 m
bias current 4A
winding number 143
diameter 0.1 m
Thrust bearing R1 0.04 m
R2 0.065 m
R3 0.08 m
Fig. 6 Signal flow in a rotor-active magnetic bearing system R4 0.1 m

Table 2 Parameters of controllers, sensors and power consists of a predictor-corrector mechanism. Starting from a
amplifiers known solution Xn at ␮ ⫽ ␮ n , the prediction of n-step is
再冋册 ⫺1
Parameter Value ⳵ H共 X, ␮ 兲 ⳵ H共 X, ␮ 兲
Xn⫹1 ⫽Xn ⫺ ⫻ ⫻⌬ ␮
A a 共1/⍀兲 1 ⳵ Xs ⳵␮ (34)
A 2 共V/m兲 7800
kp 3.8 ␮ n⫹1 ⫽ ␮ n ⫹⌬ ␮
k i 共1/s兲 200
k d 共s兲 0.01 Subsequently this periodic solution is corrected by shooting
method at ␮ ⫽ ␮ n⫹1 . Here ⳵ H(X, ␮ )/ ⳵ ␮ can be calculated by
numerical time integration of Eq. 共34兲 about the trajectory Xs (t 0
⫹t), and
⳵H d ⳵ f共 t, ␮ ,X兲 ⳵ f共 t, ␮ ,X兲
⫽J⫺I (32) 共 ␦ S␮ 兲 ⫽ ⫻ ␦ S␮ ⫹ (35)
⳵ Xs dt ⳵X ⳵␮
where J is calculated by numerical time integration of equation with ␦ S␮(t 0 )⫽0 and ␦ S␮(t 0 ⫹T)⫽ ⳵ H(X, ␮ )/ ⳵ ␮ .
共30兲 linearized at those points along the trajectory Xs (t 0 ⫹t), and
冉冊
3.3 Local Stability and Bifurcation: Floquet Theory. The
d ⳵f local stability of periodic motion of a nonlinear system can be
共 ␦ S兲 ⫽ ⫻␦S (33) determined from Floquet theory 关16 –18兴. The procedure involves
dt ⳵X
calculating the eigenvalues 共Floquet multipliers兲 of the matrix J,
with ␦ S(t 0 )⫽I and ␦ S(t 0 ⫹T)⫽J. This is called shooting method which is also called monodromy matrix. For a stable periodic
关12,19兴. solution, all the Floquet multipliers should be within the unit
3.2 Continuation of Periodic Solutions—Path-Following circle in the complex plane. When the unit circle in the complex
Technique. The problem of continuation of solutions in general plane is crossed by
is to find the whole solution branches, i.e., to obtain a solution at 1. a real eigenvalue on the positive real axis, it is a ‘‘cycle-
␮ ⫽ ␮ j⫹1 when the exact solution at ␮ ⫽ ␮ j is known. Several fold’’ or ‘‘transcritical’’ bifurcation;
continuation schemes are available. In the present study, the path 2. a pair of complex conjugate eigenvalue, it is a ‘‘secondary
following technique 关12兴 is used to investigate how a periodic Hopf’’ bifurcation;
solution is influenced by a change of ␮. In essence, the technique 3. a real eigenvalue on the negative real axis, it is a ‘‘period-
doubling’’ bifurcation.
Here, the ‘‘eigenvalue’’ above mentioned, which has the largest
magnitude among all eigenvalues of the monodromy matrix, is
called the leading Floquet multiplier.
4 Numerical Results and Discussions

The rotor system depicted in Fig. 1 is analyzed. Parameters of
the journal and thrust AMBs are given in Table 1. The parameters
of sensors, amplifiers and controllers for the five channels are the
same and given in Table 2. The axial load is taken to be 0N and
the rotational speed is 30,000rev/min. The rotor and the journal
bearings are assumed to be coaxial in static state.
The shaft is discretized into 16 elements and 17 nodes. The
cut-off rotating speed ␻ cut is chosen to be 120000rev/min which
is greater than two times the expected bandwidth of the response.
The number of retained eigenmodes ⌽k of the linear component is
8, 2 of which are rigid body modes, and other 6 modes are elastic
eigenmodes with eigenfrequencies of 4908rev/min, 12666rev/min,
Fig. 7 A e 0 À ␻ diagram of Stable and unstable regions of T
period motion
27445rev/min, 38030rev/min, 46473rev/min, 62026rev/min, re-
Fig. 8 Hopf T periodic solution without TAMB „a… Hopf T periodic solution at ␻
Ä120000 revÕmin „b… A amplitude-frequency diagram of x a

Fig. 9 Hopf T periodic solution with TAMB „a… Hopf T periodic solution at ␻ Ä70000 revÕmin „b…
A amplitude-frequency diagram of x a
Fig. 10 Quasi-periodic solution without TAMB „ e 0 Ä4 ␮ m… „a… Quasi-periodic solution at ␻

Ä113405 revÕmin „b… The orbit of the center of the rotor at journal bearing ‘a’ „c… Poincare maps „d…
Time series of x a „e… Amplitude-frequency diagram of x a

Fig. 11 Quasi-periodic solution with TAMB „ e 0 Ä4 ␮ m… „a… Quasi-periodic solution at ␻
Ä67345 revÕmin „b… The orbit of the center of the rotor at journal bearing ‘a’ „c… Poincare maps
„d… Time series of x a „e… Amplitude-frequency diagram of x a
spectively. The number of interface degrees-of-freedom is 16. Table 3 Leading Floquet multiplier
This include eight displacements and eight tilting degree-of-
freedom for the four couplings between the disks and the shaft Not consider TAMB Consider TAMB
(x a ,y a , ␸ a , ␺ a ,x b ,y b , ␸ b , ␺ b ,x c ,y c , ␸ c , ␺ c ,x d ,y d , ␸ d , ␺ d ), in e0 leading Floquet multiplier leading Floquet multiplier
which 4 degrees-of-freedom are the displacements for two journal 共␮m兲共modulus兲共modulus兲
AMBs (x a ,y a ,x c ,y c ) and 2 degrees-of-freedom are the tilting 5 0.992395⫾i0.0726290 0.992214⫾i0.0751451
angles for a TAMB ( ␸ d , ␺ d ). In present study, only unbalance 共0.995049兲共0.995055兲
forces caused by mass eccentricities of four disks (e a ⫽e b ⫽e c 10 0.992400⫾i0.0726408 0.992219⫾i0.0751573
⫽e d ⫽e 0 ) are studied. 共0.995055兲共0.995061兲
15 0.992411⫾i0.0726599 0.992230⫾i0.0751772
The combined effect of e 0 and ␻ is studied with and without 共0.995067兲共0.995074兲
the effects of the TAMB considered. The stable T period motion 20 0.992429⫾i0.0726863 0.992249⫾i0.0752049
共harmonic motion兲 and its unstable regions are plotted in an e 0 - ␻ 共0.995088兲共0.995095兲
diagram in Fig. 7, where the line is the bifurcation borderline, 22.7 0.992444⫾i0.0727040 0.960353⫾i0.279283
共0.995104兲共1.00014兲
with the modulus of leading Floquet multiplier equals to one. For 25 0.992460⫾i0.0727215 0.977981⫾i0.273811
values of e 0 and ␻ below the line, the motion of the system is 共0.995121兲共1.01559兲
stable T period motion. For that above the line, the motion is

Fig. 12 Stable T periodic motion at ␻ Ä30000 revÕmin, e 0 Ä10 ␮ m „a… With TAMB „b… Without
TAMB
Fig. 13 Stable Quasi-periodic motion at ␻ Ä30000 revÕmin, e 0 Ä22.5 ␮ m with TAMB „a… Stable
Quasi-periodic motion „b… Poincare maps
quasi-period or unstable. When values of e 0 and ␻ cross the line odic solutions are different; the amplitude of the Hopf T periodic
from below to above, a pair of Floquet multiplier crosses a unit solution with TAMB considered is smaller than that with TAMB
circle in the complex plane and secondary Hopf bifurcation not considered.
occurs. 2 Mass Eccentricity e 0 ⫽4 ␮ m. The T periodic motion of
It can be seen from Fig. 7 that the stability criteria rotating the system produces a quasi-periodic solution after the second
speed of the system is decreased when TAMB is considered in Hopf bifurcation at ␻ ⫽113400 rev/min when TAMB is not con-
modeling. As an example, the critical stability speed of system sidered. The corresponding quasi-periodic solution at ␻
with e 0 ⫽0 ␮ m is decreased by 41% from ␻ ⫽119000 rev/min to ⫽113405 rev/min is depicted in Fig. 10共a兲, and the locus of the
␻ ⫽69850 rev/min, and the critical stability speed is decreased by center of the runner is shown in Fig. 10共b兲. Its Poincare maps
42% from ␻ ⫽113100 rev/min to ␻ ⫽65870 rev/min when e 0 is
increased from 0 ␮m to 5 ␮m. This indicates that the effects of
TAMB and mass eccentricity e 0 on the stability of system are both
significant.
4.1 Effects of TAMB and Mass Eccentricity
1 Mass Eccentricity e 0 ⫽0 ␮ m. When the effect of TAMB is
not considered, the equilibrium point of the system lost stability
and a Hopf T periodic solution appears at ␻ ⫽119000 rev/min.
The corresponding Hopf T periodic solution at ␻
⫽120000 rev/min is depicted in Figure 8共a兲, and a corresponding
amplitude-frequency diagram is shown in Fig. 8共b兲. When TAMB
is considered, the Hopf bifurcation of the equilibrium point occurs
at ␻ ⫽69850 rev/min, and a stable Hopf T periodic solution ap-
pears at ␻ ⫽70000 rev/min which is depicted in Fig. 9共a兲. The
corresponding amplitude-frequency diagram is shown in Fig. 9共b兲.
It is noted that both two Hopf T periodic solutions exhibit typi-
cally the first mode whirl of system, and the frequencies are both Fig. 14 Stable periodic motion at ␻ Ä30000 revÕmin, e 0
about 24.4 Hz. However, the motions of these two Hopf T peri- Ä22.5 ␮ m without TAMB

shown in Fig. 10共c兲 reveal that the motion is quasi-periodic as the xb ⫽ interface or local nonlinearities DOF’s
Poincare points lie on closed curves. The corresponding time se- of shaft
ries is shown in Fig. 10共d兲, and its corresponding amplitude- xi ⫽ unloaded internal DOF’s of shaft
frequency diagram is shown in Fig. 10共e兲. When the TAMB is QS ⫽ external force vector of shaft
considered, the second Hopf bifurcation of the harmonic motion f S ⫽ nonlinear force vector of bearings act on
occurs at ␻ ⫽67340 rev/min. A stable quasi-periodic solution ap- shaft
pears at ␻ ⫽67345 rev/min, which is depicted in Fig. 11共a兲, and R ⫽ real number
the locus of the center of the runner is shown in Fig. 11共b兲. Its n ⫽ numbers of displacement DOF’s of shaft
Poincare maps shown in Fig. 11共c兲, reveal that the motion is n b ⫽ numbers of interface or local nonlineari-
quasi-periodic as the Poincare points lie on closed curves. The ties DOF’s of shaft
corresponding time series is shown in Fig. 11共d兲, and its corre- n i ⫽ numbers of unloaded internal DOF’s of
sponding amplitude-frequency diagram is shown in Fig. 11共e兲.
shaft
These two quasi-periodic solutions have two frequencies. One
⍀ ⫽ angular velocity of rotor
is about 24.4 Hz corresponding to the Hopf T periodic solution
␻ ⫽ angular speed of rotor
frequency of balanced rotor system, and another corresponds to
␻ cut ⫽ cut off frequency
harmonic frequency. However, the motions in the lateral direc-
x, y, z ⫽ Cartesian coordinates
tions corresponding to above two solutions are very different. The
motion of the quasi-periodic solution considering TAMB is typi- ␸, ␺ ⫽ deflection angles of shaft
cally the first mode whirl, and that of the quasi-periodic solution ⌽b ⫽ residual flexibility modes
not considering TAMB is typically the second mode whirl. ⌽k ⫽ kept elastic eigenmodes
⍀kk ⫽ diagonal matrix with the kept angular
4.2 Effects of TAMB and Mass Eccentricity on Leading eigenfrequencies
Floquet Multiplier. The leading Floquet multiplier of the T pe- T,T1 ,T2 ⫽ coordinate transformation matrix
riodic motion of the system and its modulus for different mass m ⫽ mass of whole rotor
eccentricity e 0 is shown in Table 3. It can be seen that all the Md ⫽ mass matrices of disks
moduli of leading Floquet multiplier are increased when TAMB is Gd ⫽ damping matrices of disks
considered. Figures 12共a兲 and 12共b兲 show the stable T periodic Kd ⫽ stiffness matrices of disks
motion for e 0 ⫽10 ␮ m with and without TAMB respectively. Q ⫽ force vector of the system
These two T periodic motions are almost the same. Fdex ⫽ excited forces of rotor
However, for e 0 ⫽22.5 ␮ m, the moduli of the leading Floquet e 0 ⫽ mass eccentricity of couple rotor
multiplier is increased from 0.995104 to 1.00014 when TAMB is M ⫽ mass matrices of the system
considered, which means the T periodic motion lost its stability. G ⫽ damping matrices of the system
Figures 13 and 14 show the stable quasi-periodic motion and pe- K ⫽ stiffness matrices of the system
riodic motion of the system with and without TAMB respectively. q ⫽ displacement vector of the system
So the effect of TAMB on system stability is significant. ␣ xy ⫽ geometric coupling coefficient of journal
active magnetic bearing
5 Conclusions A r ⫽ pole area of journal active magnetic
bearing
In this paper, the effect of a TAMB on an active magnetic N r ⫽ number of winding turns of journal ac-
bearing-flexible rotor system has been investigated. The finite el- tive magnetic bearing
ement method was applied to discretize the flexible shaft. Using ␮ 0 ⫽ permeability of air, 4 ␲ ⫻10⫺7 N/A2
the component mode synthesis method, the degrees of freedom of N ⫽ number of winding turns of thrust active
equation is reduced. The system equations are obtained by com- magnetic bearing
bining the equations of the reduced mechanical system and the
c r ,c z ⫽ clearance of journal and thrust active
equations of the decentralized PID controllers. Shooting method
magnetic bearing
and path-following technique are used to get the periodic motions
i x ,i y ,i z ⫽ control current of rotor journal and
and their stability margins. The local stability and bifurcation be-
thrust active magnetic bearing
haviors of the periodic motion are obtained by Floquet theory. The
f x , f y ⫽ nonlinear forces of x and y direction for
results indicate that TAMB and mass eccentricity have great in-
fluence on nonlinear stability and bifurcation of the T periodic journal active magnetic bearing
motion of system, and causes degradation of stability and bifur- I 0x ,I 0y ,I 0z ⫽ bias current of rotor journal and thrust
cation of T periodic motion. Therefore sufficient attentions should active magnetic bearing
be paid to the spillover effect due to the neglect of TAMB in the f r , f l , f t , f b ⫽ nonlinear forces of right, left, top and
dynamic analysis and design of the AMBRS in order to ensure bottom direction for journal active mag-
system reliability. netic bearing
R 1 ,R 2 ,R 3 ,R 4 ⫽ radii of inner and outer annulus for
thrust active magnetic bearing
Acknowledgment hi( 1 ) ,hi( 2 ) ,h(o1 ) ,h(o2 ) ⫽ right and left air gaps at a point on the
The work described in this paper was fully supported by a grant inner and outer annulus of thrust active
from the Research Grants Council of the Hong Kong Special Ad- magnetic bearing
ministrative Region, China 共Project No. PolyU5115/98E兲. It is F z ,M x ,M y ⫽ nonlinear forces and moment of thrust
also supported by the National Natural Science Foundation, China active magnetic bearing
共Project No. 19990511兲. Ga ,Gs ,Gc ⫽ transfer function vector of amplifier,
sensor and controller for system
Nomenclature Aa ,As ⫽ gain vector of amplifier and sensor for
system
MS ⫽ mass matrices of shaft Ta ,Ts ,Td ⫽ time constant vector of amplifier and
GS ⫽ damping matrices of shaft sensor for system
KS ⫽ stiffness matrices of shaft Kp ,Ki ,Kd ⫽ proportional gain, integral gain and de-
xS ⫽ displacement vector of shaft rivative gain

xbb ⫽ nonlinearities DOF’s of active magnetic 关8兴 Kim, Y. B., and Noah, S. T., 1990, ‘‘Bifurcation Analysis for a Modified
bearings Jeffcott Rotor with Bearing Clearance,’’ Nonlinear Dynamics, 1, pp. 221–241.
关9兴 Choi, S. K., and Noah, S. T., 1994, ‘‘Mode-Locking and Chaos in a Jeffcott
Ibb ⫽ current vector act on nonlinearities Rotor with Bearing Clearance,’’ ASME J. Appl. Mech., 61, pp. 131–138.
DOF’s of active magnetic bearings 关10兴 Nelson, H. D., Mechan, W. I., Fleming, D. P., and Kascak, A. F., 1983, ‘‘Non-
N J ⫽ number of JAMBs linear Analysis of Rotor Bearing System Using Component Mode Synthesis,’’
G ⫽ transfer function vector of the whole ASME paper No. 83-GT-303.
system 关11兴 Nataraj, C., and Nelson, H. D., 1989, ‘‘Periodic Solutions in Rotor Dynamic
X ⫽ state variables of system System With Nonlinear Supports: A General Approach,’’ ASME J. Vibr.
J ⫽ Jacobian matrix Acoust., 111, pp. 187–193.
关12兴 Fey, R. H. B., Van Campen, D. H., and de Kraker, A., 1996, ‘‘Long Term
␮ ⫽ system parameter
Structural Dynamics of Mechanical Systems With Local Nonlinearities,’’
Subscripts ASME J. Vibr. Acoust., 118, pp. 147–153.
关13兴 Zheng, T., and Hascbe, N., 2000, ‘‘Nonlinear Dynamic Behaviors of a Com-
S ⫽ shaft plex Rotor-Bearing System,’’ ASME J. Appl. Mech., 67, pp. 485– 495.
共1兲, 共2兲 ⫽ right and left annuli of thrust active 关14兴 Nelson, H. D., 1980, ‘‘A Finite Rotating Shaft Element Using Timoshenko
magnetic bearing Beam Theory,’’ ASME J. Mech. Des., 102, pp. 793– 803.
关15兴 Lalanne, M., and Ferraris, G., 1997, Rotordynamics Prediction in Engineering,
John Wiley, New York.
References
关16兴 Iooss, G., and Joseph, D. D., 1980, Elementary Stability and Bifurcation
关1兴 Virgin, L. N., Walsh, T. F., and Knight, J. D., 1995, ‘‘Nonlinear Behavior of a Theory, Springer-Verlag, New York.
Magnetic Bearing System,’’ ASME J. Eng. Gas Turbines Power, 117共7兲, pp. 关17兴 Parker, T. S., and Chua, L. O., 1989, Practical Numerical Algorithms for
582–588. Chaotic System, Springer-Verlag, New York.
关2兴 Chinta, M., Palazzolo, A. B., and Kascak, A., 1996, ‘‘Quasiperiodic Vibration
关18兴 Seydel, R., 1988, From Equilibrium to Chaos, Practical Bifurcation and Sta-
of a Rotor in a Magnetic Bearing With Geometric Coupling,’’ Proc. 5th Inter-
national Symposium on Magnetic Bearing, Kanazawa, Japan, pp. 147–152. bility Analysis, Elsevier, New York.
关3兴 Chinta, M., and Palazzolo, A. B., 1998, ‘‘Stability and Bifurcation of Rotor 关19兴 Sundararajan, P., and Noah, S. T., 1997, ‘‘Dynamics of Forced Nonlinear Sys-
Motion in a Magnetic Bearing,’’ J. Sound Vib., 214共5兲, pp. 793– 803. tems Using Shooting/Arclength Continuation Method-Application to Rotor
关4兴 Mittwollen, N., Hegel, T., and Glienicke, J., 1991, ‘‘Effect of Hydrodynamic System,’’ ASME J. Vibr. Acoust., 119共1兲, pp. 10–20.
Thrust Bearings on Lateral Shaft Vibration,’’ ASME J. Tribol., 113共4兲, pp. 关20兴 Craig, R. R., Jr., 1985, ‘‘A Review of Time-Domain and Frequency-Domain
811– 818. Component Modes Synthesis Methods,’’ Combined Experimental/Analytical
关5兴 Yu, L., and Bhat, R. B., 1995, ‘‘Coupled Dynamics of a Rotor-Bearing System Modeling of Dynamic Structural Systems Using Substructure Synthesis, D. R.
Equipped with a Hydrodynamic Thrust Bearing,’’ Shock and Vibration 2共1兲, Martinez and A. K. Miller, eds., ASCE/ASME, New York, NY, pp. 1–31.
pp. 1–14. 关21兴 Knight, J. D., Xia, Z., and McCaul, E. B., 1992, ‘‘Forces in Magnetic Journal
关6兴 Lund, J. W., and Nelson, H. B., 1980, ‘‘Instability Threshold of an Unbalanced
Bearings: Nonlinear Computation and Experimental Measurement,’’ Proc. 3rd
Rigid Rotor in Short Journal Bearings,’’ Proc. Second International Confer-
ence on Vibration in Rotating Machinery, Cambridge, UK. International Symposium on Magnetic Bearing, Alexandria, VA, pp. 441– 450.
关7兴 Brancati, R., Rocca, E., Rosso, M., and Rosso, R., 1995, ‘‘Journal Orbits and 关22兴 Ho, Y. S., Yu, L., and Liu, H., 1999, ‘‘Rotor Dynamic Coefficients of a Thrust
Their Stability for Rigid Unbalanced Rotors,’’ ASME J. Tribol., 117, pp. 709– Active Magnetic Bearing Considering Runner tilt,’’ Proc. Inst. Mech. Eng.,
716. Part J: J. Eng. Tribol., 213共J6兲, pp. 451– 462.

Torsional Vibration Analysis of
Complicated Multi-Branched
Chun-Ping Zou
School of Mechanical and Power Engineering,
Shanghai Jiao Tong University,
Shafting Systems by Modal
Shanghai, People’s Republic of China;
Department of Mechanical and Power Synthesis Method
Engineering,
East China Shipbuilding Institute, The torsional vibration calculations of the complicated multi-branched system with rigid
Zhenjiang, People’s Republic of China connection and flexible connections made up of elastic-coupling parts are very difficult to
perform using conventional methods. In this paper, a modal synthesis method of torsional
vibration analysis for the system is proposed. This approach is an improved method of
Duan-Shi Chen Hurty’s fixed-interface and Hou’s free-interface modal synthesis methods. Because of the
introduction of flexible substructure, the improved modal synthesis method can effectively
Hong-Xing Hua treat the complicated system in which there exists a rigid connection and a flexible
connection that is formed by an elastic-coupling part. When the calculation is performed,
School of Mechanical and Power Engineering, the complicated multi-branched system is divided into several substructures that are ana-
Shanghai Jiao Tong University, lyzed by FEM (finite element method) except the special elastic-coupling part that is
Shanghai, People’s Republic of China defined as flexible substructure and treated individually. The efficiency of modal synthesis
is improved by choosing suitable number of lower-frequency modes in modal synthesis. As
an example of an application of this method, the analysis of torsional vibration of a
cam-type engine shafting system is carried out both numerically and experimentally. The
results show that the above kind of multi-branched shafting system can be analyzed
effectively by the proposed method. 关DOI: 10.1115/1.1569949兴
1 Introduction can be traced to Hurty 关8,9兴. The general idea is to regard a given
structure as an assemblage of substructures acting together in
There are several kinds of methods for torsional vibration cal-
some fashion. To guarantee that the various substructures do not
culation of rotating shafting system, namely Holzer’s method,
act independently but as parts of the whole structure, Hurty intro-
transfer matrix method and FEM. The well-known Holzer’s
duced the concept of constraint modes. Generally, the constraint
method, a simple and systematic approach to calculating the fre-
modes enforce geometric compatibility at a preselected set of
quencies and mode shapes of systems, was used originally for
points on the boundary between any two adjacent substructures
analysis of an undamped system. Den Hartog and Li 关1兴 improved
关10兴. Hou advocates that substructure modes are not necessarily
the Holzer’s method by using complex numbers. Pestel and
constrained and can be free-free 关11兴. The use of unconstrained
Leckie 关2兴 introduced the transfer matrix method with the point
modes is also advocated by Goldman and Dowell 关12,13兴. Since
and field transfer matrices. Later, Sankar 关3兴 and Dawson and
publication of the original work by Hurty and Hou, there have
Davies 关4兴 improved this approach. Huang and Horng 关5兴 used
been many suggestions for improvement 关14 –19兴. One of the
complex numbers to extend the transfer matrix method with the
most important improvements is that the effect of modes not re-
Newton-Raphson technique to analyze the torsional vibration for
tained explicitly is considered by MacNeal 关19兴 and by Rubin
damped systems. The method extended by Huang and Horng
关16兴. The procedure described by MacNeal employs hybrid com-
eliminates the operation of the inverse matrix because the deriva-
ponent modes, i.e., modes obtained with some interface coordi-
tives of angular displacement and the torque are used directly with
nates free and others fixed, and uses statically derived modes to
the Newton-Raphson technique to determine the eigenvalues of
improve accuracy. The procedure suggested by Rubin employs
torsional vibration system. The major drawbacks of the transfer
residual flexibility to a second order of approximation.
matrix approach 共as noted by Firoozian and Stanway 关6兴兲 are that
The modal synthesis methods discussed above have one aspect
the information obtained is limited to stability assessment and the
in common, namely they all regard the connection between any
prediction of critical speeds, and that inherent numerical instabili-
two substructures as a rigid connection. However, flexible connec-
ties are liable to occur in the analysis of more complex systems.
tion made up of elastic-coupling parts is widely applied to engi-
Li 关7兴 used FEM to analyze the crankshaft torsional vibration of
neering structure with multi-branched shafting system, e.g., the
diesel engine. One of the characteristics of FEM is that a very
elastic coupling of marine propulsion system. The boundary dis-
large number of degrees of freedom are required. All these meth-
placement that exists in-between substructures of such structure is
ods, namely Holzer’s method, transfer matrix method and FEM,
incompatible. Hale and Meirovitch 关20兴 proposed the concept of
are very effective for the calculation of a single-spool shafting
‘‘incompatible intermediate structure’’ to treat the problem, and
system. However, when these methods are applied to calculate
the original structure is substituted by this structure approxi-
shafting systems with open-circuit and closed-circuit branches,
mately. In this paper, a modal synthesis method improved from
they are not very effective.
Hurty and Hou’s methods is presented to accommodate the vibra-
Modal synthesis method is a modeling method permitting the
tion analysis of multi-branched shafting system. It can effectively
representation of a relatively complex structure by a reduced num-
treat the complicated system in which there exists a rigid connec-
ber of degrees of freedom. The concept of substructure synthesis
tion and a flexible connection that are formed by an elastic-
coupling part. When the proposed approach is developed, the
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received November elastic-coupling part mentioned above is defined as flexible sub-
2001; revised November 2002. Associate Editor: B. Yang. structure and treated individually. Uncoordinated boundary dis-
placement of each substructure is converted into inner- 关 ⌽ 兴 s ⫽ 关兵 ⌽ 其 s1 , 兵 ⌽ 其 s2 ,¯¯ 兵 ⌽ 其 si ¯¯ 兵 ⌽ 其 sn 兴共 s⫽1,2,¯¯m 兲
deformation of flexible substructure so that the coordinative (2)
boundary condition between the flexible substructure and other
substructures could be satisfied. The other parts are naturally di- where 兵 ⌽ 其 si is the ith order column vector of dominant mode
vided into some substructures and they can be analyzed by FEM. shape of sth second-class substructure; and n is the number of
The lower-frequency normal modes of substructures are retained degrees of freedom of sth substructure.
and the higher-frequency normal modes are neglected by a fre- The dominant mass matrix is obtained according to the orthogonal
quency truncation criterion. The lower-frequency normal modes condition
are considered to be the assumed modes of Rayleigh-Ritz analysis
of whole structure. As an example of application of this method, 关 M̄ 兴 s ⫽ 关 ⌽ 兴 sT 关 M 兴 s 关 ⌽ 兴 s ⫽diag关 M s1 ,M s2 ¯M sn 兴 (3)
the analysis of torsional vibration of a cam-type engine shafting where M si
is the ith dominant mass in dominant mass matrix.
system is carried out both numerically and experimentally.
The normalized modal matrix from 共2兲 and 共3兲 is
2 Modal Synthesis Method

Many structures used in engineering are very complicated in
which there are not only rigid connections but also some kinds of
关 ⌽̄ 兴 s ⫽ 冋兵 ⌽ 其 s1 兵 ⌽ 其 s2
冑M s 冑M s
1
, 2
,¯¯
兵 ⌽ 其 si
冑M s i
,¯¯
兵 ⌽ 其 sn
冑M sn 册共 s⫽1,2,¯¯m 兲
(4)
flexible connections which exist in-between adjacent substruc-
tures. The torsional vibration model of a multi-branched shafting According to vibration theory, the relationship between system
system is shown in Fig. 1. The system is divided into subsystems coordinates 兵 Q s 其 and normal coordinates 兵 P s 其 is
in the following principles:
兵 Q s 其 ⫽ 关 ⌽̄ 兴 s 兵 P s 其 (5)
1. Main shaft 共i.e., a兲 and its direct branches are considered as
By substituting Eq. 共5兲 into 共1兲 and multiplying the left hand side
first-class substructure 共i.e., b, c, and d兲.
2. The branches belonging to first-class substructure are con- by 关 ⌽̄ 兴 sT , the following is obtained.
sidered as second-class substructure 共a consists of second-
关 ⌽̄ 兴 sT 关 M 兴 s 关 ⌽̄ 兴 s 兵 P̈ s 其 ⫹ 关 ⌽̄ 兴 sT 关 K 兴 s 关 ⌽̄ 兴 s 兵 P s 其 ⫽ 兵 0 其 (6)
class substructures 1,2,¯¯ ,m).
3. The elastic-coupling part should be taken out and defined as According to the orthogonal condition, we have
flexible substructure 共i.e., e1 , e2 ).
关 ⌽̄ 兴 sT 关 M 兴 s 关 ⌽̄ 兴 s ⫽ 关 I 兴 ; 关 ⌽̄ 兴 sT 关 K 兴 s 关 ⌽̄ 兴 s ⫽ 关 ⌳ 兴
There are two kinds of ways of connection between substruc-
tures, shown in Fig. 1. One is a rigid connection, e.g., the connec- where 关 I 兴 is unit matrix; 关⌳兴 is a diagonal matrix whose diagonal
tion between second-class substructures 1,2,¯¯ ,m. Another is a element is the square of each natural circular frequency, that is to
flexible connection, e.g., substructures a and b are connected by e1 say, 关 ⌳ 兴 ⫽diag关␻s2兴.
and substructures c and d are connected by e2 . According to the above equations, Eq. 共6兲 becomes
2.1 Modal Synthesis Between Second-Class Substructures. 关 I 兴兵 P̈ s 其 ⫹diag关 ␻ s2 兴兵 P s 其 ⫽ 兵 0 其 (7)
As an example, first-class substructure a consists of m second-
class substructures, shown in Fig. 1. Each of the second-class The normalized modal matrix 共4兲 is denoted as 关 x s 兴 , and it is
substructures is calculated by FEM. The free vibration equations divided into two parts.
of each second-class substructure 共with zero damping兲 are 关15兴
关 ⌽̄ 兴 s ⫽ 兵 x s 其 ⫽ 关 x 共s1 兲 ,x 共s2 兲 ,¯¯x 共si 兲 ,¯¯x 共sn 兲兴
关 M 兴 s 兵 Q̈ s 其 ⫹ 关 K 兴 s 兵 Q s 其 ⫽ 兵 0 其共 s⫽1,2,¯¯m 兲 (1)
⫽ 关 x sk ,x sa 兴共 s⫽1,2,¯¯m 兲 (8)
where 关 K 兴 s , 关 M 兴 s , 兵 Q̈ s 其 , 兵 Q s 其 respectively represent the total
stiffness matrix, total mass matrix, angular acceleration vector, where x s( i )
is ith order of normalized modal column vector of sth
and angular displacement vector of sth second-class substructure; second-class substructure; x sk represents the prior l order retained
and m is the number of second-class substructures. modes; x sa represents the truncation modes of the order from (l
The dominant mode shape of each second-class substructure from ⫹1) to n; and n is the number of degrees of freedom of sth
共1兲 is substructure.
Fig. 1 Torsional vibration model of complicated multi-branched shafting system

The diagonal matrix of squared frequency and normal coordinates
can also be divided into two parts, i.e., retained part and truncated
part respectively;
diag关 ␻ s2 兴 ⫽diag关 ␻ s,k
2
␻ s,a
2
兴 (9)
冋册
兵 P s其 ⫽ P a
s
P sk
(10)
According to frequency truncation criterion, the transformation

relationship between system coordinates 兵 Q s 其 and normal modal
coordinates 兵 P s 其 of sth substructure can be easily obtained by Fig. 2 The model of flexible substructure
substituting Eqs. 共8兲 and 共10兲 into 共5兲.
s
冋册
P sk
兵 Q s 其 ⫽ 关 x s 兴关 P s 兴 ⫽ 关 x sk x sa 兴 P a ⫽ 关 x sk 兴兵 P sk 其 (11) the other substructures. Assume that the shafting system consist of
first-order substructure a, b and flexible substructure e, and each
Substituting Eq. 共11兲 into 共1兲 and multiplying the left hand side by elastic unit can be simulated by a set of mass-spring elements. The
关 x sk 兴 T , the following equation is obtained according to the or- flexible substructure e, as shown in Fig. 2, is the combination of
thogonal condition. these mass-spring elements. For the ith element of flexible sub-
structure e, the following equation is obtained
再
关 I 兴兵 P̈ sk 其 ⫹diag关 ␻ s,k
2
兴兵 P sk 其 ⫽ 兵 0 其共 s⫽1,2,¯¯m 兲 (12)
K ai 共 q ai j ⫺q m
i
兲 ⫽⫺F ai j
According to equation 共12兲, we have m l q̈ m ⫹K ai 共 q a j ⫺q m 兲 ⫹K bi 共 q bi j ⫺q m
i i i i
兲 ⫽0 (20)
关 I 兴兵 P̈ L 其 ⫹ 关 K L 兴兵 P L 其 ⫽ 兵 0 其 (13) K bi 共 q b j ⫺q m 兲 ⫽⫺F b j
i i i
where 兵 P L 其 ⫽ 关 p k1 ,p k2 ¯ ,p m 兴 ; 关 K L 兴 ⫽diag关diag关␻21,k 兴 ,diag关␻22,k 兴 ,

k T
and in matrix form: 关 M 兴 e 兵 Q̈ 其 e ⫹ 关 K 兴 e 兵 Q 其 e ⫽ 兵 F 其 e (21)
¯diag关␻m,k兴]. 兵 P L 其 is the matrix of normal modal coordinates of
2
where
m second-class substructures where some coordinates are inde-
pendent of each other. Using boundary connection condition, the 兵 Q 其 e ⫽ 兵兵 q a j 其 T , 兵 q m 其 T , 兵 q b j 其 T 其 T
UN-independent coordinates can be eliminated
is displacement vector of elastic coupling unit;
兵 Q sb 其 ⫽ 关 x sb 兴兵 P s 其共 s⫽1,2,¯¯m 兲 (14)
兵 q a j 其 ⫽ 兵 q a1 j ,q a2 j ¯ ,q ai j ¯ ,q an j 其 T
where 兵 Q sb 其 is boundary displacement vector; 关 x sb 兴 is the matrix
composed of those rows corresponding to boundary connection is the boundary displacement of substructure a;
points in matrix 关 x s 兴 .
The coordinated condition of boundary displacement is 兵 q b j 其 ⫽ 兵 q b1 j ,q b2 j ¯ ,q bi j ,¯ ,q bn j 其 T
b b b b is the boundary displacement of substructure b;
关 x 11兴兵 P 1其 ⫽ 关 x 22兴兵 P 2其 ; 关 x 22兴兵 P 2其 ⫽ 关 x 33兴兵 P 3其
(15)
b b 兵 q m 其 ⫽ 兵 q m1 ,q m2 ,¯ ,q mi ¯ ,q mn 其 T
¯ ¯ ¯ ¯ ¯ ¯ ¯ 关 x mm 兴兵 P m 其 ⫽ 关 x 1 1 兴兵 P 1 其
is the inner displacement of elastic coupling unit;
By synthesizing Eq. 共15兲, and writing in blocked form
冋册
兵 F 其 e ⫽ 兵兵 F a j 其 T , 兵 F m 其 T , 兵 F b j 其 T 其 T
Pb
关 x b 兴兵 P L 其 ⫽ 关 x b1
P1 兵其
x b2 兴 ⫽ 0 (16) is the force column vector of elastic coupling unit;
that is 关 x b1 兴兵 P b 其 ⫹ 关 x b2 兴兵 P l 其 ⫽ 兵 0 其 (17) 兵 F a j 其 ⫽ 兵 F a1 j ,F a2 j ¯ ,F a1 j ,¯ ,F an j 其 T
where 兵 P b 其 and 兵 P l 其 respectively represent the matrix of un- is the boundary force of substructure a;
independent normal modal coordinates and the matrix of indepen-
dent normal coordinates among 兵 P 1 其 , 兵 P 2 其 .... 兵 P m 其 ; and 关 x b1 兴 is a 兵 F b j 其 ⫽ 兵 F b1 j ,F b2 j ,¯ ,F bi j ,¯F bn j 其 T
square matrix.
The following equation is given from Eq. 共17兲 is the boundary force of substructure b;
兵 P b 其 ⫽⫺ 关 x b1 兴 ⫺1 关 x b2 兴兵 P l 其 , 兵 F m 其 ⫽ 兵 F m1 ,F m2 ,¯ ,F mi ,¯ ,F mn 其 T
so 冋册冋
兵 P L其 ⫽ P ⫽
I
Pb ⫺ 关 x b1 兴 ⫺1 关 x b2 兴
I 册
兵 P l其 ⫽ 关 T 兴兵 P l其 (18)
is the inner force of elastic coupling unit.
Under free vibration condition, 兵 F m 其 ⫽ 兵 0¯0¯0 其 T , i.e., null
By substituting Eq. 共18兲 into 共13兲 and multiplying the left hand force is exerted inside flexible substructure. 关 M 兴 e and 关 K 兴 e are
the mass matrix and stiffness matrix of flexible substructure
side by 关 T 兴 T , the following is obtained.
respectively.
关 M l 兴兵 P̈ l 其 ⫹ 关 K I 兴兵 P l 其 ⫽ 兵 0 其 (19)
2.3 Modal Synthesis of Two Substructures Connected by
where 关 M I兴⫽关 T 兴关 T 兴,T
关 K l 兴 ⫽ 关 T 兴关 K L 兴关 T 兴
T
Flexible Substructure. The shafting system, as shown in Fig. 1,
Equation 共19兲 is free vibration equation of first-class substructure can be considered as two parts, i.e., a, e1 , b and c, e2 , d which are
a, from which the modal data of a can be solved. the combination of two first-class substructures connected by a
flexible substructure. The modal synthesis among a, e1 , b is used
2.2 Model of Flexible Substructure. For the modal synthe- to explain the modal synthesis method of two substructures con-
sis of substructures connected by flexible substructure, noncoor- nected by a flexible substructure. To perform vibration analysis of
dinative boundary displacement has to be converted into inner- the whole structure and calculate the response of the system under
deformation of flexible substructure to satisfy the coordinative external excitation, the vibration equation of substructures a and b
boundary condition at the connection of flexible substructure and is expressed as follows by Zhang 关14兴

关 M 兴 i 兵 Q̈ 其 i ⫹ 关 K 兴 i 兵 Q 其 i ⫽ 兵 F 其 i i⫽a,b (22) 兵 F̄ 其 ⫽ 关 T 兴 T 兵 F * 其 .
where 关 M 兴 i , 关 K 兴 i , 兵 Q̈ 其 i , 兵 Q 其 i , and 兵 F 其 i represent the mass ma- For free torsional vibration, the outer vector 兵 F̄ 其 ⫽ 兵 0 其 obviously.
trix, stiffness matrix, acceleration, displacement and force vector To solve Eq. 共30兲 and perform transformation of vibration modes
of ith substructure respectively. according to 共28兲, the natural frequency and modal shapes of the
The dominant mode matrixes of first-class substructure a, b are total system can be obtained.
given as 关 ⌽ 兴 a and 关 ⌽ 兴 b respectively by using the methods in
Section 2.1. Choosing lower-frequency modes 关 ⌽ 兴 la and 关 ⌽ 兴 lb 3 Torsional Vibration Analysis of Shafting System of
and denoting the corresponding mode coordinates as 兵 P 其 la and Cam-Type Engine
兵 P 其 lb , one has
3.1 Structure of the Shafting System and Its Calculation
兵 Q 其 a ⫽ 关 ⌽ 兴 la 兵 P 其 la 兵 Q 其 b ⫽ 关 ⌽ 兴 lb 兵 P 其 lb (23) Model. The shafting system of cam-type engine is shown in Fig.
There is transformation relationship 3. The combustion chamber 21 is in the left side. There are 5
cylinders located uniformly along the periphery around the axis of
兵Q其⫽关 T1兴兵 P其 (24) power output shaft. The gas with high temperature and high pres-
sure produced in the combustion chamber enters the cylinder 4
where 兵 Q 其 ⫽ 兵兵 Q 其 Ta , 兵 Q 其 Te , 兵 Q 其 Tb 其 T ,
冋册
when engine is running. The gas forces piston 5 to move recipro-
⌽ la 0 0 cally, whose movement impels cam-disc 6 to rotate. That is to say,
the axial force is converted into circumferential force, which
兵 P 其 ⫽ 兵兵 P 其 la
T
, 兵 Q 其 Te , 兵 P 其 lb
T T
其 , 关 Tl兴⫽ 0 I 0 drives the inner shaft 9 and rear propeller 13 to rotate around
0 0 ⌽ lb power output shaft. Simultaneously, the counter-moment of cam-
disc acting on piston assembles drives cylinder, cylinder-body,
According to Eq. 共23兲, the boundary displacement 兵 Q j 其 a and outer shaft, and the front propeller to rotate in the opposite direc-
兵 Q j 其 b between substructure a, b and flexible substructure e1 are tion. The output moment M i of inner shaft is equal to the output
given as follows moment M 0 of outer shaft in magnitude. However, they are in
冎
opposite direction. Gear 3 fixed on cylinder-body 15 drives fuel
兵 Q j 其 a ⫽ 关 ⌽ j 兴 la 兵 P 其 la
(25) pump, cooling water pump, and generator. Besides the reciprocal
兵 Q j 其 b ⫽ 关 ⌽ j 兴 lb 兵 P 其 lb movement, the piston assembles rotate around axis of engine’s
where, b ⌽ j c la and b ⌽ j c lb are composed of the rows corresponding output shaft together with the cylinder. This makes torsional vi-
to the boundary in 关 ⌽ 兴 la and 关 ⌽ 兴 lb . The coordinative boundary bration analysis of cam-type engine too difficult to be performed
conditions of substructures a, e1 , b are given by by conventional methods.
As the number of peak on cam-disc profile m⭓2 and the num-
兵 Q 其 e ⫽ 兵兵 Q j 其 Ta , 兵 Q j 其 Tb 其 T (26) ber of cylinder Z⭓(2m⫹1), the cam-type engine will be self-
and the coupling transformation relationship is balanced because the unbalanced inertial force and moment are
equal to zero. For the example under consideration, the output
兵 P其⫽关 T2兴兵q其
冋册
(27) torsional moment curve of the engine with m⫽2, Z⫽5 is shown
in Fig. 4. It is seen that there are 5 similar waveforms on the
I 0 curve, each of which could be attributed to one piston. The exci-
⌽ jla 0 tation of torsional moment is related to rotation speed of inner
where 兵 q 其 ⫽ 兵兵 P 其 la
T
, 兵 P 其 lb
T T
其 ; 关 T2兴⫽
0 ⌽ jlb
0 I
By substituting Eq. 共27兲 into 共24兲, the transformation relationship
of modal synthesis is derived
兵 Q 其 ⫽ 关 T 1 兴关 T 2 兴兵 q 其 ⫽ 关 T 兴兵 q 其 (28)
To assemble substructures a, e1 , b together in form, the following
is given.
关 M * 兴兵 Q̈ 其 ⫹ 关 K * 兴兵 Q 其 ⫽ 兵 F * 其 (29)
where
关 M * 兴 ⫽diag关关 M 兴 a , 关 M 兴 e , 关 M 兴 b 兴 ,
关 K * 兴 ⫽diag关关 K 兴 a , 关 K 兴 e , 关 K 兴 b 兴 ,
and
Fig. 3 Structure of cam-type engine shafting system
兵 F * 其 ⫽ 兵兵 F 其 Ta , 兵 F 其 Te , 兵 F 其 Tb 其 T
关 M 兴 i , 关 K 兴 i , 兵 F 其 i 共 i⫽a,b 兲 can be obtained from equation 共 22 兲 .
Substituting Eq. 共28兲 into 共29兲 and multiplying the left side by
关 T 兴 T , the following is obtained.
关 M̄ 兴兵 q̈ 其 ⫹ 关 K̄ 兴兵 q 其 ⫽ 兵 F̄ 其 (30)
where
关 M̄ 兴 ⫽ 关 T 兴 T 关 M * 兴关 T 兴 , 关 K̄ 兴 ⫽ 关 T 兴 T 关 K * 兴关 T 兴
and Fig. 4 The picture of output moment of torsion

quencies and mode shapes of first-class substructure b and
second-class substructures c, d, e, and f are calculated by using the
FEM. Second, the torsional vibration of first-class substructure a
can be obtained by modal synthesis based on the lower-frequency
modes of the second-class substructure. Last, the first-class sub-
structure a, b and flexible substructure s are synthesized to obtain
the torsional vibration of the whole system. Table 1 shows the
calculated results of natural frequencies according to different
stages of modal synthesis. It is seen that the calculated natural
frequencies of system decrease when the number of the reason-
able substructures involved increases.
Fig. 5 Calculation model of torsional vibration for cam-type To find the effect of truncation of higher-frequency modes, the
engine shafting system natural frequencies of the whole system are calculated with dif-
ferent number of modes retained. The results are shown in Table
2. It is seen that when the total number of modes is 67, the relative
shaft (n i ⫽2185 rpm) and outer shaft (n 0 ⫽2085 rpm) so that the error of natural frequencies ranges only from 0.508% to 2.10%
fundamental excitation frequency of torsional vibration can be while the number of modes involved decreases from 67 to 15.
calculated as Zhao 关21兴: That is to say, high accuracy of calculation can be reached with
less lower-frequency modes involved in mode synthesis.
f 0 ⫽mZ 共 n i ⫹n 0 兲 /60⫽2⫻5 共 2185⫹2085兲 /60⫽711.67共 Hz兲
In the past, the inner shaft and outer shaft of cam-type engine 4 Experiment
were treated separately, and only one of the two natural frequen-
cies of inner shaft and outer shaft was considered as the natural In order to validate the proposed modal synthesis method and
frequency of whole system so that the results of calculation are determine the dominant frequencies of excitation, the experiment
not realistic 关22兴. Owing to the existence of interaction, the inner has been conducted in two cases. The experimental set-up of full-
shaft and outer shaft should be considered as a whole system in size cam-type engine is shown in Fig. 6. The signal gear for tor-
vibration analysis. The whole system is modeled as two first-class sional vibration measurement is fixed on outer shaft.
substructures connected by a flexible substructure. The combina- Case 1: Natural Frequencies of Shafting System of Cam-type
tion of cylinder-body, outer shaft, front propeller and engine- Engine in Static Condition.
driven auxiliary machinery is defined as first-class substructure a. The simplest test to check the fundamental natural frequencies
The combination of cam-disc, inner shaft and rear propeller is of any vibratory system is the rap test. To check the natural fre-
defined as first-class substructure b. The piston assembles to link quencies of torsional vibration, the forces exerted on the system
inner shaft and outer shaft are treated as flexible substructure s in are different from the rap forces in commonly rap test. A circum-
order to treat the uncoordinated angular displacements. First-class ferential rap force is successfully applied at the shafting system by
substructure a is further divided into second-class substructures, a special instrument. The natural frequencies of the rap test and
i.e., the combination of cylinder-body, outer shaft and front pro- numerical calculation are listed in Table 3. It can be found that the
peller is defined as second-class substructure c, cooling water calculation results of natural frequencies are in good agreement
pump as second-class substructure d, fuel pump as second-class with the experimental results. So the modal synthesis method pro-
substructure e and generator as second-class substructure f, as posed in this paper is very effective for torsional vibration analy-
shown in Fig. 5. sis of the complicated multi-branched shafting system of the cam-
type engine.
3.2 Analysis of Torsional Vibration. Based on the model Case 2: The Dominant Excitation Frequencies of Shafting Sys-
shown in Fig. 5, the characteristics of torsional vibration for cam- tem of Cam-type Engine.
type engine shafting system can be determined by using the pro- To determine the dominant excitation frequencies, the tests of
posed method which is incorporated into computer software de- operation of cam-type engine are performed in two ways, namely
signed by the authors. Natural frequencies of torsional vibration no load test and low-load test. The dominant excitation frequen-
are the matter of our main concern. When the calculation is car- cies are measured under rotating conditions. The inner and outer
ried out, the following steps are followed. First, the natural fre- shafts are respectively run at 2185 rpm and 2085 rpm, but the two
Table 1 The natural frequencies of modal synthesis according to substructures involved „Hz…
Substructure 1 2 3 4 5 6
Substructure c 404.825 5431.014 6296.115 10298.600 12164.090 14839.410
Substructure a 370.310 1220.826 6124.352 6675.603 10898.827 12194.780
Substructure b 400.180 3577.616 5139.323 5981.029 6835.063 9408.980
whole system 198.010 371.843 453.629 1220.560 3576.992 5137.702
Table 2 Natural frequency and error with selected different number of modes retained „Hz…
Retained No. of modes 1 2 3 4 5 6

Total number 67 198.010 371.843 453.629 1220.560 3576.992 5137.702
Retained number 55 198.014 371.848 453.645 1220.748 3577.998 5138.896
Error 共%兲 0.002 0.001 0.003 0.015 0.028 0.023
Retained number 45 198.101 372.222 454.114 1224.303 3583.242 5142.056
Error 共%兲 0.046 0.102 0.107 0.306 0.175 0.085
Retained number 35 198.175 372.410 454.983 1228.936 3588.367 5150.724
Error 共%兲 0.083 0.152 0.298 0.686 0.318 0.253
Retained number 25 198.463 373.589 455.978 1234.871 3597.832 5160.411
Error 共%兲 0.229 0.470 0.518 1.172 0.583 0.442
Retained number 15 199.016 374.216 456.814 1246.213 3604.314 5166.865
Error 共%兲 0.508 0.638 0.702 2.10 0.764 0.568

Fig. 6 Experimental set-up of cam-type engine
shafts run in opposite directions. Results for the tests are shown in 5 Conclusion
Fig. 7. It is seen that the component of 34.75 Hz is predominant in
A modal synthesis method is proposed to analyze the torsional
Fig. 7, which is corresponding to the rotating speed of engine
vibration of multi-branched shafting system where rigid coupling
outer shaft where signal gear is fixed on. Many frequencies cor-
and flexible connection exist. When the proposed approach is de-
responding to the higher-frequency peaks are the integral times of
veloped, a new concept of connecting substructure, namely flex-
34.75 Hz. This is caused by the excitation of unbalanced moment
ible substructure, is presented. It is used to tackle the elastic-
on outer shaft. Another predominant peak is at about 710 Hz,
coupling part of multi-branched shafting system. The other parts
which is in accordance with the fundamental excitation frequency
of the system are partitioned into some substructures that are ana-
of torsional vibration mentioned above. It is caused by the pulsat-
lyzed by FEM. The lower-frequency normal modes of substruc-
ing output moment of torsion 共pulsation frequency is f
tures are retained while the higher-frequency normal modes are
⫽711.67 Hz). The reason for the phenomenon is that each cylin-
neglected. The lower-frequency normal modes are regarded as the
der’s work of cam-type engine is uneven and that the moment of
assumed modes of Rayleigh-Ritz analysis of whole structure. The
torsion amplitude acted on cam-disc is different.
proposed approach is used to analyze the torsional vibration of
cam-type engine shafting system. We can draw the following
conclusions:
Table 3 Natural Frequencies of shafting system „Hz…
1. The concept of flexible substructure and the improved modal
1 2 synthesis method presented in this paper are of value for torsional
vibration analysis of complicated system where uncoordinated
Calculation 198.010 371.843 boundary conditions exist.
Experiment 195.3 369.6
Error 共%兲 1.39 0.61 2. For the shafting system of cam-type engine, its fundamental
excitation frequency of torsional vibration 710 Hz is just in-
between the 3rd natural frequency 453.629 Hz and 4th natural
frequency 1220.56 Hz. Furthermore, the dominant excitation fre-
quencies of the cam-type engine are not coincided with natural
frequencies of the shafting system so that there is no resonance
induced. In consequence, torsion vibration of the cam-type shaft-
ing system is not harmful during its normal operation.
3. The cam-type engine shafting system that contains inner and
outer shaft assembles should be considered as a whole system as
the natural frequency decreases with the increasing of the number
of substructures involved in modal synthesis. Using the proposed
method, the inner and outer shaft assembles are analyzed simul-
taneously and the torsional vibration of the whole system is
obtained.
4. The numerical results of natural frequencies are in good
agreement with the experimental results. The reasonable accuracy
of natural frequency can be reached by considering less number of
lower-frequency modes in modal synthesis calculation.
References
关1兴 Den Hartog, J. P., and Li, J. P., 1964, ‘‘Forced Torsional Vibration with Damp-
ing: An Extension of Holzer’s Method,’’ ASME J. Appl. Mech., 31, pp. 276 –
280.
关2兴 Pestel, E. C., and Leckie, F. A., 1963, Matrix Method in Elasto Mechanics,
McGraw-Hill, New York.
关3兴 Sankar, S., 1979, ‘‘On The Torsional Vibration of Branches System Using
Extended Transfer Matrix Method,’’ ASME J. Eng. Ind., Series B, 101, pp.
546 –553.
关4兴 Dawson, B., and Davies, M., 1974, ‘‘An Improved Transfer Matrix Proce-
dure,’’ Int. J. Numer. Methods Eng., 8, pp. 111–117.
Fig. 7 The frequency spectrum of torsional vibration 关5兴 Huang, Y. M., and Horng, C. D., 1999, ‘‘Analysis of Torsional Vibration Sys-

tems by the Extended Transfer Matrix Method,’’ ASME J. Vibr. Acoust., 121, System of Contains Stiffness Coupling Component,’’ Journal of Engineering
pp. 250–255. Mechanics, 7, pp. 93–101 共in Chinese兲.
关6兴 Firoozian, R., and Stanway, R., 1989, ‘‘Design and Application of A Finite 关15兴 Xu, K. Q., 1989, ‘‘Non-coordination Dynamic Substructure Synthesis
Element Package for Modelling Turbomachinery Vibrations,’’ J. Sound Vib., Method,’’ Journal of Vibration and Shock, 31, pp. 64 – 67 共in Chinese兲.
134, pp. 115–137. 关16兴 Rubin, S., 1975, ‘‘Improved Component-Mode Representation For Structural
关7兴 Li, H. Z., 1991, ‘‘Crankshaft Torsional Vibration Calculation by Finite Ele- Dynamic Analysis,’’ AIAA J., 13, pp. 995–1006.
ment Method,’’ Journal of Internal Combustion Engines, 9, pp. 157–162 共in 关17兴 Inamura, T., Suzuki, H., and Sata, T., 1994, ‘‘An Improved Method of Dy-
Chinese兲. namic Coupling in Structural Analysis and Its Application,’’ ASME J. Dyn.
关8兴 Hurty, W. C., 1960, ‘‘Vibration of Structural System by Component Mode Syst., Meas., Control, 106, pp. 82– 89.
Synthesis,’’ J. Eng. Mech. Div., ASCE, 86, pp. 51– 69. 关18兴 Gaganis, B. J., 1999, ‘‘Modal Analysis of Rotor on Piecewise Linear Journal
关9兴 Hurty, W. C., 1965, ‘‘Dynamic Analysis of Structural System Using Compo- Bearings Under Seismic Excitation,’’ ASME J. Vibr. Acoust., 121, pp. 190–
nent Modes,’’ AIAA J., 3, pp. 678 – 685. 196.
关10兴 Craig, Jr., R. R., and Bampton, M. C. C., 1968, ‘‘Coupling of Substructures for 关19兴 MacNeal, R. H., 1971, ‘‘A Hybrid Method of Component Mode Synthesis,’’
Dynamic Analysis,’’ AIAA J., 6, pp. 1313–1319. Comput. Struct., 1, pp. 581– 601.
关11兴 Hou, S. N., 1969, ‘‘Review of Modal Synthesis Techniques and A New Ap- 关20兴 Hale, A. L., and Meirovitch, L., 1982, ‘‘A Procedure for Improving Discrete
proach,’’ The Shock and Vibration Bulletin, 40, pp. 25–39. Substructure Representation in Dynamic Synthesis,’’ AIAA J., 20, pp. 1128 –
关12兴 Goldman, R. L., 1969, ‘‘Vibration Analysis by Dynamic Partitioning,’’ AIAA 1136.
J., 7, pp. 1152–1154. 关21兴 Zhao, L. F., 1991, The Principle of Torpedo Piston Engine, Northwestern
关13兴 Dowell, E. H., 1972, ‘‘Free Vibration of an Arbitrary Structure in Terms of Polytechnic University Press, Xi’an, China, pp. 258 –269 共in Chinese兲.
Component Modes,’’ ASME J. Appl. Mech., 39, pp. 727–732. 关22兴 Zou, C. P., 1994, ‘‘FEM Analysis of Later Vibration of Torpedo Propulsion
关14兴 Zhang, H. T., 1990, ‘‘Free-Interface Mode Synthesis Method for Vibration System,’’ Journal of Torpedo Technology, 2, pp. 22–32 共in Chinese兲.

Study on the Dynamics of a Rotor
Fusheng Lin in a Maneuvering Aircraft
Guang Meng
e-mail: [email protected] This paper shows how the dynamics of a rotor in a maneuvering aircraft changes accord-
ing to the operation of the aircraft. The mathematical model of an unbalanced rotor
State Key Lab of Vibration, system located in the maneuvering aircraft is derived. The dynamic characteristics of the
Shock & Noise, rotor running at a constant angular speed or a constant acceleration are studied under
Shanghai Jiao Tong University, the assumptions that the aircraft maneuvers only in a vertical plane and that the pitching
Shanghai 200030, P. R. C. angle and the flight path inclination of the aircraft are equal. The effects of gravity and
unbalance parameter are considered. The results show that the unbalanced response of a
rotor in an aircraft is obviously influenced by the aircraft’s flying status.
关DOI: 10.1115/1.1576422兴
1 Introduction the same vertical plane with OXZ and O 1 X 2 Z 2 , and the aircraft
In the previous literatures on rotor dynamics, the supports of moves only in the vertical plane. A Jeffcott rotor model is used.
the rotor are assumed to be static and in the same horizontal or The displacements of the disk center are all measured in a body-
vertical direction 关1–2兴. This is not practical for rotors located in fixed and rotating O 1 ␰ ␩ ␨ reference frame, where ␰ and ␩ are in
aircraft because the aircraft operates under various conditions dur- the principal directions of the cross-section of the shaft and ␨
ing maneuvering. Many researchers are focused on studies of only coincides with the Y 1 axis, as shown in Fig. 2. The flexibility of
the maneuvering flight of aircraft 关3–7兴, but there is no literature the rotor is r. The temporary position of the disk center is O 3 . The
analyzed the dynamics of a rotor running in an operating aircraft. position of the mass center of the disk is C. The angle between the
This paper attempts to determine the dynamic characteristics of a axes O 1 ␩ and O 1 Y 1 is ␤. The coordinate O 1 ␰ ␩ ␨ rotates with
rotor and to simulate, more realistically, the response of such a
respect to O 1 ␨ (O 1 Z 1 ) with an angular speed ␤˙ . It is noted that e 0
rotor in a maneuvering aircraft.
is a reference eccentricity, m is the equivalent mass of the heavy
2 Equations of Motion disk mounted at the mid-span of a massless elastic shaft, e is the
eccentricity of the disk, ␪ is the angle between the orientation of
In Fig. 1, OXYZ is a space-fixed and stationary coordinate sys-
tem, and O 1 X 2 Y 2 Z 2 is an aircraft-carried coordinate system par- the eccentricity and the ␩ 1 axis, k is the stiffness coefficient of the
allel to OXYZ. It is supposed that the pitching angle and the flight shaft, and c is the external viscous damping coefficient of the shaft
path inclination of the aircraft are equal and defined as ␣, i.e., the and disk.
axis of the rotor shaft O 1 Z 1 is in the same direction as the longi- The dimensionless differential equations of motion of the rotor
tudinal axis of the fuselage. It is also assumed that O 1 X 1 Z 1 is in system can be written as follows:
冋册冉冊
冦
2␧ 1 1 2␧ 1
␰ ⬙d ⫹ ␤ ⬙ 共 ␩ d ⫹U cos ␪ 兲 ⫹ ␰ ⬘d ⫹ 2 ⫺ ␤ ⬘ 2 ⫺ ␣ ⬘ 2 共 1⫹cos 2 ␤ 兲 ␰ d ⫹2 ␤ ⬘ ␩ ⬘d ⫹ ␤ ⬘ ⫺ ␣ ⬘ 2 sin 2 ␤ ␩ d ⫺U ␤ ⬘ 2 sin ␪
⍀ ⍀ ⍀
冋册冋册
2 2
Ge 2␧
⫹ 2 cos ␣ ⫺U ␣ ⬘ sin共 ␤ ⫹ ␪ 兲 cos ␶ ⫹
2
共 x ⬘ cos ␣ ⫺z ⬘d sin ␣ 兲 ⫹x ⬙d cos ␣ ⫺z ⬙d sin ␣ cos ␶ ⫽0
⍀ ⍀ d
2␧
⍀ 冋 1
⍀
1
册
␩ d⬙ ⫺ ␤ ⬙ 共 ␰ d ⫹U sin ␪ 兲 ⫹ ␩ d⬘ ⫹ 2 ⫺ ␤ ⬘ 2 ⫺ ␣ ⬘ 2 共 1⫺cos 2 ␤ 兲 ␩ d ⫺2 ␤ ⬘ ␰ d⬘ ⫺ 冉
2␧
⍀
1
冊
␤ ⬘ ⫹ ␣ ⬘ 2 sin 2 ␤ ␰ d ⫺U ␤ ⬘ 2 cos ␪
(1)
冋册冋册
2 2
Ge 2␧
⫹ cos ␣ ⫺U ␣ ⬘ 2 sin共 ␤ ⫹ ␪ 兲 sin ␶ ⫹ 共 x ⬘ cos ␣ ⫺z ⬘d sin ␣ 兲 ⫹x ⬙d cos ␣ ⫺z ⬙d sin ␣ sin ␶ ⫽0
⍀2 ⍀ d
where,
␰ d ⫽ ␰ /e o , ␩ d ⫽ ␩ /e o , ␻ c ⫽ 冑k/m, ␧⫽c/2m ␻ c , ⍀⫽ ␻ / ␻ c ,
G e ⫽g/ ␻ 2c e o , U⫽e/e o , x d ⫽x/e o , z d ⫽z/e o . ␧ is the dimension-
tg ␣ ⫽
dx dx dz
⫽ •
dz dt dt 冉冊 ⫺1
ẋ x ⬘ x d⬘
⫽ ⫽ ⫽ ,
ż z ⬘ z ⬘d
less viscous damping coefficient, G e is the gravity parameter, U is

the unbalance parameter, and ⬘ denotes differentiation with respect
␣ ⫽arctg 冉冊x ⬘d
z d⬘
.
to dimensionless time ␶. Accordingly, ␣ ⬘ can be obtained.
If the movement of the aircraft is given, i.e., x( ␶ ) and z( ␶ ) are The numerical response of the system in O 1 ␰ ␩ ␨ coordinate can
given, then the angle ␣ can be calculated by be obtained by the Runge-Kutta integration method. Response in
O 1 X 1 Y 1 Z 1 coordinate can also be obtained through a coordinate
transformation.
Contributed by the Technical Committee on Vibration and Sound for publication Equation 共1兲 describes an unbalanced rotor system located in an
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received August 2002; aircraft moving in the vertical plane. Assuming that the rotating
Revised manuscript received March 2003. Associate editor: G. Flowers. motion of the rotor, i.e., ␤ ⬙ and ␤ ⬘ can be obtained, Eq. 共1兲 can be
Fig. 3 Influence of accelerating abruptly in vertical direction
„constant horizontal velocity component… „ U Ä1, G e Ä1, ⍀
Fig. 1 Sketch of the rotor system located in an aircraft and the
Ä0.8, z ⬙d Ä0, z ⬘d Ä z ⬘d 0 Ä100, x ⬘d 0 Ä100…
space-fixed stationary coordinates
simplified and then solved. When the rotor rotates at a constant sionalized as the ratio of the actual amplitude of the disk center to
the reference eccentricity e 0 . In each figure, the amplitude is
speed ⍀, then ␤ ⬙ ⫽0 and ␤ ⬘ ⫽⍀ 0 ⫽⍀. If the rotor accelerates at
shown in the form of the natural logarithm with base 2.71828 of
a constant rate of acceleration C a , then ␤ ⬙ ⫽C a ⬎0 and ␤ ⬘
⫽⍀ 0 ⫹C a ␶ , where ⍀ 0 is the initial angular speed of the rotor. the dimensionless amplitude r e ⫽ 冑␰ 2d ⫹ ␩ 2d /e 0 . The velocity and
the acceleration are assumed to change suddenly at the beginning
of the aircraft’s action. In all cases, ␧⫽0.08. Data was taken for
3 Numerical Analysis 50 revolutions after the response became stable, following this,
When the aircraft flies at a constant rate of acceleration, the the status of the aircraft changes.
dimensionless x ⬘d0 and z ⬘d0 , x ⬘d and z ⬘d , and x d⬙ and z d⬙ are assumed
to be the initial velocity components, the velocity components, 3.1 Constant Rotor Rotating Speed
and the acceleration components in the directions of the X and the 3.1.1 Influence of the Aircraft Velocity on Rotor Response
Z axes, respectively. Then z d⬘ ⫽z d0 ⬘ ⫹z d⬙ ␶ and x d⬘ ⫽x d0
⬘ ⫹x d⬙ ␶ . If the The numerical results show that the steady-state amplitudes for
flight path of an aircraft is a sine curve in the vertical plane OXZ different horizontal velocities are the same, which means that
and the horizontal component of the aircraft velocity is constant, changing the horizontal velocity z d⬘ has no influence on the re-
the motion can be written as x d ⫽x d0 ⫹⌫ sin(␭zd⬘␶). Thereby the sponse amplitude. When the aircraft climbs at a constant velocity
acceleration components of the aircraft can be obtained. When the and the vertical velocity component is set to different values, the
rotor rotates at a constant speed, the responses at three different amplitude varies with x d⬘ . Larger x d⬘ corresponds to smaller am-
speed ratios (⍀⫽0.8,0.3,1.6) are calculated. When the rotor ro- plitude. When ⍀ changes, the difference between the amplitudes
tates at a constant acceleration, two cases of the initial speed ratio in different vertical velocities changes also. The influence of the
(⍀ 0 ⫽0.8,0.3) are investigated. vertical velocity component of the aircraft when ⍀⫽0.3 is weaker
The numerical responses of some cases are shown from Fig. 3 than when ⍀⫽0.8.
to Fig. 7. In these figures the amplitude refers to the maximum
amplitude of the disk center in every revolution and is nondimen- 3.1.2 Influence of the Aircraft Acceleration on Rotor Response
When ⍀⫽0.8, the amplitude of the disk increases suddenly when
the aircraft accelerates abruptly in the vertical direction but the
horizontal velocity component remains unchanged. The amplitude
decreases with time until reaching a steady value 共Fig. 3兲. The
steady-state amplitude is smaller than the value when x ⬙d ⫽0. The
increasing amount of response amplitude depends on the value of
x d⬙ when the aircraft accelerates in the vertical direction. Larger
values of x d⬙ correspond to larger amplitudes. The difference be-
tween the steady-state amplitudes for different values of x ⬙d is
small. Changing G e hardly affects the steady-state amplitude
when the aircraft is accelerating. However, a larger value of the
parameter U will cause larger steady-state amplitude.
When ⍀ is small, the increase in amplitude is less and the time
needed to reach steady state is shorter.
3.1.3 Rotor Response When Aircraft Flies in a Sine Curve in
the Vertical Plane OXZ. It is supposed that x d ⫽x d0
⫹⌫ sin(␭z⬘d␶) and the horizontal velocity of the aircraft is unal-
tered, i.e., z ⬙d ⫽0 and z d⬘ ⫽z ⬘d0 ⫽constant. In this case, ␭z ⬘d denotes
the period of the motion. The maneuvering range in the vertical
direction is reflected by ⌫. The following two cases are calculated.
Fig. 2 The reference frame O 1 ␰ ␩ ␨ and parallel frame O 3 ␰ 1 ␩ 1 ␨ 1 共1兲 ⌫ changes while ␭ and z d⬘ are invariant; 共2兲 z ⬘d changes while ⌫

Fig. 4 Rotor responses when the aircraft flies in a sine curve „⍀Ä0.8, U Ä0.5, G e Ä0.5, z d⬙ Ä0, ␭Ä0.00002… „a… Influence
of ⌫ „ z ⬘d Ä100… „b… Influence of z ⬘d „⌫Ä50000…
and ␭ are unaltered. The response when the aircraft moves only in When ⍀ becomes smaller, the increased amplitude due to the
the horizontal direction and ⌫⫽0 is taken as the basis for com- aircraft’s maneuver is less and the fluctuation is reduced.
parison. The aircraft first flies in a horizontal direction at a con-
stant velocity. When the free vibration vanishes away and the
steady state amplitude is reached, data of 50 revolutions are taken. 3.2 Constant Angular Acceleration of the Rotor
Afterwards, the aircraft maneuvers for several periods. The corre-
3.2.1 Influence of the Aircraft Velocity on Rotor Response
sponding period in dimensionless time is 1200⫻2 ␲ .
If the aircraft maneuvers periodically, its velocity and accelera- ( ␤ ⬙ ⫽0.00025). If the aircraft is static, the steady state unbal-
tion will also vary periodically. The amplitude response of the ance response of the rotor at constant speed acceleration appears
disk center changes periodically except at the beginning of aircraft to fluctuate rapidly after the critical rotating speed. When x ⬘d ⬎0
action. In Fig. 4共a兲 the amplitude decreases suddenly at the very and is invariable, different z ⬘d lead to obviously different response
beginning of the aircraft maneuver. The vibration response is pe- at the beginning of the rotor acceleration. With the rotor speed
riodic when the second period of aircraft movement begins, and increasing, the difference between responses becomes smaller and
the period is the same as that of sine curve (500⫻2 ␲ ). For most then vanishes. A smaller value of amplitude corresponds to a large
part of each period 共almost 80%兲 the amplitude is less than that x ⬘d . When z ⬘d remains constant, the response curves for different
for ⌫⫽0, but for some parts the amplitude is larger than that for x ⬘d are clearly different at the beginning of acceleration. ⍀ 0 ap-
⌫⫽0. The amplitude fluctuates acutely when the aircraft starts to pears to have no effect on the response.
maneuver. Then the fluctuation decreases. The peak amplitude
value for ⌫⫽100000 is almost 50% greater than that for ⌫⫽0, 3.2.2 Influence of the Aircraft Acceleration on Rotor Response
and is almost 26% greater than that for ⌫⫽50000 (G e ⫽0.5, U ( ␤ ⬙ ⫽0.00025). For different aircraft accelerations, the re-
⫽0.5). In Fig. 4共b兲, larger values of z d⬘ correspond to smaller sponses of the rotor at low speed range are also different, but the
aircraft maneuvering periods, relatively smaller periods of vibra- responses at high speed range may be almost the same 共Fig. 5共a兲兲
tion response and larger peak values. This indicates that quick or very different 共Fig. 5共b兲兲 depending on the values of z ⬙d0 and
maneuvering of the aircraft will cause the vibration response of z ⬙d . G e has little influence on the response curve, but larger values
the rotor system to change dramatically, although the maneuver of U correspond to larger amplitude. The responses for ⍀ 0 ⫽0.3
range in the vertical plane is unchanged. are like those for ⍀ 0 ⫽0.8.
Fig. 5 Influence of the aircraft acceleration in horizontal and vertical directions on rotor accelerating response „⍀ 0
Ä0.8, x ⬙d Ä0, 10, 50, x ⬘d Ä z ⬘d Ä100, G e Ä1, U Ä1.0… „a… z ⬙d 0 Ä z ⬙d Ä0 „b… z ⬙d 0 Ä z ⬙d Ä100

Fig. 6 Rotor accelerating response when the aircraft flies in a sine curve „⍀ 0 Ä0.8, U Ä0.5, G e Ä1.5, z ⬙d Ä0… „a… Influ-
ence of ⌫ „ z ⬘d Ä100, ␭Ä0.00002… „b… Influence of z ⬘d „⌫Ä100, ␭ z ⬘d Ä0.002…
3.2.3 Rotor Response When Aircraft Flies in a Sine Curve in the rotor running at a constant speed increases suddenly at the
Vertical Plane OXZ ( ␤ ⬙ ⫽0.00025). When the aircraft flies in a very beginning of aircraft acceleration and eventually becomes
sine curve and ⍀ 0 ⫽0.8, the rotor response amplitude at the be- stable, except when the aircraft moves and accelerates only in the
ginning of the rotor acceleration decreases and the response curve horizontal direction. When the rotor rotates at a constant accelera-
fluctuates. The response is not influenced by the parameters of the tion, the amplitude fluctuates before the critical speed. Different
sine curve when the rotor speed is near or at the critical speed. aircraft acceleration lead to much different rotor responses when
When the rotor speed is much greater than the critical speed, the the rotor runs at a constant rotating speed or at a constant accel-
response curve is significantly influenced by aircraft maneuvering eration.
and fluctuates slowly 共Fig. 6兲. The fluctuation at the beginning of 2兲 The response of the rotor rotating at a constant speed is
rotor acceleration and after the critical speed is reduced if U in- clearly influenced by the amplitude and the period of the flight
creases or G e decreases. path when the aircraft moves in a sine curve in the vertical plane.
When the initial rotating speed is small (⍀ 0 ⫽0.3), the re- The response curve appears to be periodic except at the beginning
sponses are different for different values of ⌫ 共Fig. 7兲. Before of the aircraft action, the period is the same as that of the sine
reaching the critical speed, the response fluctuates slowly. If ⌫ curve. The gravity parameter affects the response significantly,
⫽0, the response has many peak values equal to the value on the especially the trend of the response curve. The unbalance param-
⌫⫽0 curve, this means that the amplitude of the rotor with con- eter affects only the amplitude value.
stant acceleration is less than that of the rotor with lower rotating 3兲 When the aircraft flies in a sine curve in the vertical plane,
speed and no aircraft maneuver. Aircraft maneuvering has no in- the response of the rotor running at a constant acceleration is
fluence on the response peak amplitude. different from that when the aircraft has no maneuvering, mainly
at the beginning of the acceleration and after the rotor speed
4 Conclusions passes through the critical speed. The response curve is obviously
1兲 The operation of an aircraft affects the dynamics of the rotor influenced by the gravity parameter when the initial speed is low
located in it. When the aircraft maneuvers at a constant accelera- in the case of acceleration.
tion in the vertical and/or horizontal directions, the amplitude of
Acknowledgment
The support from China ‘‘863’’ Project 共No. 2002AA412410兲
and The Doctor Subjects Research Founding of China University
共No. 20020248053兲 is gratefully acknowledged.
References
关1兴 Lee, A. C., Kang, Y., Tsai, K. L., and Hsiao, K. M., 1992, ‘‘Transient Analysis
of an Asymmetric Rotor-Bearing System during Acceleration,’’ ASME J. Ind.,
114共4兲, pp. 465– 475.
关2兴 Ganesan, R., and Sankar, T. S., 1993, ‘‘Resonant Oscillations and Stability of
Asymmetric Rotors,’’ Proc. of the 14th Biennial ASME Conference on Me-
chanical Vibration and Noise, ASME DE, 56, pp. 19–22.
关3兴 Spence, A. M., and Cele, R., 1995, ‘‘Coupled Rotor Fuselage Dynamics and
Aero-Elasticity in Turning Flight,’’ J. Am. Helicopter Soc., 40共1兲, pp. 47–58.
关4兴 Cao, Y., 1999, ‘‘Modelling the Unsteady Aerodynamic Forces of a Maneuver-
ing Rotor,’’ Aircraft Engineering and Aerospace Technol., 71共5兲, pp. 444 – 450.
关5兴 Bagai, A., Leishman, J. G., and Park, J., 1999, ‘‘Aerodynamic Analysis of a
Helicopter in Steady Maneuvering Flight Using a Free-Vortex Rotor Wake
Model,’’ J. Am. Helicopter Soc., 44共2兲, pp. 109–120.
关6兴 Park, J. S., and Leishman, J. G., 1999, ‘‘Investigation of Unsteady Aerody-
namics on Rotor Wake Effects in Maneuvering Flight,’’ Annual Forum
Proceedings-American Helicopter Society, 1, pp. 467– 480.
Fig. 7 Rotor accelerating response when the aircraft flies in a 关7兴 Krothapalli, K. R., Prasad, J. V. R., and Peters, D. A., 2001, ‘‘Helicopter Rotor
sine curve „⍀ 0 Ä0.3… „ U Ä0.5, G e Ä1.5, z ⬘d Ä100, z ⬙d Ä0, ␭ Dynamic Inflow Modelling for Maneuvering Flight,’’ J. Am. Helicopter Soc.,
Ä0.00002, ⌫Ä0, 50000, 100000… 46共2兲, pp. 129–139.

Control of Self-Excited Vibration
Jinhao Qiu of a Rotor System With Active Gas
Junji Tani
Bearings
Taekyu Kwon This paper presents an experimental study on the active control of self-excited vibrations
in a rotor-bearing system supported on a pair of externally pressurized thrust bearings in
Institute of Fluid Science, Tohoku University, the axial direction and on actively controlled journal gas bearings in the radial direction.
Katahira 2-1-1, Aoba-ku, Sendai 980-8577, The active journal gas bearings used are of the tilting-pad type with one traditional
Japan passive pad and two active pads with embedded piezoelectric actuators. Feedback control
systems are constructed with gap sensors for measuring the vibration of the rotor, embed-
ded piezoelectric actuators, and PID controllers. The experimental results show that the
self-excited vibration can be effectively suppressed with the designed feedback control
system, if the gains of the PID controllers are properly tuned. 关DOI: 10.1115/1.1576423兴
1 Introduction bearing system using the control force generated by the two active
journal gas bearings. The effectiveness of the active pads was
Since gas bearings use gas as a lubricant, they can work in a
confirmed by experimental results.
wide range of temperatures and have been widely used in turbo
compressors and expanders of helium liquefiers 关1,2,3兴. One of
the main problems with gas bearings is their intrinsic fluid-related 2 Configuration and Vibration Characteristics of the
hydrodynamic instability as the rotor exceeds twice the natural Rotor-Bearing System
frequency of the rotor rigid body mode. The instability is due to
the interaction between the journal and the gas used as the lubri- Figure 1 shows the configuration of the rotor-bearing system,
cant. The stability of the gas bearings has been greatly improved originally designed for use in a high-expansion-ratio turbo ex-
with better designs such as spiral grooves, tilting pads and elastic pander of a helium liquefier, shown in Fig. 2. The rotor consists of
foils. In these cases, rotors supported by gas bearings can reach a shaft, a turbine impeller at one end of the shaft for the genera-
very high rotational speeds 关1,4兴. However, self-excited vibrations tion of the driving moment, and a blower impeller for braking at
in rotor systems supported by gas bearings have always been an the other end. The rotor of the expander is supported by two
important issue, and a further enhancement in stability is required tilting-pad journal gas bearings in the radial direction and a pair of
in practical applications. externally pressurized thrust gas bearings in the axial direction.
In some cases, a modification of the configuration or some pa- The rotor can reach a maximum speed of 35,000 rpm at a driving
rameters of the design can prevent gas bearings from undergoing air pressure of 0.5 MPa on the turbine impeller.
self-excited vibration 关5,6兴. For example, Ihara 关6兴 designed a The rotor is installed vertically and its weight is supported by
movable pivot with an integrated spring, which can adjust the the thrust bearings. Since the load in the radial direction is very
pressing force between the pad and the journal, to increase the small, tilting-pad journal gas bearings are used. The total length of
damping effect of the bearing system. The new pivot had the the rotor is 181.35 mm and the two tilting bearings are located at
ability to suppress the self-excitation of the rotor system, but the 23.54 mm above and 26.46 mm below the center of mass of the
friction between the pads and shaft also prevented the shaft from rotor, as shown in Fig. 2共a兲. The cross-sectional view of the
starting rotation when the torque was too small. tilting-pad journal gas bearing used in the rotor-bearing system is
Recently, the active control of the self-excited vibration in gas shown in Fig. 2共b兲 and Fig. 3. There are three tilting pads in each
bearings has begun to attract attention, due to the increasing de- bearing. The width of a pad is 15.4 mm and its spanning angle ␤
mands for high rotation speed and stability 关7–11兴. Recent ad- is 110 deg. The nominal clearance C r is 20 ␮m and the preload of
vances in sensor and actuator technology and electronics have the pads is 0.5. The parameters of the rotor-bearing system are
also made the active control of gas bearings possible. Many stud- summarized in Table 1. The dynamic characteristics of the tilting-
ies on the active control of externally pressurized bearings have pad gas bearings are discussed in 关4兴.
also been reported, but no research has been reported on the active The externally-pressurized thrust bearing was used to support
control of tilting-pad journal gas bearings, probably due to their the high thrust load resulting from the impeller. The thrust bear-
complicated configuration and vibration mechanism. ings are a pair of the inherently compensated annular bearings
In this study, an active pad, supported by a pivot with an em- shown in Fig. 4. Their outer and inner diameters are 32 mm and
bedded piezoelectric actuator, is proposed for active tilting-pad 16.4 mm, respectively. There are 18 nozzles uniformly distributed
journal gas bearings and used in the active control of the self- in a circle of r a ⫽22.4 mm in each thrust bearing. The diameters
excited vibration of a rotor-bearing system. The rotor is supported of the upper and lower bearings are 0.3 mm and 0.6 mm, respec-
on a pair of externally-pressurized thrust bearings 共EPTB兲 in the tively. The parameters of the thrust gas bearings are summarized
axial direction, and on active journal gas bearings in the radial in Table 2. The design and dynamic characteristics of the
direction. The radial position of the active pads in the journal externally-pressurized thrust bearings are discussed in 关1兴.
bearing can be actively adjusted by applying a voltage to the The vibration of the rotor is measured using eddy-current gap
embedded piezoelectric actuators. Experiments were performed sensors with a sensitivity of 0.5 ␮m. The four gap sensors are
on the active control of the self-excited vibration in the rotor- divided into two groups and fixed at two planes at 10.84 mm
above and 14.16 mm below the center of mass of the rotor. The
two gap sensors in the same plane are mounted 90 deg apart and
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received March 2002; measure the lateral displacement of the shaft in that plain, as
Revised March 2003. Associate Editor: G. Flowers. shown in Fig. 3. The positions of the sensors are labeled as LR,
Fig. 1 Configuration of the rotor-bearing system
LL, UR and UL. The LR and LL sensors are installed in the lower rotor is supported by the pressure of the gas films in the radial
plane and the UR and UL sensors are installed in upper plane. direction. The pressure distribution of the gas film, and therefore
Moreover, the LL and UL sensors measure the displacement of the the resultant force due to pressure, are dependent upon the clear-
shaft in the y direction, and the LR and UR sensors measure the ance distribution between the shaft and pads. The resultant force
displacement in the x direction of the right-handed coordinate sys- can be actively controlled by changing the radial position of the
tems shown in Figs. 1 and 3. pivot. Therefore, the vibration of the shaft can be actively sup-
The characteristics of the rotor vibration are measured using the pressed when an appropriate voltage is applied to the piezoelectric
following procedure. Firstly, a 40 V DC voltage is applied to all actuators.
four piezoelectric actuators as a bias voltage in the vibration con- In order to investigate the static characteristics of the PZT ac-
trol of the rotor system. Secondly, the thrust bearing is activated tuators in the rotor-bearing system, the bias displacements of the
by supplying pressurized air. Thirdly, the valves of the gas sup- rotor are measured at the four sensors when a DC voltage was
plies for the driving impeller and the braking impeller are turned applied to the UR actuator during operation. The bias displace-
on to start the rotation of the shaft. The rotation speed of the shaft, ments are the DC components of the rotor displacement, that is,
determined by the relative pressure of the air supplied to the driv- the displacements of the whirling center of the rotor. Figure 6
ing impeller and the braking impeller, is raised to 30,000 rpm by shows the results measured at 11,600 rpm. As the voltage in-
increasing the flow rate of the air to the driving impeller and then creased from 0 V to 150 V, the displacements of the shaft mea-
reduced to 0 rpm again by decreasing the flow rate. The vibration sured by the LR and UR gap sensors decreased from 0 ␮m to
of the shaft is measured when the speed of the rotor reduces from ⫺4.3 ␮m, while the displacements of the shaft at the LL and UL
30,000 rpm to 0 rpm. sensors increased from 0 ␮m to 2.2 ␮m at a rotation speed of
Figure 5 shows the waterfall plot of the vibration characteristics 11,600 rpm. This means that, due to the static displacement gen-
in the y direction measured by the LL sensor. It can be seen that erated by the PZT actuator in the UR pad, the rotor moves away
both the synchronous vibration with the frequency equal to the from the LR and UR sensors but closer to the LL and UL sensors.
rotation speed, and the asychronous vibration 共self-excited vibra-
tion兲, are induced in the rotor system. The sychronous vibration 4 Control System and Experimental Setup
reaches the maximum amplitude at 11,600 rpm, which equals the
Figure 7 shows the configuration of the experimental setup of
natural frequency of the first rigid mode of the rotor. The fre-
the rotor system and the feedback control system used. Since it is
quency of the self-excited vibration is about 57 Hz, regarless of
difficult to derive a mathematical model of the dynamical system,
the rotational speed, but its amplitude also reaches the maximum
a PID control method that does not require a mathematical model
value at around 11,600 rpm. The fluid-related instability in both
of the control object is used in this study. It is also assumed that
the journal bearings and the thrust bearings can induce the self-
the interaction between the actuators is negligible and that each
excited vibration. A dynamic coupling between the journal bear-
actuator can be controlled independently. The input-output rela-
ings and the thrust bearings may also occur in the rotor-bearing
tionship of the controller for each actuator is defined by the fol-
system. The objective of this study is to control the self-excited
lowing equation:
vibration of the rotor-bearing system.
3 Active Gas Bearings

U共 s 兲
Y共s兲
Ki
s 冉
⫽K p ⫹K d s⫹ ⫽K p 1⫹T d s⫹
1
T is 冊 (1)
In order to control the vibration of the shaft, piezoelectric ac- where Y (s) is the feedback signal 共in volts兲, U(s) is the control
tuators were embedded in two of the three pivots in each bearing, input 共the voltage applied to the actuators in volts兲, and K p , K d
as shown in Fig. 3. The piezoelectric actuators are 5⫻5 and K i are the proportional, derivative and integral gains, respec-
⫻10 mm in size and can generate 6.1 ␮m of displacement at 200 tively. Moreover, T i ⫽K p /K i and T d ⫽K d /K p are the integral and
V input voltage. The piezoelectric actuators, which have a wide derivative constants.
frequency bandwidth and can generate a large force output, are Different gains were selected for different actuators in the ex-
suitable for this kind of application. The positions of the four periment. For the actuator installed in the same direction as the
actuators are also labeled as LR, LL, UR and UL. The LL and UL sensor, the output of the sensor was used directly as the feedback
actuators are embedded in the pivots on the y-z plane and the LR signal Y (s). For the actuators at the LR and UR positions, the
and UR actuators are embedded in the pivots on the plane at 30 feedback signal Y (s) was estimated from the output of the two
degrees clockwise from the x-y plane. sensors at the nearest measuring plane.
When the shaft rotates, a pressurized gas film is formed be- Figure 8 shows a block diagram of the active control system.
tween the shaft and each pad due to the viscosity of the gas. The The analog signals from the sensors are sampled at a frequency of

Fig. 2 Configuration of the high-expansion-ratio expander „a… Longitudinal
cross-section „b… Transverse cross-section

Fig. 3 Configuration of the tilting-pad journal gas bearings
Fig. 4 Configuration of the externally-pressurized thrust gas
bearings „a… Distribution of nozzles „b… Upper and lower
10 kHz through an A/D converter and then transmitted to a DSP. clearances
The designed controller was implemented digitally in the DSP,
and the control input calculated on the basis of the sensor signals.
The calculated control input was added to a bias voltage, ampli- duced. However, the synchronous vibration remained almost un-
fied and then applied to the actuators. The bias voltage was super- changed. The small peak at 50 Hz is due to the electromagnetic
posed on the input voltage calculated from Eq. 共1兲 in order to noise of the power supplies.
increase the maximum output displacement of the actuators. The Figure 10 show the rotor frequency response measured by the
gain of the power amplifier is 20 and the bias voltage after am- four sensors at 11,600 rpm. K d was set to 0.01, and the propor-
plification is 40 V. The gain of the sensor and the power amplifier tional gains K p of the UL and UR actuators were set to the opti-
are not included in the transfer function in Eq. 共1兲.
5 Results and Discussions Table 2 Parameters of the thrust bearing

Due to the influence of the pressurized air supplied to the driv- Outer diameter, r 1 32 mm
ing impeller, the actuators at LL and LR were less effective in the Inner diameter, r 0 16.4 mm
control of the self-excited vibration than the other two actuators. Diameter of nozzle position circle, r a 22.6 mm
Therefore, the gains of these two actuators were fixed at K d Number of nozzles, n 18
Diameter of nozzles in the lower bearing, r s1 0.6 mm
⫽0.01, K i ⫽0.001 and K p ⫽50, while the gains of the other two Diameter of nozzles in the upper bearing, r s2 0.3 mm
actuators at UL and UR were varied to investigate their influence. Unloaded clearance of the lower bearing h n1 23.4 ␮m
Moreover, the integral gain of the two actuators at UL and UR Unloaded clearance of the upper bearing h n2 16.6 ␮m
was also set to 0.001, since it has little influence on the control Nondimensional axial displacement ⌬h n /(h n1 ⫹h n2 ) ⬍0.22
effect. Control experiments were performed for different combi-
nations of K p and K d for the actuators at UL and UR. Figure 9
shows the shaft frequency response measured by the sensors at
UR and LR for a rotation speed of 10,000 rpm with gains K p and
K d for the actuators at UL and UR set to 10 and 0.02, respectively.
The asynchronous vibration 共about 57 Hz兲 was significantly re-
Table 1 Parameters of the rotor-bearing system
Shaft length, l 181.35 mm l1 23.54 mm

Shaft mass, m 0.28 kg l2 26.46 mm
Shaft radius, R 8 mm l3 10.84 mm
Bearing width, L 15.4 mm l4 14.16 mm
Length of pad, ␤ 110° l5 90.25 mm
Clearance, C r 20 ␮m l6 91.1 mm
Pivot position ␤ p / ␤ 0.65 l7 38.74 mm
Dimensionless preload 0.5 l8 6 mm Fig. 5 Waterfall chart of the frequency response of the rotor at
the LL gap sensor „experimental result…

Fig. 6 Bias displacement of the shaft versus DC voltage ap-
plied to the UR actuator
Fig. 8 Block diagram of the control system
mal values of 150 and 100, respectively. Since the rotor speed Figure 12 shows the rotor frequency response measured by the
equals the resonant frequency of the rotor-bearing system, the UR sensor when the K p of the actuators at UL and UR was set to
amplitude of the asynchronous vibration increases significantly at 10, and K d was set to 0.2 for Case 共a兲 and 0.01 for Case 共b兲. In
the uncontrolled state. However, the asynchronous vibration is Case 共b兲, the asynchronous vibration of 57 Hz was satisfactorily
almost completely suppressed after control. Despite this, little ef- suppressed, though a low peak remained due to the small propor-
fect could be observed on the synchronous vibration. The other tional gain. However in Case 共a兲, another asynchronous vibration
components of the asynchronous vibration with frequencies of 25 of 67 Hz was significantly excited. Moreover, the gains used in
Hz, 79 Hz and 132 Hz were also suppressed. Case 共b兲 are the same as those in Fig. 9, except that K d decreased
The influence of the feedback gain on the control results was from 0.02 to 0.01. This means that the same gains can yield good
further investigated. Fig. 11 shows the control results when the control performance for a range of rotor speeds. The equivalent
proportional gains of the actuators at UL and UR were set to 200, stiffness and damping coefficients of the closed-loop system cor-
while the rotor speed and the other gains were kept the same as responding to the proportional and differential gains cannot be
those in the former case. The asynchronous vibration at a fre- estimated quantitatively since the mathematical model of the con-
quency of 57 Hz was almost completely suppressed. However, a trol plant is unknown. However, their variation can be observed
second asynchronous vibration of 93 Hz was excited. from the position and shape of the peak corresponding to the
asynchronous vibration in the power spectra shown in Fig. 9–12.
The decrease in the stiffness coefficients moves the peak of the
asynchronous vibration to the left side 共Fig. 9兲, while the peak
becomes sharper as the damping coefficient decreases 共Fig. 10兲.
Fig. 9 Power spectra of the shaft at the LR sensor position

„Rotation speed: 10,000 rpm, PID gain: K p of UL act.Ä10, K p of
UR act.Ä10…
Fig. 7 Schematic diagram of the active control system

Fig. 10 Power spectra of the shaft at different sensor positions „Rotation speed: 11,600 rpm, PID gain: K p of UL act.Ä150, K p
of UR act.Ä100… „a… UL sensor position „b… LL sensor position „c… UR sensor position „d… LR sensor position
Fig. 12 Power spectra of the shaft at the UR sensor position

Fig. 11 Power spectra of the shaft at the UL sensor position „Rotation speed: 11,600 rpm, „a… PID control: K d of UL Act.
„Rotation speed: 11,600 rpm, PID gain: K p of UL act.Ä200, K p of Ä0.2, K p of UL act.Ä10, „b… K d of UR act.Ä0.01, K p of UR act.
UR act.Ä200… Ä10…

6 Conclusions References
A novel tilting-pad journal gas bearing system with active pads 关1兴 Ino, N., Machida, A., Tsugawa, K., and Hashimoto, H., 1991, ‘‘Development
of Externally Pressurized Thrust Bearing for High-Expansion-Ratio Helium
was studied for the control of self-excited vibration in a rotor Expander, Trans. Jpn. Soc. Mech. Eng., Ser. C, 共in Japanese兲 58共550兲, pp.
bearing system. Each active pad contains an embedded piezoelec- 1885–1891.
tric actuator, used to adjust the clearance between the shaft and 关2兴 Togo, T., 2000, Investigation Report on the Research of Gas Bearings, Japa-
pads. The novel gas bearings were used in the rotor system of a nese Society of Tribologists.
关3兴 Andres, L. S., 1996, ‘‘Turbulent Flow, Flexure-Pivot Hybrid Bearings for
high-expansion-ratio turbo expander for helium liquefier, and a Cryogenic Applications,’’ ASME J. Tribol., 118, pp. 190–200.
feedback control system was constructed with four gap sensors, 关4兴 Qiu, J., Takagi, T., Tani, J., Machida, A., Tsugawa, K., Yanagi, H., and Ino, N.,
embedded actuators and PID controllers. It was confirmed by ex- 1993, ‘‘Dynamic Characteristics of a Tilting-pad Bearing System for High
periment that the self-excited vibration in the rotor system could Expansion-Ratio Expander,’’ Adv. Cryog. Eng., 39, pp. 909–914.
关5兴 Bently, D. E., and Muszynska, A., 1989, ‘‘Anti-Swirl Arrangements Prevent
be effectively suppressed with the feedback control if the gains of Rotor/Seal Instability,’’ Stress, Reliab. Des. 111共2兲, pp. 156 –162.
the controllers were appropriately chosen. 关6兴 Ihara, K., 1987, ‘‘Gas Expander for Lower Temperature,’’ Journal of Turbine
It was also found that little control was achieved for the syn- Machinery 共in Japanese兲, 15共11兲, pp. 57– 62.
chronous vibration of the rotor system using the present configu- 关7兴 Palazzolo, A. B., Lin, R. R., Alexander, R. M., Kascak, A. F., and Montague,
J., 1989, ‘‘Piezoelectric Pushers for Active Vibration Control of Rotating Ma-
ration. The active control of synchronous vibration will be the chinery,’’ ASME J. of Vibration, Acoustics, Stress, Reliab. Des. 111, pp. 298 –
subject of future work. 305.
关8兴 Palazzolo, A. B., Jagannathan, S., Kascak, A. F., Montague, G. T., and Kiraly,
L. J., 1993, ‘‘Hybrid Active Vibration Control of Rotorbearing Systems Using
Piezoelectric Actuators,’’ J. Vibr. Acoust., 135, pp. 111–119.
关9兴 Horikawa, O., and Shimokohbe, A., 1990, ‘‘An Active Air Bearing,’’ JSME,
Acknowledgment Int. Journal, 33共1兲, pp. 55– 60.
The authors thank Dr. Yanagi, and Mr. Tsugawa of Mayekawa 关10兴 Muszynska, A., Franklin, W. D., and Bently, D. E., 1988, ‘‘Rotor Active ‘Anti-
Swirl’ Control,’’ Stress, Reliab. Des. 110共2兲, pp. 143–150.
Mfg. Co. Ltd., MYCOM Advanced Technology Laboratory, 关11兴 Sun, L., Krodkiewski, J. M., and Cen, Y., 1998, ‘‘Self-Tuning Adaptive Con-
Ibaraki, Japan, for their effort and cooperation in the design and trol of Forced Vibration in Rotor Systems using an Active Journal Bearing,’’ J.
manufacturing of the rotor-bearing system. Sound Vib., 213共1兲, pp. 1–14.

Vibration of Flex Circuits in Hard
Disk Drives
A flex circuit connects the stationary electronic components in a hard disk drive to the
rotating arm that carries the read/write heads and positions them above data tracks on
the disk. Flex circuits are conventionally formed as a laminate of polyimide substrate,
adhesive, and copper conductors. Deformation of a flex circuit is discussed in the context
J. A. Wickert of the following stages: the initial unstressed shape, configurations in which stresses set
Fellow ASME, and relax in response to elevated temperature, equilibrium, and small amplitude vibra-
Department of Mechanical Engineering, tion. The model involves displacements of the flex circuit in the directions tangent and
Carnegie Mellon University, normal to the local equilibrium shape, and those motions couple with the arm’s dynamics.
Pittsburgh, PA 15213 Nonlinearity associated with finite curvature, partial elastic springback, and the arm’s
geometry and inertia properties are incorporated within the vibration model to predict
system-level natural frequencies, mode shapes, and coupling factors between the circuit
and the arm. Laboratory measurements using noncontact laser interferometry validate the
model with respect to the circuit’s shape, stiffness, restoring moment, and natural frequen-
cies. The primary degrees of freedom for optimizing flex circuit design are the thicknesses
of the individual layers within the circuit, free length, and the locations and slopes of the
circuit’s attachment points to the arm and electronics block. The model’s predictions and
trends developed from a case study in free length are discussed with a view toward
reducing coupling between the circuit and arm in certain vibration modes.
关DOI: 10.1115/1.1547661兴
1 Introduction The mechanical behavior and vibration of a hard drive’s disks,

motor, spindle, bearings, voice coil, and other elements have been
The storage density of hard disk drives, as measured by the
the subject of substantial engineering development effort since the
number of data bits captured within a unit area, has grown at a first introduction of a hard disk drive in 1956. As drives have
historical rate of about 60% per year, and that rate has accelerated become more sophisticated, sources of vibration that had previ-
recently to more than 100% per year. In a similar vein, the cost ously been within tolerance limits have become the focus of fur-
associated with storing a megabyte of data has fallen by over four ther improvement efforts. In particular, the vibration of flex cir-
orders of magnitude during the past two decades. With the physi- cuits and their coupling with motion of the read/write heads are
cal limit for the density of magnetic disk recording now appearing now factors in high density recording applications. No longer
on the horizon, higher-precision requirements are being placed on viewed as a lightweight appendix to the arm and voice coil, the
the vibration of each structural and actuation component. flex circuit has dynamics that couple through the arm and produce
By way of background and motivation, the construction of a track positioning errors.
hard disk drive is shown in the photograph of Fig. 1. Data is Reducing flex circuit vibration in certain modes is a potential
stored as magnetic transitions on the thin media layer that coats strategy for improving transient settling response after a seek op-
the glass disks. The arm pivots about a fixed bearing, and it is eration to a new data track. From the modeling perspective, an
driven by a voice coil that swings between two permanent mag- objective of this investigation is to better understand the flex cir-
nets. The read/write heads are located at the tip of the arm above cuit’s equilibrium shape involving finite deformation and partial
each disk surface, but they are not discernible in Fig. 1 because of elastic springback, and the character of its small amplitude vibra-
their small size. The particular drive shown in the figure has a tion about equilibrium. Of particular interest are the flex circuit’s
total of ten read/write heads. static shape at a specified arm rotation angle, natural frequencies,
The voice coil and its companion servo system slew the heads and coupling to the arm and read/write heads in each mode. In
to a desired cylinder of data, and follow it in the presence of disk what follows, the roles played by such parameters as the circuit’s
runout, vibration, windage, and other disturbances. Electrical laminated structure, finite curvature, free length, and boundary
leads are routed to each read/write head to carry the recording and conditions at the arm and electronics block are explored.
readback signals, and other larger leads power the voice coil. Each
head and coil wire is integrated within a flat and flexible circuit 2 Material Characteristics and Loading Sequence
that conveys all of the electrical signals between the 共rotating兲 arm
and the 共stationary兲 electronics on the drive’s body. This overall In Fig. 3, the two micrographs of a flex circuit’s cross-section
construction is shown schematically in Fig. 2. In a typical embodi- describe the internal morphology and dimensions of the several
ment, the flex circuit is a polyimide film laminate having rolled material layers. In optical diagnostics, a segment of the flex circuit
annealed copper wires in its conductor layer. Flex circuits replace was embedded in epoxy, diced, and polished smooth. A thin layer
conventional multi-lead wiring and combine electrical functional- of gold was then sputtered onto the surface of the cross-section to
ity with mechanical flexibility, which in turn can potentially intro- enhance imaging. The particular circuit of Fig. 3 is formed as a
duce unwanted vibration. Quite aside from hard disk drives, other symmetric sandwich laminate of two h p ⫽31 ␮ m thick polyimide
types of flex circuits are used in avionics packages, gyroscopes, layers on either side of flat copper electrical leads. Kapton® is
hearing aids, and cardiac pacemakers. often used as the substrate material, and it is chosen on the basis
of electrical, mechanical, and thermal properties. In Fig. 3共a兲, a
matrix of epoxy bonds the polyimide layers to the conductors, and
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002; the thickness of the intervening adhesive is h a ⫽13 ␮ m. The cir-
revised October 2002. Associate Editor: Chin An Tan. cuit comprises twenty conductors having common h c ⫽26 ␮ m
Fig. 1 Photograph depicting certain mechanical components within a hard
disk drive used for computer data storage
thickness, and one of four different widths depending on the type counterparts in the circuit’s natural state where no reactions are
of electrical signal being carried. Some ␦ ⫽67% of the flex cir- applied through attachment to the arm and electronics block. Fur-
cuit’s b⫽11.15 mm width is associated with the conductors, and ther, the flex circuit’s stresses, particularly in the adhesive layer,
the remaining fraction of space between the conductors remains are known to partially relax with time and temperature. Deforma-
insulated by adhesive. Figure 3共b兲 depicts a second portion of the tion from the initial free shape, through finite motion to equilib-
same flex circuit, and its image is spaced width-wise relative to rium, and ultimately to small amplitude vibration are broken down
Fig. 3共a兲. Three adjacent conductors, slight variation in the adhe- in the following sequence of loading states:
sive thickness, and indentation of the conductor into the lower Initial S( 0 ) The flex circuit is specified to be initially unstressed
substrate are visible features. The cumulative thickness is h⫽h c and straight. Imperfections that are present as a result of the cir-
⫹2(h p ⫹h a )⫽114 ␮ m, and the flex circuit has linear density cuit’s production or the drive’s final assembly could be incorpo-
2.38 g/m in the direction along its free length L. rated at this stage by specifying a functional form of curvature
Vibration of the flex circuit depends in part on its equilibrium k ( 0 ) (s), where s苸(0,L) denotes arc length, but such effects are
curvature and internal loads, which are distinguished from their not considered at this first level of approximation.
Set S( 1 ) Referring to Fig. 2, the flex circuit is bent from S ( 0 ) and
attached to the arm and electronics block. Elements within it are
subjected to tension T(s), shear force N(s), and bending moment
M (s). During the process to S ( 1 ) , the circuit undergoes finite
Fig. 3 „a… Brightfield and „b… differential interference contrast

micrographs of flex circuit cross-sections. The images depict
Fig. 2 Schematic of the equilibrium and vibration model for the construction morphology and the thicknesses of the poly-
the arm and flex circuit mechanism imide, conductor, and adhesive layers.

deformation to the new shape of curvature k ( 1 ) . The coordinates Table 1 Baseline parameters for the arm and flex circuit’s
(x L ,y L ) of the contact point and the angle ␪ L of approach at the equilibrium and vibration models
electronics block are specified. Rotation ␣ of the arm is an inde- Flex Circuit
pendent variable in S ( 1 ) , and together with the geometric param-
eters ␤, ␥, and r defined in Fig. 2, its value sets the circuit’s Width, b 11.1 mm
Thickness
conditions Polyimide layer, h p 31 ␮m
Adhesive layer, h a 13 ␮m
x 共 0 兲 ⫽r cos共 ␣ ⫺ ␤ 兲 y 共 0 兲 ⫽r sin共 ␣ ⫺ ␤ 兲 (1) Conductor layer, h c 26 ␮m
Cumulative, h 114 ␮m
Conductor fraction, ␦ 67%
␪共 0 兲⫽␣⫺␥ (2) Free length, L 31 mm
Composite linear density, ␳ A 2.38 g/m
Modulus
at its endpoint s⫽L. In S ( 1 ) , the arm is held by an external agent Polyimide, E p 2.75 GPa
at an angle that could correspond to the disk’s inner 共ID兲 or outer Adhesive, E a 1.03 GPa
共OD兲 diameters, or the load-unload 共LU兲 support where the arm is Conductor, E c 115 GPa
parked when the drive is not in use. Springback ratio, p 0.55
Composite bending stiffness, EI 4.78⫻(10⫺6 ) N•m2
Springback S( 2 ) During both its production and day-to-day usage, Composite axial stiffness, EA 2.45⫻(104 ) N
the flex circuit is exposed to elevated temperature. Certain fabri-
cation stages take place at 80°C; an assembly-level wash/dry Arm
cycle occurs at 85°C; and the internal environment of a perfor- Attachment radius, r 10.6 mm
mance drive can reach 50– 60°C. By comparison, the glass tran- Read/write head radius, R 43.8 mm
sition temperature of an adhesive commonly used within flex cir- Attachment angle, ␤ 95°
cuits lies in the range 50–95°C. In the springback stage, while the Offset angle, ␥ 8°
Angles
flex circuit remains attached to the arm and electronics block in Load-unload, ␣ LU 23°
S ( 1 ) , stresses are allowed to partially relax and the circuit assumes Disk OD, ␣ OD 32°
the new natural state S ( 2 ) . To characterize this process, thermal Disk ID, ␣ ID 64°
Inertia, m 17.2 g
cycling experiments were conducted with the arm locked at the Radius of gyration, ␬ 11.9 mm
disk’s inner diameter ( ␣ ID⫽64 deg) and the assembly held at
45°C for one hour. After having cooled to ambient temperature, Electronics Block
the arm and flex circuit assumed nearly that point as the new Coordinates, (x L ,y L ) 共24, ⫺11兲 mm
equilibrium configuration. The arm was then pivoted to the disk’s Tangency angle, ␪ L ⫺90°
outer diameter ( ␣ OD⫽32 deg), thermally cycled, and observed to
assume a nearby position in equilibrium. During those processes,
however, the stresses within the flex circuit were only partially
reduced. Indeed, when the circuit was cut free and released, it directions tangent and normal to the local equilibrium, and those
assumed a new natural state having shape and curvature some- displacements are denoted by u(s,t) and v (s,t), respectively.
what between k ( 0 ) and k ( 1 ) . The empirical factor p is introduced to
quantify the amount of elastic springback that occurs when the 3 Equilibrium Shape, Loads, and Stresses
flex circuit is unloaded, and the curvature in S ( 2 ) is defined k ( 2 )
⫽(1⫺p)k ( 1 ) . When p⫽1, the circuit behaves elastically, is unaf- The flex circuit’s shape in S ( 3 ) is defined parametrically by the
fected by exposure to elevated temperature, and returns to its ini- coordinates (x(s),y(s)). With the nomenclature (䊉) ⬘ ⫽d/ds,
tial straight shape when unloaded from S ( 1 ) . On the other hand, equilibrium in the local tangential and normal directions is gov-
when p⫽0, the flex circuit assumes S ( 1 ) as the new natural state. erned by the force and moment balances 关1兴
In practice, the actual loading process and k ( 2 ) lie between these T ⬘ ⫽Nk, N ⬘ ⫽⫺Tk, k ⬘ ⫽⫺N/EI⫹k 共 2 兲 ⬘ (3)
two extremes, and they change in a time-, temperature-, and
history-dependent manner as the arm continuously slews between in which the constitutive relation M ⫽EI(k⫺k ) has been em-
(2)
the disk’s ID and OD. In the baseline parameters of Table 1, the bedded. The circuit’s bending stiffness is given by
value p⫽0.55 was determined by experience in thermally cycling
several disk drives and examining the circuits when they were cut
and released from the arm and electronics block. The flex circuits
EI⫽
1
12 c c
1
E h 3 ␦ b⫹ E a h 3c 共 1⫺ ␦ 兲 b⫹2E a
12
1
冉 1
bh 3 ⫹ bh 共 h
12 a 4 a c
were photographed, and their natural shapes were compared to
those predicted by the equilibrium model described below at vari-
ous levels of springback. Parameter p was then adjusted by cut-
冊冉
⫹h a 兲 2 ⫹2E p
1
12
1
bh 3p ⫹ bh p 共 h c ⫹h p ⫹2h a 兲 2
4 冊 (4)
and-try until the predicted and measured natural shapes were where values for the elastic constants of the conductor E c , poly-
agreeably close. imide E p , and adhesive E a are listed in Table 1. While Eq. 共4兲 is
Equilibrium S( 3 ) From the natural configuration S ( 2 ) , the flex specific to the cross-sectional construction of Fig. 3, the treatment
circuit is imagined to be re-attached to the arm and electronics can be adapted for other geometries.
block. With no external torque M o about the pivot point being On the basis of measured layer thicknesses and published elas-
applied by the voice coil, the arm rotates to equilibrium in re- tic constants 关2兴 in Table 1, the circuit’s composite bending stiff-
sponse to the combined influence of the tension, shear force, and ness is EI⫽4.78⫻10⫺6 N•m2 . The conductor layer contributes
bending moment at s⫽0. At this stage, the arm’s equilibrium 26% to the stiffness, the polyimide layer 71%, and the adhesive
angle and the circuit’s curvature are denoted by ␣ and k layer only 3%. This value of EI was validated by both static bend-
* *
⫽k ( 3 ) , respectively. ing and natural frequency measurements conducted with a circuit
Deformed S( 4 ) In this loading stage, the flex circuit is elastically segment that was embedded as a cantilever in an epoxy casting. In
deformed about equilibrium, corresponding to static slew at a cer- the static test, the segment was mounted on a micrometer transla-
tain angle, to small amplitude vibration of the flex circuit about tion stage, and under specified displacement, the force applied to
equilibrium, or to finite amplitude vibration. The tension, shear it’s tip was measured. The signal from a planar beam sensor 共Fu-
force, and bending moment are each incremented relative to their tek FR-1020兲 was conditioned and amplified to provide a cali-
equilibrium values. Motion of the flex circuit is resolved into the brated and linear force response for loads up to 295 mN. The

Fig. 4 Predicted „ … and measured shapes of the flex circuit
in equilibrium with attachment to the arm and electronics Fig. 5 Equilibrium tension and shear force along the circuit’s
block, and in its natural state. The shapes extracted from pho- arc length
tographs of the circuit in the two states are denoted by the „䊊…
and „䊉… data points, respectively.
along the circuit’s length from the arm. While the average strain
over the cross-section is only T/EA⬇⫺1.3⫻10⫺6 with
stiffness recorded on this basis was 4.60⫻10⫺6 N•m2 . In the sec- EA⫽E c h c b ␦ ⫹2E a h a b⫹E a h c b 共 1⫺ ␦ 兲 ⫹2E p h p b (6)
ond validation test, the value 4.91⫻10⫺6 N•m2 was determined the peak bending strain h c 兩 k ⫺k 兩 /2 within the conductor layer
(2)
by measuring the segment’s two lowest natural frequencies and *
is 0.12%. For other free lengths, arm positions, or endpoint loca-
matching them to the values expected for a cantilever. In the cal- tions, local yielding could occur within the ductile conductor.
culations described below, the stiffness value determined from Eq. As the arm slews about equilibrium, the flex circuit bends fur-
共4兲 was used.
The circuit’s equilibrium curvature is determined by integrating ther to S ( 4 ) as shown in Fig. 6. The voice coil applies the static
Eqs. 共3兲, and its shape is found subsequently from the kinematic bias torque
relations M o ⫽T 共 0 兲 r sin共 ␤ ⫺ ␥ 兲 ⫹N 共 0 兲 r cos共 ␤ ⫺ ␥ 兲 ⫹M 共 0 兲 (7)
x ⬘ ⫽cos ␪ , y ⬘ ⫽sin ␪ , ␪ ⬘ ⫽k 共 s 兲 (5) which increases in Fig. 7 from zero at equilibrium to the maxi-
subject to position and slope conditions at s⫽0 and L. While mum value 0.35 N•mm at the disk’s ID. Multiple measurements
some rotation between the circuit and arm does occur locally at of the torque were made for one disk drive at four different slew
s⫽0, that effect is neglected in this first approximation, recogniz- angles, and those results are also shown in Fig. 7. The indicated
ing that the resulting model may overestimate the circuit’s actual variation of M o is representative of such measurements and cap-
boundary stiffness. tures hysteresis in the circuit, and friction in the pivot bearing and
Solutions to Eqs. 共3兲 and 共5兲 are found through a shooting
method in which the boundary value problem in x, y, and ␪ is
converted to an initial value problem, and then solved iteratively.
Estimates for 共initially unknown兲 N(0), T(0), and k(0) are as-
signed, and the system of differential equations is integrated nu-
merically. The values x(L), y(L), and ␪ (L) which result from
that calculation do not, in general, satisfy the endpoint constraints.
In iteration, N(0), T(0), and k(0) are then adjusted through a
nonlinear minimization scheme until each kinematic condition is
satisfied within desired tolerance.
On the basis of the parameters in Table 1, Fig. 4 depicts simu-
lated and measured shapes of the flex circuit in its natural (S ( 2 ) )
and equilibrium (S ( 3 ) ) states. The data points shown in the figure
represent coordinate locations as extracted from photographs of
the circuit in the two configurations. Curvature in the natural state
was set with the arm held at the disk’s OD and the springback
constant being p⫽0.55. At the equilibrium angle ␣ ⫽26.4 deg,
*
the tension, shear force, and bending moment applied by the flex
circuit to the arm produce no resultant torque about the pivot
bearing. Their variations in response to first-order changes in ␣,
however, are captured by stiffness 10.2 mN•mm/deg.
The equilibrium tension and shear force distributions along the
circuit are shown in Fig. 5. The circuit is compressed over its Fig. 6 Variation of the flex circuit’s static shape for arm posi-
entire length with the minimum, mean, and maximum loading tions which range between the disk’s outer and inner diam-
values for T being ⫺55, ⫺33, and ⫺14 mN, respectively. The eters. For each arm position, the locations of the circuit’s end-
zero crossing in shear occurs at a distance approximately 64% points are denoted by „䊊….

Fig. 7 Predicted „ … and measured „䊊… dependencies of the static restoring mo-
ment for arm positions between the disk’s outer and inner diameters
supports. The trend in Fig. 7 is agreeably linear over the operating T⫽T ⫹ ⑀ T 1 ⫹¯ (9)
range, although the circuit’s stiffness dM o /d ␣ does decrease in *
the disk’s mid-diameter range, and grow again for slew angles k⫽k ⫹ ⑀ k 1 ⫹¯ (10)
*
near the ID. about their equilibrium values, denoted by (䊉) . Here ⑀ Ⰶ1 is a
Figure 8 depicts the manner in which the flex circuit’s equilib- *
dimensionless scaling parameter used in the linearization, and the
rium shape changes as a function of its free length. Results are first-order corrections are written 关3,4兴
shown for simulations in which L varies between 70% and 130%
of its nominal value. With the other parameters of Table 1 remain- N 1 ⫽⫺EI 共 v ,ss ⫹ 共 k u 兲 ,s 兲 ,s (11)
*
ing fixed, the corresponding changes to the arm’s equilibrium
angle and the circuit’s static stiffness are listed in Table 2. T 1 ⫽EA 共 u ,s ⫺k v 兲 (12)
*
k 1 ⫽ 共 v ,s ⫹ 共 k u 兲兲 ,s (13)
*
4 Natural Frequencies, Vibration Modes, and Cou- in terms of the circuit’s tangential and normal displacements. Here
pling the comma-subscript notation signifies partial differentiation. The
For small amplitude vibration in S ( 4 ) , the shear, tension, and equations of motion become
curvature are expanded ␳ Au ,tt ⫺T 1,s ⫹N k 1 ⫹k N 1 ⫽0 (14)
* *
N⫽N ⫹ ⑀ N 1 ⫹¯ (8) ␳ A v ,tt ⫺N 1,s ⫺T k 1 ⫺k T 1 ⫽0 (15)
* * *
where ␳ A is the circuit’s mass-per-unit-length.
Vibration of the flex circuit and arm couple through
m ␬ 2 ␣¨ 1 ⫽T 1 共 0,t 兲 r sin共 ␤ ⫺ ␥ 兲 ⫹N 1 共 0,t 兲 r cos共 ␤ ⫺ ␥ 兲 ⫹M 1 共 0,t 兲
(16)
where m is the arm’s mass, ␬ is it’s radius of gyration about the
pivot bearing, and ␣ 1 is the first-order rotation about ␣ . The
*
time-dependent incremental loads in Eq. 共16兲 are evaluated at the
circuit’s connection point to the arm. Although not considered
here, the effects of the voice coil’s driving torque and damping in
Table 2 Dependence of the equilibrium angle and the circuit’s

stiffness with respect to free length over a range 70% to 130%
of nominal
Equilibrium Stiffness
Length ratio Length 共mm兲 angle 共deg兲共mN•mm/deg兲
70% 21.7 41.3 63.6
80% 24.8 33.0 40.0
90% 27.9 28.4 20.2
100% 31.0 26.4 10.2
110% 34.1 29.3 5.6
Fig. 8 Flex circuit equilibrium shapes predicted for free 120% 37.2 36.9 5.0
length’s which range between 70% and 130% of the nominal 130% 40.3 44.2 5.8
value in Table 1

Fig. 9 First six vibration modes „ … shown superposed on
the equilibrium shape „ …. Each element of the figure is
Fig. 11 Measured spectrum and natural frequencies of one
annotated with the predicted natural frequency and displace-
disk drive’s flex circuit in modes two, three, and four
ment ratio ␩.
the pivot bearing could be incorporated in Eq. 共16兲 at this stage

for direct numerical simulation or control system studies. Motion
of the circuit and arm are also constrained through
u⫽ 冕0
s
k v ds⫹ēs⫹r sin共 ␤ ⫹ ␥ 兲 ␣ 1
*
(19)
u 共 0 兲 ⫽r sin共 ␤ ⫹ ␥ 兲 ␣ 1 (17) where ē is the average longitudinal strain. By embedding Eq. 共19兲
into Eqs. 共11兲–共13兲 and Eq. 共15兲, the working form of the trans-
v共 0 兲 ⫽r cos共 ␤ ⫹ ␥ 兲 ␣ 1 , (18) verse equation of motion involves only the dependent variables v
and ␣ 1 , and it is applied to characterize the lower modes.
and the conditions u⫽0, v ⫽0, and v ,s ⫹k u⫽0 at s⫽L.
* The vibration model is discretized in terms of ␣ 1 and nodal
The flex circuit is substantially stiffer for displacements in u
values for v that are evenly-spaced along the circuit’s length, and
than v , and the ratio of longitudinal to flexural stiffness is mea-
the natural frequencies and mode shapes are determined numeri-
sured by the dimensionless parameter (EAL 2 )/(EI), which is cally. Figure 9 depicts the lowest six modes on the basis of the
O(106 ) for the problem at hand. On the time scales of the lower parameters listed in Table 1. Each element of the figure is anno-
transverse modes, tension variations propagate almost instanta- tated to indicate the mode’s natural frequency and the displace-
neously, and the explicit appearance of u in the equations of moment ratio
tion can be suppressed by approximating T 1 as a function of time
alone. After integrating Eq. 共12兲 in this manner, the longitudinal R 兩 ␣ 1兩
displacement field is approximately 关5兴 ␩⫽ , (20)
max共冑u 2 ⫹ v 2 兲
which measures the relative motions of the read/write head and
flex circuit. At only some three Hertz, the fundamental mode is
dominated by motion in ␣ 1 and set primarily by the arm’s inertia
and the circuit’s static stiffness. Because of its low frequency, this
mode’s dynamics generally do not contribute tracking errors to the
extent that the modes at 374 Hz, 837 Hz, 1.39 kHz, and higher
frequencies do.
Fig. 10 Test stand used for measuring flex circuit transient

responses and natural frequencies. A 90 deg prism directs the
target beam from the laser head to the flex circuit. A small
patch of retroreflective tape „not visible in the photograph… was
placed on the circuit to reduce measurement sensitivity to mis- Fig. 12 Ring-down of a flex circuit in its second mode, which
alignment and rotation of the circuit during vibration. for this disk drive was placed at 332 Hz. The damping ratio is
1.1%.

eter measured changes in the two light path lengths through the
interference fringes generated by superposition of coherent beams
that reflected from 共i兲 a stationary reference surface and 共ii兲 the
moving flex circuit. The target beam was directed onto the flex
circuit by a right-angle prism located on the concave side of the
flex circuit. To ensure that sufficient light was returned by the flex
circuit into the optical head, a small patch of retroreflective tape
was attached to the circuit at the measurement point. Particles
within the retroreflective medium ensured that a portion of the
incident light was returned into the source optical fiber regardless
of the flex circuit’s 共potentially兲 large displacement or slope. As
the arm or circuit was impacted, the displacement or velocity
signal was captured on an digital oscilloscope 共HP 54600A兲, and
its frequency content was characterized by using an dynamic sig-
Fig. 13 Transient response in the arm and flex circuit’s 3.08 Hz nal analyzer 共HP 35665A兲. Peaks in the autocorrelation record
fundamental mode following an impact provided the natural frequencies. With this technique, vibration
measurements were readily made with a strong signal-to-noise
ratio, and with displacement resolution and bandwidth exceeding
the test’s requirement.
Although the coupling ratios ␩ are only a fraction of a percent
in modes two and higher, they do have design implications be- In the spectrum for one disk drive shown in Fig. 11, the natural
cause of the high precision required of this mechanism. A data frequencies of modes two, three, and four were measured at 356
track on a typical performance drive may be only 0.5 ␮m wide, Hz, 844 Hz, and 1.23 kHz. Several of the flex circuit’s torsion
with positioning tolerances for the heads on the track of ⫾5% modes were also present in the illustrated 1.6 kHz frequency
共⫾25 nm兲 for write and ⫾10% 共⫾50 nm兲 for read operations. In range, but their content in Fig. 11 was suppressed by judicious
a situation where the flex circuit vibrates at an amplitude corre- placement of the impact and measurement points relative to the
sponding to one h⫽114 ␮ m thickness and with the coupling ratio torsion modes’ nodes. Despite its layered construction and attach-
being 0.23% in the second mode, the head would in turn respond ments to the arm and electronics block, the flex circuit presents a
with amplitude 260 nm, some ten times greater than the write damping ratio of only 1.1% in the second mode as indicated by
inhibit limit. the time record for ring-down in Fig. 12. By contrast, the mecha-
For vibration modes that involve significant flexure of the cir- nism is highly damped in the fundamental sway mode, with a
cuit, the natural frequencies are relatively insensitive to T and measured frequency at 3.08 Hz. Figure 13 depicts the arm and flex
* circuit’s transient response in that mode, and just over one cycle
N , at least for the ranges of parameters considered here. When
* of motion occurred following impact.
those forces are artificially set to zero, for instance, the differences
in the calculated natural frequencies for modes two through five Figures 14 and 15 show trends for the natural frequencies and
are only 3.2%, 2.1%, 1.4%, and 0.9%, respectively. Similarly, the coupling ratios which are predicted in a parameter study of free
differences in the calculated ␩ for those modes are only 3% or length L. For each length, the tension, shear force, and curvature
less. were determined on the basis of the equilibrium configurations
Figure 10 depicts a photograph of the test stand used for mea- shown in Fig. 8. The natural frequencies for modes two through
suring the flex circuit’s natural frequencies. Transverse vibration six decrease monotonically in Fig. 14 as L is examined over a
was measured by using a Michelson-style laser interferometer range 30% below, and 30% above, the nominal value. Also for
共Polytec OFV-3000兲, and fiber optic leads were used to establish these modes, Fig. 15 depicts the behavior of the displacement
the paths for the reference and target light beams. The interferom- ratio ␩. The fundamental mode is dominated by the arm’s sway
Fig. 14 Dependence of the natural frequencies in modes two through six on the flex
circuit’s free length

Fig. 15 Dependence of the displacement ratios ␩ in modes two through six on the
flex circuit’s free length
with the flex circuit responding nearly statically, and ␩ for that study in free length is discussed with a view toward exploring the
mode increases gradually within the range 357– 412%. In the mechanism’s design space and reducing arm-circuit coupling in
共even兲 modes two, four, and six, ␩ is relatively insensitive to certain vibration modes. Parameter and optimization studies with
design changes in L. On the other hand, ␩ has a zero crossing for respect to the model’s other degrees of freedom, and the implica-
the 共odd兲 modes three and five at nearly the same free length (L tions of flex circuit vibration for control system design, are sub-
⬇26 mm) which corresponds to some 83% of the baseline value. jects of current investigation.
At that design point, the natural frequencies have increased rela-
tive to their values at L⫽31 mm in Fig. 14, and the static stiffness
has likewise grown in Table 2. However, the modal displacement Acknowledgment
ratios can be reduced significantly or precisely driven to zero in This work was supported by a grant from IBM Corporation.
modes three and five. That trade-off may be desirable in certain The author appreciates the assistance of Matthew Brake in con-
applications. In short, the equilibrium and vibration model can be ducting the static stiffness and natural frequency measurements.
used to advantage for optimizing flex circuit designs with respect Kapton® is a registered trademark of E.I. du Pont de Nemours
to their vibration and load transmission performance. and Company.
5 Summary
The primary degrees of freedom for optimizing the design of References
flex circuits are the thickness of the layers within the circuit, its 关1兴 Love, A. E. H., 1944, A Treatise on the Mathematical Theory of Elasticity,
Fourth Edition, Dover Publications, New York.
free length, and the locations and angles of the attachment points 关2兴 Anonymous, 1996, General Specifications, Bulletin GS-96-7, DuPont Films,
at the arm and electronics block. Those parameters can be selected E.I. du Pont de Nemours and Company.
with respect to three criteria: 共i兲 the static bias torque applied by 关3兴 Perkins, N. C., 1990, ‘‘Planar Vibration of an Elastica Arch: Theory and Ex-
the voice coil to position the arm at a certain location on the disk; periment,’’ ASME J. Vibr. Acoust., 112, pp. 374 –379.
关4兴 Brush, D. O., and Almroth, B. O., 1975, Buckling of Bars, Plates, and Shells,
共ii兲 the circuit’s natural frequencies relative to the control system’s McGraw-Hill, New York.
bandwidth; and 共iii兲 the extent of vibration coupling between the 关5兴 Wickert, J. A., 1992, ‘‘Non-linear Vibration of a Traveling Tensioned Beam,’’
flex circuit and the arm in certain vibration modes. The parameter Int. J. Non-Linear Mech., 27, pp. 503–517.

Vibration Control of a Traveling
M. Saigo
Research Manager, Suspended System Using Wave
Mechanical Engineering Laboratory,
Agency of Industrial Science and Technology,
1-2 Namiki, Tsukuba, Ibaraki 305-8564, Japan
Absorbing Control
This paper describes vibration control of a suspended system using a wave absorbing
K. Tani method. Here, we treat a system that accepts a traveling command. This system is called
Professor, a traveling system. In the previous paper, we treated a system that performs only the
Gifu University, vibration control, where the support of the suspended system moves only for vibration
1-1 Yanagito, Gifu 501-1193, Japan control and eventually settles at the original position. This system is called a nontraveling
e-mail: [email protected] system. In a traveling system, the support moves both for traveling and for vibration
control. We present a new control strategy for these two different aims by applying the
H. Usui vibration control method developed in the previous paper. A traveling multiple-pendulum
Nippon Steel Corporation, system and a traveling wire-and-load system are treated. The wire-and-load system has a
2-6-3 Otemati, Chiyodaku, Tokyo 100-8071, small rigid pendulum between the support and the wire. The vibration control is per-
Japan formed by monitoring this small rigid pendulum. The wire-and-load system is extended to
e-mail: [email protected] a model crane system that has a motor system to roll up and down the suspended mass
like a real crane. The same program with different parameter values controls these three
systems. Both numerical simulation and experiment have been conducted, and the devel-
oped control method has been shown to be quite effective. 关DOI: 10.1115/1.1569515兴
1 Introduction while suppressing the vibration. Both a multiple-pendulum system

and a wire-and-load suspended system are treated. Furthermore,
Recently, traveling-wave control has been studied as an alter-
the method is applied to a model crane system and shown to be
native to mode-based vibration control by several researchers.
useful for the vibration suppressing of a practical crane.
Wave control has several advantages over mode-based vibration
control; it has no control and observation spillovers that may oc-
cur in the vibration control when there are some imperfections in
system modeling or inaccuracy in sensor locations; it has a better 2 Equation of Motion
control performance than vibration control at low frequencies; and 2.1 Multiple-Pendulum System. Figure 1 shows a travel-
it is basically a local control method to which we have paid atten- ing multiple rigid-pendulum system and a traveling wire-and-load
tion in our studies. The last feature means that we can suppress the system. The equations of motion of a traveling multiple-pendulum
vibration of a system using no more than the information about system of n degrees of freedom 共DOF兲 are obtained using the
the dynamic states of the element nearest the actuator. This is Lagrange’s equation of motion. The kinetic energy T k and the
quite advantageous to the system whose parameters are change- potential energy U k of the k-th pendulum are expressed as
冉冊冉冊
able during control operations.
n 2 n
The studies of traveling-wave control include Vaughan 关1兴, Von
Flotow 关2,3兴, Millar 关4兴, Mace 关5兴, Fujii 关6兴, Tanaka 关7兴 and Ut- T k ⫽m k ẏ 0 ⫹ 兺
j⫽k⫹1
l j ␪˙ j /2⫹I k ␪˙ 2k /2⫹m k ẏ 0 ⫹ 兺
j⫽k⫹1
l j ␪˙ j ␪˙ k h k
sumi 关8兴. These studies of traveling-wave control of elastic beams
or strings have used the theoretical solution expressed in the form
of traveling wave and derived the nonreflecting condition of
waves at the control point. On the other hand, O’Connor 关9兴
再
U k ⫽⫺m k g h k 共 1⫺ ␪ 2k /2兲 ⫹ 兺
n
j⫽k⫹1
l j 共 1⫺ ␪ 2j /2兲冎
treated a mass-and-spring system and presented a method of the where ␪ k is the angle of the k-th pendulum numbered from the
wave-absorption in a discrete vibration system. In these studies, free end and assumed to be so small that the terms having powers
few attempts have been made to apply the wave control strategy higher than second of ␪ k are negligible, h k is the distance between
to practical vibrating systems to show the above-mentioned ad- the center of gravity of the k-th pendulum and the axis of the k-th
vantages of traveling-wave control. connecting pin, I k is the moment of inertia of the k-th pendulum
In the previous paper 关10兴, we presented a new practical wave about the axis of the k-th connecting pin, m k is the mass of k-th
control strategy that is easy to build in a control computer with pendulum, l k is the distance between the axes of the kth and (k
on-line calculation of the imaginary wave-propagating system. We ⫺1)-th connecting pins, and y 0 is the position of the support of
applied the presented method to the vibration suppression of a the pendulum system. The Lagrangian L⫽ 兺 k⫽1 n
(T k ⫺U k ) gives
multiple-pendulum system and showed the effectiveness of the the equations of motion of k-th pendulum as
冉冊兺冉兺冊冉兺
method experimentally. There, vibration control by controlling the
k⫺1 k⫺1 j⫺1 k⫺1
support movement was addressed and no support movement for
traveling toward a given target position was considered.
In the present paper, the wave control method presented in the
I k ⫹l 2k 兺
i⫽1
m i ␪¨ k ⫹l k
j⫽1
lj
i⫽1
m i ⫹h j m j ␪¨ j ⫹ l k
i⫽1
mi
冊兺冉兺冊
previous paper is expanded to consider the combination of travel- n k⫺1
ing control and vibration control. In this case, the control system ⫹h k m k 共 l i ␪¨ i 兲 ⫹g l k m i ⫹h k m k ␪ k ⫽0 (1)
has to achieve a system displacement to a desired target position i⫽k⫹1 i⫽1

From the equations obtained by replacing k by k⫹1 and by k
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2000; ⫺1 in Eq. 共1兲, as well as Eq. 共1兲 itself, the following equation is
Revised Jan. 2002. Associate Editor: R. L. Clark. obtained.
3 Control Strategy
The concept of our vibration control is to connect the system
whose vibration should be suppressed to a virtually infinite system
that can absorb vibration energy endlessly. In the previous paper,
we have presented a control strategy in which the real pendulum
system is suspended by the imaginary energy-absorbing multiple-
pendulum system whose dynamics is simulated by on-line com-
putation. Since we have to use a finite-DOF energy absorbing
system in practice, we introduced initialization methods for the
energy absorbing system. At the time of initialization, the deflec-
tions and velocities of all the imaginary pendulums other than the
lowest are set to zero, and the deflection and velocity of the lowest
are set to fit the present position and velocity of the support of the
real system.
In this paper, we treat vibration control of a suspended system
that accepts a traveling command. The traveling command is
given in term of the acceleration of the suspended system ÿ c as a
function of time. Two types of imaginary multiple-pendulum sys-
tem are possible as shown in Fig. 2. One is the nontraveling
Fig. 1 Traveling multiple-pendulum system and wire-and-load imaginary system 共NTIS, Fig. 2共a兲兲 and the other the traveling
system imaginary system 共TIS, Fig. 2共b兲兲. The NTIS does not accept the
traveling command ÿ c and thus its algorithm is the same as that of
the vibration control of the non-traveling system treated in the
previous paper. The vibration control calculated using NTIS is
⫺ 兵 q 1 p 1 ⫹J 1 / 共 r 1 m 1 兲其 ␪¨ 1 ⫹q 2 p 1 ␪¨ 2 ⫹g 共 m 1 ⫹r 2 m 2 兲 p 1 共 ⫺ ␪ 1 ⫹ ␪ 2 兲 added with the traveling command to produce position control of
the support of the real system. The influence of the traveling is
⫽0 actually regarded as a disturbance appearing on the uppermost
pendulum of the real system. With TIS, on the other hand, the
q k⫺1 p k⫺1 ␪¨ k⫺1 ⫺ 关 q k p k ⫹ 兵 J k ⫹l k 共 1⫺r k⫺1 兲 m k⫺1 其 p k⫺1 兴 ␪¨ k support of the imaginary system is moved according to the trav-
再冉兺冊
eling command ÿ c and the movement is propagated through the
k⫺2
imaginary system down to the real system. For both cases in Fig.
⫹q k⫹1 p k ␪¨ k⫹1 ⫹g m j ⫹r k⫺1 m k⫺1 p k⫺1 ␪ k⫺1 2, the value of x 0 ⫽ 兺 k⫽1
n
l k ␸ k is the distance between the horizon-
j⫽1
tal positions of the support and of the lowest end of the imaginary
⫺ 冉兺k⫺1
j⫽1
冊
m j ⫹r k m k 共 p k ⫹p k⫺1 兲 ␪ k ⫹ 冉兺 k
j⫽1
mj
system. The initialization is performed based on this value.
Through numerical simulations we have found the control per-
formance using NTIS is better than that using TIS. It is considered
冊冎
that the initialization using TIS brings about a larger initial deflec-
tion and velocity to the lowest imaginary pendulum because the
⫹r k⫹1 m k⫹1 p k ␪ k⫹1 ⫽0 共 k⫽1,n 兲
acceleration of the imaginary system due to ÿ c produces a larger
(2) value of x 0 . This causes the vibration energy flow back into the
controlled real system. Therefore, we use NTIS in the following
q n⫺1 p n⫺1 ␪¨ n⫺1 ⫺ 兵 J n ⫹l n 共 1⫺r n⫺1 兲 m n⫺1 其 p n⫺1 ␪¨ n
work.
⫹gp n⫺1 再冉兺 n⫺2
j⫽1
冊冉兺
m j ⫹r n⫺1 m n⫺1 ␪ n⫺1 ⫺
n⫺1
j⫽1
mj
冊冎
⫹r n m n ␪ n ⫽⫺ÿ 0
r k ⫽h k /l k , J k ⫽I k /l k , p k ⫽1/兵共 1⫺r k 兲 m k ⫹r k⫹1 m k⫹1 其 ,
q k ⫽J k ⫺h k m k
From the above equation, we see the term of the support move-
ment appears explicitly only in the equation of the uppermost
pendulum.
2.2 Wire-and-Load System. The wire-and-load system
shown in Fig. 1 has a small rigid pendulum between the support
and the wire, and there is a load at the bottom of the wire. The
wire length is fixed. Assuming the equation of the wire is ex-
pressed by that of a dangling string and applying the finite differ-
ence method to the equation of motion, we obtain a system of
equations similar to that of a multiple simple-pendulum system
共see Appendix兲. In the following numerical simulation, we will
treat the wire-and-load system as a non-homogenous multiple- Fig. 2 N-DOF imaginary system for traveling pendulum sys-
pendulum system consisting of the uppermost rigid pendulum and tem „ ÿ c : traveling command, ẍ 0 : vibration control… „a… Non-
a large-DOF series of simple pendulums, among them the lowest traveling Imaginary System „NTIS…, „b… Traveling Imaginary
having a mass equal to that of the load. System „TIS…

geous first to move the system near the destination with less vi-
bration and then to control the final position accurately. It is an
easy job for the control computer, given the present and target
positions, to cancel the error of the final position.
When the real pendulum system is connected to the imaginary
system, the acceleration of the lowest end of the imaginary system
is represented as
共 i 2 /l⫺h 兲 ␪¨ ⫹g 共 ␮ 0 ⫹h/l 兲 ␪ ⫺g 共 1⫹ ␮ 0 ⫹ ␮ 兲 ␸
ẍ 0 ⫽ (3)
1⫺h/l⫹ ␮
where ␪ is the angle of the uppermost rigid pendulum, ␸ is the
angle of the lowest imaginary system pendulum, ␮ 0 is the ratio of
the mass of the total real system to that of the uppermost rigid
pendulum, ␮ is the ratio of the mass of an imaginary system
pendulum to that of the uppermost rigid pendulum, and l, h and i 2
are the length, the distance between the supporting point and cen-
ter of gravity, and the square of radius of gyration on the support-
ing point, of the uppermost rigid pendulum, respectively. As stated
for the non-traveling case investigated in the previous paper, ẍ 0 is
used here also as the control of the movement of the support of
the real pendulum system for vibration suppression. A homog-
enous multiple simple-pendulum system is used as the imaginary
system in Eq. 共3兲 for simplicity same as in the previous paper. A
measured value of ␪, the numerically approximated value of ␪¨ and
the computed value of ␸ from the imaginary system give the
vibration control ẍ 0 . Note that Eq. 共3兲 includes no dimensional
parameters of the suspended pendulums except those of the up-
permost one. So, it can give vibration control for the multiple
rigid-pendulum system as well as for the wire-and-load system
with an uppermost rigid pendulum. In other words, the control
does not depend on the length of the wire.
The parameter ␮ is introduced in Eq. 共3兲 for the adjustment of
the performance of the control system. For a large value of ␮, the
wave propagation in the imaginary system becomes slower and
the control gain for the vibration suppression smaller. A smaller
gain makes the control of the system more stable but less effec-
tive. In the experiment, we obtained the practical values for ␮
considering the system stability and the limitations of the actual
DC servo motor system. Thus, the wave propagating characteris-
tics in the imaginary system and the control performance of the
vibration suppression can be designed by changing the values of
␮. The parameter ␬, the ratio of the length of the imaginary sys-
tem pendulum to that of the uppermost rigid pendulum, can also
change the wave propagating characteristics in the imaginary sys-
tem.
4 Experiment
4.1 Experiments of the Multiple-Pendulum System and
the Wire-and-Load System
Fig. 3 Three types of initializing method for NTIS. „a… PI 4.1.1 Experimental Apparatus. Figure 4 shows a schematic
method: initialization when x 0 Ä0, „b… VI method: initialization diagram of the experimental apparatus. The uppermost pendulum
when ẋ 0 Ä0, „c… VI’ method: initialization when ẋ 0 Ä0 with sup- is connected to and supported by the nut of the ball-screw. The
port shift
ball-screw is driven by a 350W DC servomotor to generate the
horizontal movement of the support of pendulum. The DC servo-
motor is a velocity feedback type with an integrated tacho-
Three types of initializing timing are investigated for NTIS as generator. A rotary encoder is attached to the uppermost pendu-
shown in Fig. 3. The cases 共a兲 and 共b兲 in Fig. 3 are the same as lum.
those used in the previous paper, while 共c兲 in Fig. 3 is a new The multiple-pendulum system is made of three same-size alu-
method presented in this paper. In the case 共a兲, the PI method, minum plates connected serially by pins allowing free rotation.
initialization is made when x 0 ⫽0(⫽ 兺 k⫽1
n
l k ␸ k ), and in the case The width and thickness of each plate are 40 mm and 10 mm,
共b兲, the VI method, initialization is made when ẋ 0 ⫽0. In the case respectively. The distance between the center axes of the connect-
共c兲, the VI’ method, initialization is made when ẋ 0 ⫽0 as in the ing pins of each plate is 300 mm. The adjacent plates can be fixed
case 共b兲 but also has a position shift of the support of imaginary rigidly with braces to form a pendulum system having less than
system. The vibration control performance in this case is better three degrees of freedom.
than in the cases 共a兲 and 共b兲, except that it may cause an error in The wire-and-load system is made up of a rigid pendulum, a
the final support position of the real system. It may be advanta- wire and a load. The length, width and thickness of the rigid

Fig. 4 Experimental apparatus „wire-and-load system…
pendulum are 60 mm, 40 mm and 15 mm, respectively. The di-

ameter of the wire is 1 mm. The length of the wire and the weight
of the load can be changed.
The computation of the control is conducted by a DSP
共TMS320C30兲 for the rigid-pendulum system and a personal com-
puter with 200 MHz CPU for the wire-and-load system. The sam-
pling period of A/D conversion is 0.1 ms for the rigid-pendulum
system and 2 ms for the wire-and-load system. A 10-DOF system
of simple pendulums has been used as the imaginary system. Fig. 5 Experimental results of 3 DOF rigid-pendulum system
„␮Ä3, ␬Ä1…. „a… No control, „b… Control with PI method, „c… Con-
4.1.2 Experimental Results. The following system move- trol with VI’ method
ment pattern is used as the traveling command in the experiments;
the acceleration ÿ c is 4.26 m/s2 for the time period between 0 s
and 0.0352 s and ⫺0.01883 m/s2 between 0.0352 s and 8 s. Using pendulum system and the vibration of the wire is practically neg-
this acceleration pattern the pendulum system should travel the ligible. Figures 8 show the effect of the wire length l and the load
distance of 0.60 m in 8 seconds. weight w on the control performance of the VI’ method when
„a… The case of the multiple-pendulum system ␮⫽1. We can see the control performance is independent of the
Figures 5 show the effects of the different initializing methods, change of the wire length and the load weight. Besides, the con-
the PI and VI’ methods, for the 3-DOF traveling rigid-pendulum trol performances in Fig. 8 for ␮⫽1 are better than that in Fig.
system with ␮⫽3. In these figures, the curve rising to the right- 7共c兲 for ␮⫽3, which confirms that the quick wave propagation in
hand side is the position of the support and the vibration wave- the imaginary system brings about a better performance in vibra-
form is the angle of the uppermost pendulum. From these figures, tion suppression while the system is stable.
we can confirm that our method is effective for a traveling pen-
dulum system as well as for a nontraveling system. Figures 6
show the control performance of the PI method and the VI’
method on the 1-DOF pendulum system 共three pendulums are
connected rigidly兲. Both initializing methods have excellent vibra-
tion suppressing effects. Similarly, the control performance on the
non-homogeneous 2-DOF pendulum system 共the lowest and the
middle pendulums are connected rigidly兲 is well confirmed 共fig-
ures are not shown兲. The control program for the 2-DOF system is
the same as that for the 3-DOF system, for the uppermost pendu-
lum and the total mass of the system are the same.
„b… The case of the wire-and-load system
Figures 7 show the effects of the different initializing methods,
the PI and VI’ methods, for the traveling wire-and-load system
with the wire length l⫽0.5 m, the load weight w⫽12.3 N and
␮⫽3. In these figures, the curve rising to the right-hand side is the
position of the support and the vibration waveform is the angle of
the uppermost pendulum. From these figures we can see that the
VI’ method has quite an excellent damping performance 共Fig.
7共c兲兲. The PI method is accurate in positioning the pendulum sys-
tem at the traveling destination, but the vibration control perfor-
mance is not so good. The performance of the VI method is not so
good 共figure is not shown兲. The vibration waveform shown in Fig.
7共c兲 resembles well that of the 1-DOF rigid pendulum shown in Fig. 6 Experimental results of 1 DOF rigid-pendulum system
Fig. 6共b兲. This means the dynamic characteristics of the experi- „␮Ä1, ␬Ä1…. „a… Control with PI method, „b… Control with VI’
mental wire-and-load system is similar to that of the 1-DOF rigid- method

Fig. 9 Experimental results of wire-and-load system for final
positioning „ l Ä0.3 m, w Ä12.3 N, ␮Ä10, ␬Ä10…. „a… Interruption
of traveling command „overrun case…, „b… Position correction
after traveling command
process can position the support accurately at the target position

with small vibration. Figure 10 shows a non-traveling case where
a combination of initializing methods is used for a large initial
amplitude. This process combines the PI method that absorbs
Fig. 7 Experimental results of wire-and-load system „ l quickly large amplitudes with the VI method that absorbs quickly
Ä0.5 m, w Ä12.3 N, ␮Ä3, ␬Ä10…. „a… No control, „b… Control with
PI method, „c… Control with VI’ method
small amplitudes. The PI method was used between 0 s and 6 s
and the VI method was used afterwards. This, combinations of
different initialization methods can be used effectively to adapt to
various situations.
Figures 9 show the results of the operation in which final posi- In order to understand the characteristics of the initializing
tioning accuracy is concerned when the VI’ method is used for methods obtained in the experiments, several numerical simula-
vibration control. As stated earlier, this method can produce a tions have been conducted. Figures 11 show the simulation results
positioning error, and some measures should be taken when accu- for the wire-and-load system corresponding to Fig. 7共b兲 and 共c兲. In
rate positioning at the target is needed. Figure 9共a兲 is an overrun the simulation, dry friction is assumed at each connecting pin of
case where the vibration control is switched to the VI method and the pendulum to represent the wire. The ratio of the mass of the
the traveling command is suppressed the instant the support wire to that of the uppermost pendulum is assumed to be 0.0001.
reaches the exact target position. Figure 9共b兲 is a case where the Small wire vibrations occur at the starting period in the simulation
positioning correction to the exact target position is carried out results, which are not observed in the experiments. In Fig. 11 we
using the VI method after the traveling command is over. Either show the waveform of the swing angle of the load in stead of that
of the uppermost pendulum because the latter is affected by the
small wire vibration. There is little difference between the results
in Fig. 11 and Fig. 7. From these figures, we can confirm that the
experiments have been performed successfully and that the angle
of the uppermost rigid pendulum is virtually equal to the swing
angle of the load.
As is easily understood, the vibration of the load would propa-
gate up to the uppermost rigid pendulum. So, our wave-absorbing
system, which attends the uppermost rigid pendulum, is eventu-
ally effective in suppressing the vibration of the load.
Fig. 8 Experimental results of wire-and-load system for differ- Fig. 10 Experimental result of nontraveling wire-and-load sys-
ent wire length and load weight with VI’ method „␮Ä1,␬Ä10…. „a… tem for large amplitude with PI and VI methods „ l Ä0.3 m, w
l Ä0.3 m, w Ä12.3 N „b… l Ä0.9 m, w Ä31.9 N Ä12.3 N, ␮Ä10, ␬Ä10….

Fig. 11 Numerical results of wire-and-load system „ l Ä0.5 m,
w Ä12.3 N, ␮Ä3, ␬Ä10…. „a… Control with PI method, „b… Control
with VI’ method
The VI’ method shows best vibration suppression among the

three initializing methods. The PI and VI methods have inferior
vibration control performances, especially for a small value of ␮.
Therefore, it is suitable to use a relatively large value of ␮
depressing the vibration control performance for the PI or VI
methods.
4.2 Experiment of a Model Crane
4.2.1 Experimental Apparatus. Figure 12 shows the experi- Fig. 13 Experimental results of crane system for raising load
mental model of a crane, which has a load suspended by a wire „total weightÄ17.6N…. „a… No control, „b… Control with PI method
and a pulley. One end of the wire is fixed to the motor shaft for „␮Ä10, ␬Ä10…, „c… Control with VI’ method „␮Ä1, ␬Ä10…
winding and the other end is fixed to a small rigid pendulum that
is attached to the nut of the ball-screw with free rotation. The
distance between the axis of the rigid pendulum and the wire
return on the returning pulley is equal to the diameter of the pul-
ley. Then, the angle of the rigid pendulum is practically equal to tionary direction of the pendulum is always vertical and the vibra-
the swing angle of the load independent of its height when the tion suppression strategy for the wire-and-load system is applied
small vibrations of the wire can be ignored. This means the sta- directly by monitoring one half of the load suspension system.
Our experiment has shown that it is possible to neglect the small
wire vibration in the wire-and-load system as well as in most
practical crane systems. The velocity pattern used of winding-up
and rewinding-down is 0.1 m/s and ⫺0.1 m/s, respectively. The
load moves between the vertical positions of 0.9 m and 0.3 m
during the time period between 0 s and 6 s.
4.2.1 Experimental Results. Figures 13 show the experimen-
tal results for the case of winding up the load; 共a兲 in Fig. 13 is the
case where no vibration control is used, 共b兲 is the case with wave
control in the PI method, and 共c兲 is the case with wave control in
the VI’ method. Figures 14 show the cases of rewinding down the
load with controls similar to those in Fig. 13. The system move-
ment pattern is the same as in the case for the wire-and-load
system. In Fig. 13 we can see the amplitude of the vibration in
winding-up without control becomes larger as the wire length be-
comes shorter, due to the instability in winding-up of a suspended
load. The contrast in these figures demonstrates the effectiveness
of the stabilization using the wave-absorbing control. In addition,
the wave control method presented has shown an excellent control
performance regardless of the wire length. Figures 13共b兲 and
14共b兲 with the PI method for a large value of ␮ show relatively
good results in the final state of the load, that is, accurate final
position and small vibration. Thus, we can use the PI method to
position the system accurately at the target position if its vibration
Fig. 12 Model crane system suppression performance is acceptable. Even if the VI’ method is

␮ ⫽ ratio of the mass of an imaginary system pendu-
lum to that of the uppermost rigid pendulum
␬ ⫽ ratio of the length of an imaginary system pendu-
lum to that of the uppermost rigid pendulum
␪ ⫽ angle of the uppermost rigid pendulum
␸ ⫽ angle of the lowest imaginary system pendulum
␪ k ⫽ angle of the k-th rigid pendulum numbered from
the free end
h k ⫽ distance between the center of gravity of k-th
rigid pendulum and the axis of k-th connecting
pin
I k ⫽ moment of inertia of k-th rigid pendulum about
the axis of k-th connecting pin
m k ⫽ mass of k-th rigid pendulum
l k ⫽ distance between the axes of the k-th and
(k⫺1)-th connecting pins of the rigid pendulum
␸ k ⫽ angle of the k-th imaginary system pendulum
numbered from the lowest end
x 0 ⫽ 兺 k⫽1
n
l k ␸ k 共n: degrees of freedom of imaginary
system兲
y 0 ⫽ horizontal displacement of the support 共imaginary
or real兲
y c ⫽ traveling command
␩ ⫽ deflection of the wire
␩ i ⫽ deflection of the i-th finite element of the wire
z ⫽ coordinates of the wire measured from the lower
共free兲 end
Appendix
We assume the equation of motion of the wire is expressed by
that of a dangling string. By balancing the horizontal component
of forces on an infinitely small element z⬃(z⫹dz), the following
equation is obtained as
Fig. 14 Experimental results of crane system for lowering load
„total weightÄ17.6 N…. „a… No control, „b… Control with PI method
„␮Ä10, ␬Ä10…, „c… Control with VI’ method „␮Ä1, ␬Ä10…
冉冊
M
␳
⫹z
⳵ 2␩ ⳵ ␩ 1 ⳵ 2␩
⳵z2
⫹ ⫺
⳵z g ⳵t2
⫽0 (A1)
where ␩ is the lateral deflection of wire, z is the coordinate mea-

sured from the lowest end, ␳ is the mass of string per unit length
used, the final position errors are not very significant. Figures 13 and M is the mass of load.
and 14 have shown our wave absorbing method is useful for the Appling the finite difference method to the above equation us-
actual crane system. ing the following approximations,
5 Concluding Remarks ⳵ 2␩ ␩ i⫹1 ⫺2 ␩ i ⫹ ␩ i⫺1 ⳵ ␩ ␩ i ⫺ ␩ i⫺1
⫽ , ⫽ , z⫽i⌬z
In this paper we propose a wave control method using a non- ⳵z 2
⌬z 2 ⳵z ⌬z
traveling imaginary multiple-pendulum system applied to vibra-
tion control of the traveling suspended system. We have shown we obtain
that the method presented, using little information about the sus-
M
pended system states, is quite effective for a traveling suspended 共 ␩ ⫺2 ␩ i ⫹ ␩ i⫹1 兲 ⫹ 共 i⫺1 兲 ␩ i⫺1 ⫺ 共 2i⫺1 兲 ␩ i ⫹i ␩ i⫹1
system whose dynamics are changeable during operation. The ini- ␳ ⌬z i⫺1
tializing methods with and without shifting the support of the
⌬z
imaginary system have their respective merits for the application. ⫽ ␩¨ (A2)
The former realizes accurate final positioning while the latter re- g i
alizes excellent vibration suppression. Combination of methods where ␩ i is the lateral deflection of i-th mesh point numbered
can produce excellent overall performances. This control method from the lowest end of the wire and ⌬z is the finite difference
can readily be applied to real crane systems that have a moving mesh.
pulley and parallel wiring. From Eq. 共A2兲 and the equation obtained by replacing i by i
⫹1 in Eq. 共A2兲, we obtain the following equation of motion by
Nomenclature substituting ␪ i ⫽( ␩ i ⫺ ␩ i⫹1 )/⌬z,
g ⫽ Gravitational acceleration
⌬z ¨ M
( ˙ ), ( ¨ ) ⫽ first and second order differentiation with respect ␪ i⫽ 共 ␪ ⫺2 ␪ i ⫹ ␪ i⫺1 兲 ⫹ 共 i⫺1 兲 ␪ i ⫺2i ␪ i ⫹ 共 i⫹1 兲 ␪ i⫹1
* * to time g ␳ ⌬z i⫹1
i 2 ⫽ square of radius of gyration on the supporting (A3)
point of the uppermost rigid pendulum The above equation is the same as the equation of motion of a
l ⫽ length of the uppermost rigid pendulum multiple simple-pendulum system obtained from Eq. 共2兲 if we
h ⫽ distance between the supporting point and the regard ⌬z as the length of the simple pendulum and M as the
center of gravity of the uppermost rigid pendulum additional mass to the lowest pendulum. This means that we can
␮ 0 ⫽ ratio of the mass of the total real system to that of simulate the dynamics of wire in the form of a system of simple
the uppermost rigid pendulum pendulums of length ⌬z.

The influence of a wire on the total dynamics of the pendulum 关4兴 Miller, D. W., and von Flotow, A. H., 1989, ‘‘A Travelling Wave Approach to
Power Flow in Structural Networks,’’ J. Sound Vib., 128, pp. 145–162.
system is quite small as can be seen in the experimental results.
关5兴 Mace, B. R., 1984, ‘‘Wave Reflection and Transmission in Beams,’’ J. Sound
The exact dynamical formulation for a wire is not so important in Vib., 97, pp. 237–246.
our study 共M is much greater than ␳ ⌬z). So, we will not try to 关6兴 Fujii, H., and Ohtsuka, T., 1992, ‘‘Experiment of a Noncollocated Controller
develop a more accurate mathematical model of the wire in this for Wave Cancellation,’’ AIAA J., 15共3兲, pp. 93–97.
paper. 关7兴 Tanaka, N., and Kikushima, Y., 1992, ‘‘Active Wave Control of a Flexible
Beam,’’ JSME Int. J., Ser. III, 35共1兲, pp. 236 –244.
关8兴 Utsumi, M., 1999, ‘‘Analytical Implementation of Wave-Absorbing Control
References for Flexible Beams Using Synchronization Condition,’’ ASME J. Vibr. Acoust.,
关1兴 Vaughan, D. R., 1968, ‘‘Application of Distributed Parameter Concepts to 121, pp. 468 – 475.
Dynamic Analysis and Control of Bending Vibrations,’’ ASME J. Basic Eng., 关9兴 O’Connor, W., and Lang, D., 1998, ‘‘Position Control of Flexible Robot Arms
90, pp. 157–166. Using Mechanical Waves,’’ ASME J. Dyn. Syst., Meas., Control, 120, pp.
关2兴 Von Flotow, A. H., 1986, ‘‘Traveling Wave Control for Large Spacecraft Struc- 334 –339.
tures,’’ AIAA J., 9, pp. 462– 468. 关10兴 Saigo, M., Tanaka, N., and Tani, K., 1998, ‘‘An Approach to Vibration Control
关3兴 Von Flotow, A. H., 1986, ‘‘Disturbance Propagation in Structural Networks,’’ of Multiple-Pendulum System by Wave Absorption,’’ ASME J. Vibr. Acoust.,
J. Sound Vib., 106, pp. 433– 450. 121, pp. 524 –533.

Dynamic Stiffness Formulation
and Its Application for a
J. R. Banerjee Combined Beam and a Two
Professor of Structural Dynamics,
School of Engineering
and Mathematical Sciences,
Degree-of-Freedom System
City University, Northampton Square, London
EC1V OHB
This paper is concerned with the dynamic stiffness formulation and its application for a
Bernoulli-Euler beam carrying a two degree-of-freedom spring-mass system. The effect of
a two degree-of-freedom system kinematically connected to the beam is represented ex-
actly by replacing it with equivalent stiffness coefficients, which are added to the appro-
priate stiffness coefficients of the bare beam. Numerical examples whose results are ob-
tained by applying the Wittrick-Williams algorithm to the total dynamic stiffness matrix
are given and compared with published results. Applications of the theory include the free
vibration analysis of frameworks carrying two degree-of-freedom spring-mass
systems. 关DOI: 10.1115/1.1569943兴
1 Introduction trated or lumped mass/inertia, or a spring, to be added at a node or

joint of a structure 关38兴, without compromising the accuracy, and
There are many important engineering structures that can be
hence retaining the exactness of results. However, a difficulty
modelled as beams carrying one, two or multi degree-of-freedom
arises when the beam is attached to a discrete structural element
spring-mass systems. Examples of such practical applications in- such as a two degree-of-freedom system because the member dy-
clude components of buildings and machine tools, vehicle suspen- namic stiffness matrix, the basic building block in the method of
sions, rotating machinery, accessories of machine structures and such a combined system, is not currently available. This paper
robotics amongst others. Because of these wide-ranging applica- develops the dynamic stiffness matrix of a Bernoulli-Euler beam
tions the vibration behavior of beams carrying discrete structural to which has been attached a two degree-of-freedom spring-mass
elements such as a one, two or multi degree-of-freedom spring- system, which is free to translate and rotate. The resulting eigen-
mass systems has received considerable attention for many years. values of the combined system, which yield the natural frequen-
The research has developed from the simpler problem of the free cies of the combined structure, are determined using the Wittrick-
vibration analysis of beams with lumped masses 关1–9兴 or spring Williams algorithm 关37兴. In order to illustrate the method several
supports 关10–16兴 to that of beams carrying elastically mounted cases have been evaluated and some representative results have
masses 关17–25兴 or a one or two degree of freedom spring-mass been compared with published ones. This paper is thought to be
systems 关26 –31兴. 共For further reading on combined dynamical the first in which the dynamic stiffness matrix of a combination of
systems in general, Refs. 关32兴, 关33兴 are recommended.兲 All these continuous and discrete structural elements has been developed.
investigations have shown that the natural frequencies and mode Although the simple case of the Bernoulli-Euler beam has been
shapes of a beam carrying spring-mass systems may deviate con- considered here as the parent structural member for illustrative
siderably from those of the beam alone. purposes, the procedure is completely general and can easily be
The main drawback of the existing solution procedures for such extended to Timoshenko beams with or without an axial load 关39–
problems is that by and large they concentrate solely on uniform 42兴 carrying a two degree-of-freedom spring-mass system.
and unidirectional beams, and cannot be generally applied to
frameworks. This paper presents an entirely different approach
based on the dynamic stiffness method which is free from such 2 Theory
limitations and yields exact results, some of which can be directly
compared to the solution of the complete governing differential 2.1 Dynamic Stiffness Formulation for a Bernoulli-Euler
equations. It does not seem to be widely recognized that the dy- Beam. The dynamic stiffness matrix of a Bernoulli-Euler beam
namic stiffness method allows an assembly technique 共similar to has been available in the literature 关34,36兴 for many years. Ban-
that used in finite element methods兲 that enables exact free vibra- erjee 关43兴 has given a general formulation that leads to the devel-
tion analysis of either a single structural element or a combination opment of the dynamic stiffness matrix of a structural element.
of structural elements with different orientations. Furthermore, in For a Bernoulli-Euler beam the procedure is simple and is de-
contrast to finite element methods, the results using the dynamic scribed as follows. 共Readers who are not familiar with the subject
stiffness method are not only exact, but also independent of the are referred to the Appendix, which gives details of the dynamic
number of elements used in the analysis and thus offers much stiffness development.兲 In both axial and transverse motion the
better computational efficiency. governing partial differential equations in free vibration are de-
The free vibration analysis of simple beams or frameworks has rived, and the corresponding total differential equations for har-
been solved using the dynamic stiffness method without difficulty monic motion established. These are then integrated in terms of
关34,35兴, because the dynamic stiffness properties of a beam 关36兴 two arbitrary constants for axial motion, and in terms of four more
and the solution technique known as Wittrick-Williams algorithm for the flexural motion. The associated end loads are then obtained
关37兴 are both well established. The method also allows a concen- in terms of the arbitrary constants, so that for the longitudinal
motion, the axial force at each end, as well as the axial displace-
ments are both expressible in terms of the two arbitrary constants.
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002; Eliminating the two constants then gives the relationship between
Revised Dec. 2002. Associate Editor: B. Yang. the two end forces and two end displacements. This embodies the
Fig. 1 Amplitudes of displacements and forces at the ends of
a Bernoulli-Euler Beam in free vibration
Fig. 2 A two degree-of-freedom system kinematically con-

共frequency dependent兲 dynamic stiffness matrix for axial motion nected to a beam element
relating the amplitudes of forces and displacements at the ends of
the freely vibrating Bernoulli-Euler beam.
In a similar way, the shear force and bending moment at each d 1 ⫽W 3 ␭ 3 共 S ␭ C h␭ ⫹C ␭ S h␭ 兲 /⌬, d 2 ⫽W 2 ␭ 2 S ␭ S h␭ /⌬,
end of the beam, together with the bending displacement and ro-
tation at each end, are expressible in terms of four constants of d 3 ⫽W 1 ␭ 共 S ␭ C h␭ ⫺C ␭ S h␭ 兲 /⌬
integration when dealing with the free vibration in flexural mo- (5)
tion. On eliminating these constants, a relationship between the d 4 ⫽⫺W 3 ␭ 共 S ␭ ⫹S h␭ 兲 /⌬,
3
d 5 ⫽W 2 ␭ 2 共 C h␭ ⫺C ␭ 兲 /⌬,
four end loads and the four kinematic conditions at the ends, is d 6 ⫽W 1 ␭ 共 S h␭ ⫺S ␭ 兲 /⌬
established. The resulting 4⫻4 matrix linking the two is the fre-
quency dependent dynamic stiffness matrix in flexural motion. with
The axial and flexural dynamic stiffness properties can now be
compounded into a 6⫻6 dynamic stiffness matrix of the EI EI EI
W 1⫽ , W 2⫽ , W 3⫽ (6)
Bernoulli-Euler beam, relating the amplitudes of the forces to the L L 2
L3
corresponding amplitudes of the displacements at the ends of the
vibrating beam 共see Fig. 1兲. In the usual notation when the beam
parameters EA, EI, m and L represent the extensional rigidity,
bending rigidity, mass per unit length and length of the beam
␭⫽ 冑4 m ␻ 2L 4
EI
(7)
respectively, the dynamic stiffness matrix takes the following C ␭ ⫽cos ␭, S ␭ ⫽sin ␭, C h␭ ⫽cosh ␭, S h␭ ⫽sinh ␭ (8)
form 共see Appendix for details兲
冋册冤冥冋册
and
a1 0 0 a2 0 0 ⌬⫽1⫺C ␭ C h␭ (9)
F x1 U1
0 d1 d2 0 d4 d5
F y1 V1 2.2 Dynamic Stiffness Formulation for a Two Degree-of-
M1 0 d2 d3 0 ⫺d 5 d6 ␪1 Freedom Spring-Mass System Attached to a Beam. Figure 2
⫽ (1) shows a rigid mass M having a mass moment of inertia I ␣ about
F x2 a2 0 0 a1 0 0 U2
F y2 V2 its centroidal axis and mounted at the ends 共nodes兲 of a Bernoulli-
0 d4 ⫺d 5 0 d1 ⫺d 2 Euler beam by means of two linearly elastic springs of spring
M2 ␪2 constant K S1 and K S2 , respectively. The distances of the center of
0 d5 d6 0 ⫺d 2 d3
gravity 共g兲 of the mass from the location of the springs are D 1 and
or D 2 respectively, as shown.
The governing differential equations of motion for the two
degree-of-freedom spring mass system shown in Fig. 2, undergo-
F⫽K␦ (2)
ing free vibration in translation and rotation, are
where F and ␦ are respectively the force and displacement vectors ⳵ 2v M
and K is the frequency dependent dynamic stiffness matrix whose M ⫺ f k1 ⫺ f k2 ⫽0 (10)
elements k(i, j) (i⫽1,2 . . . 6; j⫽1,2, . . . 6). are given by a 1 , a 2 ⳵t2
and d 1 ⫺d 6 .
In Eq. 共1兲, F x1 , F y1 , M 1 and U 1 , V 1 and ␪ 1 are the amplitudes ⳵ 2␣ M
I␣ ⫹ f k1 D 1 ⫺ f k2 D 2 ⫽0 (11)
of the forces and displacements at end 1 and F x2 , F y2 , M 2 and ⳵t2
U 2 , V 2 and ␪ 2 are the corresponding amplitudes of the forces and
displacements at end 2, respectively 共see Fig. 1兲. The elements of where v M and ␣ M are the transverse displacement and angular
the dynamic stiffness matrix a 1 , a 2 and d 1 ⫺d 6 are as follows rotation of the rigid mass, t is time, and f k1 and f k2 are the inter-
共see Appendix for details兲. active spring forces between the spring-mass system and the beam
at the two connecting points at nodes 1 and 2 respectively 共see
EA EA Fig. 2兲. These forces are given by
a 1⫽ ␮ cot ␮ , a 2 ⫽⫺ ␮ cos ec␮ (3) f k1 ⫽K S1 共 v 1 ⫺ v M ⫹D 1 ␣ M 兲 (12)
L L
f k2 ⫽K S2 共 v 2 ⫺ v M ⫺D 2 ␣ M 兲 (13)
with
where v 1 and v 2 are the transverse bending displacements of the
␮⫽ 冑 m ␻ 2L 2
EA
(4)
beam at nodes 1 and 2 respectively.
If harmonic oscillation with circular 共or angular兲 frequency ␻ is
assumed, then
and v M 共 x,t 兲 ⫽V M e i ␻ t , ␣ M 共 x,t 兲 ⫽A M e i ␻ t ,

冋册冤冥
v 1 共 x,t 兲 ⫽V 1 e i ␻ t , v 2 共 x,t 兲 ⫽V 2 e i ␻ t (14) a1 0 0 a2 0 0
F x1
and 0 d 1 ⫹e 1 d2 0 d 4 ⫹e 2 d5
F y1 ⫹F k1
f k1 共 x,t 兲 ⫽F k1 e i ␻ t , f k2 共 x,t 兲 ⫽F k2 e i ␻ t (15) M1 0 d2 d3 0 ⫺d 5 d6
⫽
where V M , A M , V 1 , V 2 , F k1 and F k2 are the amplitudes of free F x2 a2 0 0 a1 0 0
vibration. F y2 ⫹F k2 d 4 ⫹e 2
0 ⫺d 5 0 d 1 ⫹e 3 ⫺d 2
Substituting Eqs. 共14兲 and 共15兲 into Eqs. 共10兲–共13兲 gives
冋册
M2
0 d5 d6 0 ⫺d 2 d3
共 ⫺M ␻ 2 ⫹K S1 ⫹K S2 兲 V M ⫹ 共 K S2 D 2 ⫺K S1 D 1 兲 A M
U1
⫺K S1 V 1 ⫺K S2 V 2 ⫽0 (16) V1
共 ⫺K S1 D 1 ⫹K S2 D 2 兲 V M ⫹ 兵 ⫺I ␣ ␻ 2 ⫹ 共 K S1 D 21 ⫹K S2 D 22 兲其 A M ␪1
⫻ (27)
U2
⫹K S1 D 1 V 1 ⫺K S2 D 2 V 2 ⫽0 (17)
V2
Equations 共16兲 and 共17兲 are solved for V M and A M to give ␪1
V M ⫽ 关 K S1 兵 K S2 D 2 共 D 1 ⫹D 2 兲 ⫺I ␣ ␻ 2 其 V 1 or
⫹K S2 兵 K S1 D 1 共 D 1 ⫹D 2 兲 ⫺I ␣ ␻ 2 其 V 2 兴 /Z (18) F⫽K␦ (28)
A M ⫽ 关 K S1 兵 D 1 M ␻ 2 ⫺K S2 共 D 1 ⫹D 2 兲其 V 1 where K, F, and ␦ are the modified dynamic stiffness matrix, and
force and displacement vectors respectively.
⫹K S2 兵 K S1 共 D 1 ⫹D 2 兲 ⫺D 2 M ␻ 2 其 V 2 兴 /Z (19)
where 3 Application of the Dynamic Stiffness Matrix
Z⫽M I ␣ ␻ 4 ⫺ ␻ 2 兵 K S1 共 I ␣ ⫹D 21 M 兲 ⫹K S2 共 I ␣ ⫹D 22 M 兲其 The dynamic stiffness matrix of Eqs. 共27兲–共28兲 can now be
used to compute the natural frequencies and mode shapes of either
⫹K S1 K S2 共 D 1 ⫹D 2 兲 2 (20) a single beam with various end conditions or an assembly of
beams, for example a plane or space frame, carrying one or more
From Eqs. 共12兲 and 共13兲 the spring force amplitudes are given by two degree-of-freedom spring-mass systems. When several ele-
F k1 ⫽K S1 共 V 1 ⫺V M ⫹D 1 A M 兲 (21) ments are to be used the overall dynamic stiffness matrix of the
complete structure must be assembled. The associated natural fre-
F k2 ⫽K S2 共 V 2 ⫺V M ⫺D 2 A M 兲 (22) quencies and mode shapes are then extracted using the Wittrick-
With the expressions for V M and A M given by Eqs. 共18兲 and 共19兲, Williams algorithm, see Refs. 关37兴 and 关34,35兴. The algorithm
Eqs. 共21兲 and 共22兲 can now be written in terms of V 1 and V 2 only, guarantees that no natural frequency and its associated mode are
in the form missed. This is, of course, not possible in the conventional finite
冋册冋册冋册
element method. Interested readers will fill the procedure detailed
F k1 e1 e2 V1 in Refs. 关34,35,37兴. Note that the algorithm requires information
⫽ (23) about the clamped-clamped natural frequencies of all elements
F k2 e2 e3 V2
关34,35,37兴 within the structure to provide exact results. For the
where, after some reduction, e 1 , e 2 and e 3 are given by present problem the determination of the clamped-clamped natu-
ral frequencies of the structural element shown in Fig. 2 will
e 1 ⫽K S1 ␻ 2 兵 M I ␣ ␻ 2 ⫺K S2 共 I ␣ ⫹D 22 M 兲其 /Z (24)
involve the calculation of natural frequencies of the Bernoulli-
e 2 ⫽K S1 K S2 ␻ 共 I ␣ ⫺D 1 D 2 M 兲 /Z
2
(25) Euler beam as well as those of the two degree-of-freedom system
when both ends 共nodes 1 and 2兲 are built-in. The frequency equa-
e 3 ⫽K S2 ␻ 兵 M I ␣ ␻
2 2
⫺K S1 共 I ␣ ⫹D 21 M 兲其 /Z (26) tion for a Bernoulli-Euler beam with clamped-clamped end con-
ditions can be derived using standard procedure 关44兴 whereas for
Equation 共23兲 provides the required dynamic stiffness matrix of
the two degree-of-freedom system connected to the Bernoulli-
the two degree-of-freedom system 共see Fig. 2兲 where e 1 , e 2 and
Euler beam shown in Fig. 2, the corresponding clamped-clamped
e 3 are the equivalent stiffnesses linking forces which will be trans-
natural frequencies can be obtained by substituting Z⫽0 of Eq.
mitted to the beam at nodes 1 and 2 to the associated displace-
共20兲. In order to ensure that no natural frequencies of the structure
ments V 1 and V 2 . The stiffnesses account for the attached two
are missed the actual requirement for applying the Wittrick-
degree-of-freedom system properly, and no approximation is in-
Williams algorithm 关37兴 is to identify how many such clamped-
volved in their derivation.
clamped natural frequencies exist below an arbitrarily chosen trial
The above analysis enables a two degree-of-freedom spring
frequency rather than calculating them precisely 关34,35,43兴.
mass system to be represented exactly by essentially a pair of
springs, loading the beam at nodes 1 and 2 with stiffnesses vary-
ing dynamically according to Eq. 共23兲. 4 Numerical Results and Discussion
To illustrate the above theory two examples are given. The first
2.3 Dynamic Stiffness Matrix of the Combined Bernoulli- is taken from Ref. 关28兴 in which a beam clamped at both ends
Euler Beam and the Two Degree-of-Freedom Spring-Mass carries a two degree-of-freedom spring-mass system as shown in
System. The dynamic stiffness matrix of the combined Fig. 3. The data used in the analysis are as follows:
Bernoulli-Euler beam and the two degree-of-freedom system can
EA⫽3.0⫻1010 N, EI⫽6.25⫻108 Nm2 ,
now be obtained by incorporating the dynamic stiffness force-
displacement relationship of the two degree-of-freedom system of m⫽1179 kg/m, L 1 ⫽1.4 m, L 2 ⫽1.2 m,
Eq. 共23兲 into the dynamic stiffness force-displacement relation-
ship of the Bernoulli-Euler beam given by Eq. 共1兲. This is realized L 3 ⫽1.4 m, M ⫽200 kg, I ␣ ⫽70.833 kgm2 ,
by adding respectively, e 1 to k(2,2) term, e 2 to k(2,5) term, e 2 to
D 1 ⫽D 2 ⫽0.6 m and K S1 ⫽K S2 ⫽1010 N/m
k(5,2) term and e 3 to k(5,5) term of the stiffness matrix K of Eqs.
共1兲–共2兲. Thus the modified force displacement relationship of the The first three natural frequencies of the beam in flexural vibration
combined dynamical system can be written as are shown in Table 1 together with those reported in Ref. 关28兴 and

Fig. 3 A clamped-clamped beam carrying a two degree-of-
freedom spring-mass system
also with those obtained with the two degree-of-freedom spring-

mass system missing. Except for the second natural frequency the
results agreed completely with those of Ref. 关28兴 in which a
Laplace transformation technique with respect to the spatial vari-
able is used, and is expected to give exact results. The author was
unable to pin-point the exact reason for the small discrepancy
共less than 0.04%兲 in the second natural frequency, but this may be
attributed to the accuracy used in the input data of Ref. 关28兴. In
Ref. 关28兴 the built-in end conditions were simulated by using very
large linear and rotational springs 共of the order of 1020) and are
thought unlikely to alter the results. The mode shapes correspond-
ing to these first three natural frequencies are shown in Fig. 4,
which agree very well with the ones illustrated in Fig. 3 of Ref.
关28兴.
Although the present theory is capable of determining the natu-
ral frequencies in axial vibration, these were suppressed by disal-
lowing the axial deformation of the beam, so that the results are
directly comparable with those of Ref. 关28兴. The axial natural
frequencies are uncoupled from the bending ones and they are not
of interest as they will remain unaffected by the presence of the
two degree-of-freedom spring-mass system. 共If the spring connec-
tions are oblique or the two degree-of-freedom system is con-
nected to a framework a coupling would occur in which case the Fig. 4 The first three natural frequencies and mode shapes of
axial motion of the beam would have to be included in the analy- the clamped-clamped beam carrying a two degree-of-freedom
sis.兲 The results of Table 1 show that the presence of the two shown in Fig. 3
degree-of-freedom spring-mass system has changed the first three
natural frequencies by 3.43%, 3.97% and 0.05%, respectively. Of
course, any change in the natural frequencies will be dependent on
the parameters of the two degree-of-freedom spring-mass system
and its location on the beam.
The next set of results for this example was obtained to dem-
onstrate the effect of the spring constants K S1 共or K S2 ) on the
fundamental natural frequency of the beam. 共Note that K S1 and
K S2 are equal for the present problem.兲 Figure 5 shows the varia-
tion of the fundamental natural frequency ( ␻ 1 ) of the beam
against the variation of the spring constant K S1 共or K S2 ). For
smaller values of stiffness coefficients the fundamental natural
frequency increases with the increase in spring constant, but be-
yond a large stiffness constant it approaches a steady 共constant兲
value and remains virtually unaltered. This is to be expected and
is in accord with an earlier investigation 关28兴. Other parameters
Table 1 Natural frequencies of a combined beam and a two

degree-of-freedom system
Natural frequencies 共rad/s兲

Combined beam and two DOF system
Frequency Number Present theory Ref. 关28兴 Beam alone
1 983.19 983 1018.1
2 2695.1 2694 2806.4 Fig. 5 The effect of spring stiffness on the fundamental natu-
3 5499.0 5499 5501.8 ral frequency of the clamped-clamped beam carrying a two
degree-of-freedom shown in Fig. 3

Table 2 Natural frequencies of a plane frame carrying a two
degree-of-freedom system
Natural frequency 共rad/s兲

Frequency Frame with Frame without %
Number two DOF system two DOF system difference
1 176.62 224.76 21.4
2 229.40 245.70 6.63
3 246.09 267.38 7.96
degree-of-freedom system is ignored.兲 The results shown in Table

2 indicate that the natural frequencies vary significantly as a result
of the attached two degree-of-freedom system.
5 Conclusions
Fig. 6 A plane frame carrying a two degree-of-freedom spring A dynamic stiffness matrix for a combined beam and a two
mass system
degree-of-freedom spring-mass system is developed starting from
the classical differential equation theory of its free natural vibra-
tion. By applying the Wittrick-Williams algorithm the resulting
dynamic stiffness matrix is used to determine the free vibration
such as the lengths of the beam segments (L 1 , L 2 and L 3 ) and/or characteristics of a beam and a plane frame which are kinemati-
the mass and mass moment of inertia 共M and I ␣ ) of the two cally connected to a two degree-of-freedom spring-mass system.
degree-of-freedom system, see Fig. 3, can also be varied and their The results for the beam carrying a two degree-of-freedom spring-
subsequent effects on the natural frequencies can be studied. A mass system agreed with published results. The theory allows any
detailed investigation of this type is beyond the scope of this number of two degree-of-freedom spring-mass systems to be
paper because it will no-doubt take it further than it is intended. added to a structure consisting of beams. This useful extension of
However, it should be stressed that the proposed method enables a the dynamic stiffness method to combined system to cover frame-
parametric study to be made to position appropriately the natural works is the context in which the theory is expected to be most
frequencies within or without a certain band. This may sometimes advantageous, particularly in the avoidance or attenuation of un-
be necessary to avoid resonance, for example in the mounting of a welcome vibration. The method used is exact and can be applied
machine. to validate finite element or other approximate methods.
The second illustrative example is a plane frame consisting of
beam members as shown in Fig. 6. One of the structural members
carries a two degree-of-freedom spring-mass system as shown. Appendix
The natural frequencies of the frame without the two degree-of- Derivation of the Dynamic Stiffness Matrix of a Bernoulli-
freedom spring-mass system have been obtained by Howson and Euler Beam. The axial and bending deformations for a
Williams using a published computer program 关34兴 based on the Bernoulli-Euler beam are uncoupled. Therefore, the derivation of
dynamic stiffness matrix method. However, when the two degree- the dynamic stiffness matrix for each of these two cases can be
of-freedom spring-mass system is included the problem becomes carried out separately, and later can be combined together.
very different and cannot generally be solved using earlier ap-
proaches. The application of the present theory is particularly use- Axial Stiffnesses. A uniform Bernoulli-Euler beam of length L
ful for solving such problems. As in Ref. 关34兴 it is assumed that is shown in Fig. 7 in a rectangular right handed Cartesian co-
each member 共beam兲 of the frame shown in Fig. 6 has the same ordinate system with the X-axis coinciding with the axis of the
properties which are as follows: beam. The governing differential equation of motion of the beam
in free axial 共or longitudinal兲 vibration is given by 关36,44兴
EA⫽8.0⫻108 N, EI⫽4.0⫻106 Nm2 and m⫽30 kg/m
⳵ 2u ⳵ 2u
The length of each member of the frame can be deduced from the EA ⫺m ⫽0 (A1)
geometry shown. The parameters of the two degree-of-freedom ⳵x 2
⳵t2
system connected to one of the members in the frame as shown in where EA and m are the axial 共or extensional兲 rigidity and mass
Fig. 6 are per unit length of the beam respectively, and u(x,t) is the axial
displacement of the cross-section at a distance x, and t is time.
M ⫽1500 kg, I ␣ ⫽75 kgm2 , K S1 ⫽108 N/m,
K S2 ⫽1.25⫻108 N/m, D 1 ⫽D 2 ⫽1.5 m.
The nodes of the frame are numbered as shown. Following the

usual procedure generally used in the finite element method, the
dynamic stiffness matrix of the complete frame is assembled from
the dynamic stiffness matrices of all the individual elements in the
frame. Of course, for the element connecting the nodes 2 and 4,
the dynamic stiffness theory developed in this paper is used. The
first three natural frequencies of the frame including and exclud-
ing the two degree-of-freedom system were obtained using the
Wittrick-Williams algorithm and are shown in Table 2 together
with the percentage difference between the two sets of results.
共The published program of Howson and Williams 关34兴 gives the Fig. 7 Coordinate system and notation for a Bernoulli-Euler
same natural frequencies as the present theory when the two beam

冋册冋
F x1
F x2
⫽
a1
a2
册冋册
a2 U1
a1 U2
(A11)
where the frequency dependent elements of the 2⫻2 dynamic

stiffness matrix are given by
EA EA
a 1⫽ ␮ cot ␮ , a 2 ⫽⫺ ␮ cos ec␮ (A12)
L L
Bending Stiffnesses. The governing differential equation of
motion of the Bernoulli-Euler beam shown in Fig. 7 in bending
共or flexural兲 natural vibration is given by 关36,44兴
⳵ 4v ⳵ 2v
EI ⫹m ⫽0 (A13)
⳵x 4
⳵t2
where EI and m are the bending 共or flexural兲 rigidity and mass per
unit length of the beam respectively, and v (x,t) is the bending 共or
Fig. 8 End conditions for a Bernoulli-Euler beam in free vibra- flexural兲 displacement of the cross-section at a distance x and t is
tion time.
Assuming harmonic oscillation with angular 共or circular兲 fre-
quency ␻ so that
Assuming harmonic oscillation with angular 共or circular兲 fre- v共 ␰ ,t 兲 ⫽V 共 ␰ 兲 e i ␻ t (A14)

quency ␻ so that
where V( ␰ ) is the amplitude of bending 共or flexural兲 vibration and
u 共 x,t 兲 ⫽u 共 ␰ ,t 兲 ⫽U 共 ␰ 兲 e i ␻ t (A2) ␰ ⫽x/L. Substituting Eq. 共A14兲 into Eq. 共A13兲 gives
where U( ␰ ) is the amplitude of axial vibration and ␰ ⫽x/L.
Substituting Eq. 共A2兲 into Eq. 共A1兲 gives d 4V
⫺␭ 4 V⫽0 (A15)
2 d␰4
d U
⫹ ␮ 2 U⫽0 (A3) where
d␰2
where
␭⫽ 冑
4 m ␻ 2L 4
(A16)
␮⫽ 冑 m ␻ 2L 2
EA
(A4)
EI
The solution of the differential Eq. 共A15兲 is given by 关36,44兴
The solution of the differential Eq. 共A3兲 is given by V 共 ␰ 兲 ⫽C 3 cos ␭ ␰ ⫹C 4 sin ␭ ␰ ⫹C 5 cosh ␭ ␰ ⫹C 6 sinh ␭ ␰
U 共 ␰ 兲 ⫽C 1 cos ␮ ␰ ⫹C 2 sin ␮ ␰ (A5) (A17)
where C 1 and C 2 are constants. where C 3 – C 6 are constants.
The expression for axial force F(x) in the positive direction of The expressions for rotation 共␪兲, bending moment 共M兲 and
X 共see Fig. 7兲 can be expressed with the help of Eq. 共A5兲 as shear force 共S兲 consistent with the positive direction of the co-
ordinate system at the left hand end of the beam can be written as
dU EA dU
F 共 x 兲 ⫽F 共 ␰ 兲 ⫽⫺EA ⫽⫺ 1 dV ␭
dx L d␰
␪共 x 兲⫽␪共 ␰ 兲⫽ ⫽ 共 ⫺C 3 sin ␭ ␰ ⫹C 4 cos ␭ ␰ ⫹C 5 sinh ␭ ␰
L d␰ L
EA
⫽ ␮ 共 C 1 sin ␮ ␰ ⫺C 2 cos ␮ ␰ 兲 (A6) ⫹C 6 cosh ␭ ␰ 兲 (A18)
L
Now referring to Fig. 8共a兲, the boundary 共or end兲 conditions for EI d 2 V EI
displacements and forces can be applied as follows. M 共 x 兲 ⫽M 共 ␰ 兲 ⫽⫺ ⫽ ␭ 2 共 C 3 cos ␭ ␰ ⫹C 4 sin ␭ ␰
L d␰
2 2
L2
At x⫽0 共 i.e. ␰ ⫽0 兲 , U⫽U 1 and F⫽F x1 (A7)
⫺C 5 cosh ␭ ␰ ⫺C 6 sinh ␭ ␰ 兲
At x⫽L 共 i.e. ␰ ⫽1 兲 , U⫽U 2 and F⫽⫺F x2 (A8)
(A19)
Substituting Eqs. 共A7兲 and 共A8兲 into Eqs. 共A5兲 and 共A6兲, the fol-
lowing matrix relationships can be obtained EI d 3 V EI
冋册冋册冋册
S 共 x 兲 ⫽S 共 ␰ 兲 ⫽ ⫽ ␭ 3 共 C 3 sin ␭ ␰ ⫺C 4 cos ␭ ␰
U1 1 0 C1 L d␰
3 3
L3
⫽ (A9)
U2 cos ␮ sin ␮ C 2
⫹C 5 sinh ␭ ␰ ⫹C 6 cosh ␭ ␰ 兲 (A20)
and
Now referring to Fig. 8共b兲, the boundary 共or end兲 conditions for
冋册
F1
F2
⫽
EA
L
␮
0
⫺sin ␮
冋 ⫺1
册冋册
C1
cos ␮ C 2
(A10)
the displacements and forces can be applied as follows
At x⫽0 共 i.e. ␰ ⫽0 兲 , V⫽V 1 , ␪⫽␪1 , S⫽F y1

The constants C 1 and C 2 can now be eliminated from Eqs. 共A9兲
and 共A10兲 to give the dynamic stiffness formulation of the axially and M ⫽M 1 (A21)
vibrating beam relating amplitudes of the forces and displace-
ments at its ends as follows At x⫽L 共 i.e. ␰ ⫽1 兲 , V⫽V 2 , ␪⫽␪2 , S⫽⫺F y2

and M ⫽⫺M 2 (A22) or
Substituting Eqs. 共A21兲 and 共A22兲 into Eqs. 共A17兲–共A20兲, the fol- F⫽K␦ (A31)
lowing two matrix equations can be obtained for displacements where F and ␦ are respectively the force and displacement vectors
冋册冋册冋册
and forces, respectively, in terms of the constants C 3 – C 6 . and K is the frequency dependent 6⫻6 dynamic stiffness matrix
1 0 1 0 whose elements k(i, j) (i⫽1,2 . . . 6; j⫽1,2, . . . 6) are given by
V1 C3 a 1 , a 2 and d 1 ⫺d 6 defined in Eqs. 共A12兲, 共A27兲 and 共A28兲. Note
␪1 0 ␭/L 0 ␭/L C4 that K is symmetric as expected.
⫽
V2 C␭ S␭ C h␭ S h␭ C5
␪2 ⫺␭S ␭ /L ␭C ␭ /L ␭S h␭ /L ␭C h␭ /L
C6 References
(A23) 关1兴 Chen, Y., 1963, ‘‘On the Vibration of Beams or Rods Carrying a Concentrated
Mass,’’ ASME J. Appl. Mech., 30, pp. 310–311.
冋册冋册
and 关2兴 Pan, H. H., 1965, ‘‘Transverse Vibration of an Euler Beam Carrying a System
of Heavy Bodies,’’ ASME J. Appl. Mech., 32, pp. 434 – 437.
0 ⫺W 3 ␭ 3 0 W 3␭ 3 关3兴 Laura, P. A. A., Pombo, J. L., and Susemihl, E. A., 1974, ‘‘A Note on the
F y1 Vibration of a Clamped-Free Beam with a Mass at the Free End,’’ J. Sound
M1 W 2␭ 2
0 ⫺W 2 ␭ 2
0 Vib., 37, pp. 161–168.
⫽ 关4兴 Parnell, L. A., and Cobble, M. H., 1976, ‘‘Lateral Displacements of a Vibrating
F y2 ⫺W 3 ␭ S ␭ W 3 ␭ C ␭ ⫺W 3 ␭ S h␭ ⫺W 3 ␭ 3 C h␭
3 3 3
Cantilever Beam with a Concentrated Mass,’’ J. Sound Vib., 44, pp. 499–511.
M2 关5兴 Gurgoze, M., 1984, ‘‘A Note on the Vibrations of Restrained Beams and Rods
⫺W 2 ␭ 2 C ␭ ⫺W 2 ␭ 2 S ␭ W 2 ␭ 2 C h␭ W 2 ␭ 2 S h␭
冋册
with Point Masses,’’ J. Sound Vib., 96, pp. 461– 468.
关6兴 Burch, J. C., and Mitchell, T. P., 1987, ‘‘Vibration of a Mass-Loaded Clamped-
C3 Free Timoshenko beam,’’ J. Sound Vib., 114, pp. 341–345.
C4 关7兴 Laura, P. A. A., Filipich, C. P., and Cortinez, V. H., 1987, ‘‘Vibrations of
⫻ (A24) Beams and Plates Carrying Concentrated Masses,’’ J. Sound Vib., 117, pp.
C5 459– 465.
C6 关8兴 Wu, J. S., and Lin, T. L., 1990, ‘‘Free Vibration Analysis of a Uniform Can-
tilever Beam with Point Masses by an Analytical-and-Numerical-Combined
where Method,’’ J. Sound Vib., 136, pp. 201–213.
关9兴 Abramovich, H., and Hamburger, O., 1991, ‘‘Vibration of a Cantilever Ti-
C ␭ ⫽cos ␭, S ␭ ⫽sin ␭, C h␭ ⫽cosh ␭, S h␭ ⫽sinh ␭, moshenko Beam with a Tip Mass,’’ J. Sound Vib., 148, pp. 162–170.
关10兴 Massalas, C., and Soldatos, K., 1978, ‘‘Free Vibration of a Beam Subjected to
Elastic Constraints,’’ J. Sound Vib., 57, pp. 607– 608.
EI EI EI 关11兴 Davies, H. G., and Rogers, R. J., 1979, ‘‘The Vibration of Structures Elasti-
W 1⫽ , W 2⫽ , W 3⫽ (A25) cally Constrained at Discrete Points,’’ J. Sound Vib., 63, pp. 437– 447.
L L 2
L3
关12兴 Lau, J. H., 1981, ‘‘Fundamental Frequency of a Constrained Beam,’’ J. Sound
The constants C 3 – C 6 can now be eliminated to give the 4⫻4 Vib., 78, pp. 154 –157.
关13兴 Verniere, P., Ficcadenti, G., and Laura, P. A. A., 1984, ‘‘Dynamic Analysis of
dynamic stiffness matrix of the Bernoulli-Euler beam in flexural a Beam with an Intermediate Elastic Support,’’ J. Sound Vib., 96, pp. 381–389.
冋册冋册冋册
vibration as follows 关14兴 Lau, J. H., 1984, ‘‘Vibration Frequencies and Mode Shapes for a Constrained
Cantilever,’’ ASME J. Appl. Mech., 51, pp. 182–187.
F y1 d1 d2 d4 d5 V1 关15兴 Maurizi, M. J., and Bambill de Rossit, D. V., 1987, ‘‘Free Vibration of a
Clamped-Clamped Beam with an Intermediate Elastic Support,’’ J. Sound Vib.,
M1 d2 d3 ⫺d 5 d6 ␪1 119, pp. 173–176.
⫽ (A26) 关16兴 Rao, C. K., 1989, ‘‘Frequency Analysis of Clamped-Clamped Uniform Beams
F y2 d4 ⫺d 5 d1 ⫺d 2 V2
with Intermediate Elastic Support,’’ J. Sound Vib., 133, pp. 502–509.
M2
d5 d6 ⫺d 2 d3
␪2 关17兴 Jacquot, R. G., and Gibson, J. D., 1972, ‘‘The Effects of Discrete Masses and
Elastic Supports on Continuous Beam Natural Frequencies,’’ J. Sound Vib., 23,
where pp. 237–244.
关18兴 Laura, P. A. A., Maurizi, M. J., and Pombo, J. L., 1975, ‘‘A Note on the
d 1 ⫽W 3 ␭ 3 共 S ␭ C h␭ ⫹C ␭ S h␭ 兲 /⌬, d 2 ⫽W 2 ␭ 2 S ␭ S h␭ /⌬, Dynamic Analysis of an Elastically Restrained-Free Beam with a Mass at the
Free End,’’ J. Sound Vib., 41, pp. 397– 405.
关19兴 Laura, P. A. A., Susemihl, E. A., Pombo, J. L., Luisoni, L. E., and Gelos, R.,
d 3 ⫽W 1 ␭ 共 S ␭ C h␭ ⫺C ␭ S h␭ 兲 /⌬ (A27) 1977, ‘‘On the Dynamic Behavior of Structural Elements Carrying Elastically
Mounted, Concentrated Masses,’’ Appl. Acoust., 10, pp. 121–145.
d 4 ⫽⫺W 3 ␭ 3 共 S ␭ ⫹S h␭ 兲 /⌬, d 5 ⫽W 2 ␭ 2 共 C h␭ ⫺C ␭ 兲 /⌬, 关20兴 Bapat, C. N., and Bapat, C., 1987, ‘‘Natural Frequencies of a Beam with
Nonclassical Boundary Conditions and Concentrated Masses,’’ J. Sound Vib.,
d 6 ⫽W 1 ␭ 共 S h␭ ⫺S ␭ 兲 /⌬ (A28) 112, pp. 177–182.
关21兴 Ercoli, L., and Laura, P. A. A., 1987, ‘‘Analytical and Experimental Investiga-
with tion on Continuous Beams Carrying Elastically Mounted Masses,’’ J. Sound
Vib., 114, pp. 519–533.
⌬⫽1⫺C ␭ C h␭ (A29) 关22兴 Larrondo, H., Avalos, D., and Laura, P. A. A., 1992, ‘‘Natural Frequencies of a
Bernoulli Beam Carrying an Elastically Mounted Concentrated Mass,’’ Ocean
Combination of Axial and Bending Stiffnesses. A simple su- Eng., 19, pp. 461– 468.
perposition is now possible to put the axial and bending dynamic 关23兴 Abramovich, H., and Hamburger, O., 1992, ‘‘Vibration of a Cantilever Ti-
moshenko Beam with Translational and Rotational Springs and with Tip
stiffnesses together in order to express the force-displacement re- Mass,’’ J. Sound Vib., 154, pp. 67– 80.
lationship of the Bernoulli-Euler beam. Superposing Figs. 8共a兲关24兴 Rossi, R. E., Laura, P. A. A., Avalos, D. R., and Larrondo, H. O., 1993, ‘‘Free
and 8共b兲, see Fig. 1, and using Eqs. 共A11兲 and 共A26兲, one obtains Vibrations of Timoshenko Beams Carrying Elastically Mounted, Concentrated
the dynamic stiffness relationship of a Bernoulli-Euler beam in Masses,’’ J. Sound Vib., 165, pp. 209–223.
关25兴 Gurgoze, M., 1996, ‘‘On the Eigenfrequencies of a Cantilever Beam with
冋册冤冥冋册
free vibration, as follows Attached Tip Mass and a Spring-Mass System,’’ J. Sound Vib., 190, pp. 149–
162.
a1 0 0 a2 0 0 关26兴 Jen, M. U., and Magrab, E. B., 1993, ‘‘Natural Frequencies and Mode Shapes
F x1 U1 of Beams Carrying a Two-Degree-of-Freedom Spring-Mass System,’’ ASME
0 d1 d2 0 d4 d5
F y1 V1 J. Vibr. Acoust., 115, pp. 202–209.
M1 0 d2 d3 0 ⫺d 5 d6 ␪1 关27兴 Wu, J. S., and Huang, C. G., 1995, ‘‘Free and Forced Vibrations of a Timosh-
⫽ enko Beam with any Number of Translational and Rotational Springs and
F x2 a2 0 0 a1 0 0 U2 Lumped Masses,’’ Int. J. Commun. Num. Meth. Eng., 11, pp. 743–756.
F y2 V2 关28兴 Chang, T. P., and Chang, C. Y., 1998, ‘‘Vibration Analysis of Beams with a
0 d4 ⫺d 5 0 d1 ⫺d 2 Two Degree-of-Freedom Spring-Mass System,’’ Int. J. Solids Struct., 35, pp.
M2 ␪2 383– 401.
0 d5 d6 0 ⫺d 2 d3 关29兴 Wu, J. S., and Chou, H. M., 1998, ‘‘Free Vibration Analysis of a Cantilever
(A30) Beam Carrying any Number of Elastically Mounted Pointed Masses with the

Analytical-and-Numerical-Combined Method,’’ J. Sound Vib., 213, pp. 317– 关37兴 Wittrick, W. H., and Williams, F. W., 1971, ‘‘A General Algorithm for Com-
332. puting Natural Frequencies of Elastic Structures,’’ Q. J. Mech. Appl. Math., 24,
关30兴 Wu, J. S., and Chou, H. M., 1999, ‘‘A New Approach for Determining the pp. 263–284.
Natural Frequencies and Mode Shapes of a Uniform Beam Carrying any Num- 关38兴 Anderson, M. S., Williams, F. W., Banerjee, J. R., Durling, B. J., Herstrom, C.
ber of Sprung Masses,’’ J. Sound Vib., 220, pp. 451– 468. L., Kennedy D., and Warnaar, D. B., 1986, ‘‘User Manual BUNVIS-RG: An
关31兴 Wu, J. J., and Whittaker, A. R., 1999, ‘‘The Natural Frequencies and Mode Exact Buckling and Vibration Program for Lattice Structures, with Repetitive
Shapes of a Uniform Beam with Multiple Two-DOF Spring-Mass Systems,’’ J. Geometry and Substructuring option,’’ NASA Tech. Memo. 87669.
Sound Vib., 227, pp. 361–381. 关39兴 Cheng, F. Y., 1970, ‘‘Vibration of Timoshenko Beams and Frameworks,’’ J.
关32兴 Dowell, E. H., 1979, ‘‘On Some General Properties of Combined Dynamical
Struct. Div. ASCE, 96, pp. 551–571.
Systems,’’ ASME J. Appl. Mech., 46, pp. 206 –209.
关40兴 Wang, T. M., and Kinsman, T. A., 1971, ‘‘Vibration of Frame Structures Ac-
关33兴 Nicholson, J. W., and Bergman, L. A., 1986, ‘‘Free Vibration of Combined
cording to the Timoshenko Theory,’’ J. Sound Vib., 14, pp. 215–227.
Dynamical Systems,’’ J. Eng. Mech., 112, pp. 1–13.
关34兴 Howson, W. P., and Williams, F. W., 1977, ‘‘Compact Computation of Natural 关41兴 Howson, W. P., and Williams, F. W., 1973, ‘‘Natural Frequencies of Frames
Frequencies and Buckling Loads for Plane Frames,’’ Int. J. Numer. Methods with Axially Loaded Timoshenko Members,’’ J. Sound Vib., 26, pp. 503–515.
Eng., 11, pp. 1067–1081. 关42兴 Cheng, F. Y., and Tseng, W. H., 1973, ‘‘Dynamic Stiffness Matrix of Timosh-
关35兴 Howson, W. P., Banerjee, J. R., and Williams, F. W., 1983, ‘‘Concise Equations enko Beam Columns,’’ J. Struct. Div. ASCE, 99, 527–549.
and Program for Exact Eigensolutions of Plane Frames including Member 关43兴 Banerjee, J. R., 1996, ‘‘Dynamic Stiffness Formulation for Structural Ele-
Shear,’’ Adv. Eng. Software, 5, pp. 137–141. ments: A General Approach,’’ Comput. Struct., 63, pp. 101–103.
关36兴 Kolousek, V., 1973, Dynamics in Engineering Structures, Butterworths, Lon- 关44兴 Clough, R. W., and Penzien, J., 1975, Dynamics of Structures, McGraw-Hill,
don. Singapore.

Effect of Damping on Asymmetric
Systems
Paolo Gallina
Associate Professor, This paper addresses the phenomenon of the destabilizing effect of slight damping on
Department of Energetics, asymmetric linear systems. Previous works had showed that the destabilizing effect, re-
University of Trieste, garded for a long time as a ‘‘paradox,’’ depends upon the ratio of the damping coeffi-
Via A. Valerio 10-34127 Trieste, Italy cients. This work extends those results to n-dof systems. In fact, conditions for a general
e-mail: [email protected] asymmetric n-dof slightly damped system to be stable are obtained. Also, a useful suffi-
cient condition is carried out. This practical design tool gives optimum damping ratios
and takes into consideration the parameter uncertainties as well. The solution is based on
the eigenvalue sensitivity analysis. Moreover, a formal physical explanation of the desta-
bilizing effect of damping is given. Eventually, the theory is validated by means of a
simple example. 关DOI: 10.1115/1.1569945兴
1 Introduction the eigenvalues will move. We proved that destabilizing effect due
to slight damping occurs when a system of homogeneous linear
The dynamic behavior of a general linear discrete undamped
inequalities is not satisfied. This general condition is an extension
system can be described by a vector differential equation of the
of the results obtained by Bolotin 关5兴, Herrmann et al. 关6兴
form
The paper is divided into three parts: in the first one a detailed
analysis of slight damping effects on a general n-dof stable un-
Mẍ共 t 兲 ⫹Kx共 t 兲 ⫽0 (1)
damped system is carried out. A stability condition is obtained.
where M and K are the mass and stiffness matrices respectively. Moreover, using the cone algorithm 关10兴, a simplified sufficient
If mass and stiffness matrices are symmetric positive definite, condition for the slight damped system to be stable is obtained.
the system is stable in the BIBO sense, where the acronym BIBO This result is of practical interest because it gives the optimum
stands for ‘‘bounded input bounded output.’’ In most cases, the damping ratios for the system to be stable. Damping coefficient
damping matrix is neglected as far as stability analysis is con- uncertainties are taken into account as well.
cerned. In fact, the structural damping of the real system would In the second part, a physical explanation of the destabilizing
have the effect of moving the eigenvalues of the undamped sys- effect of damping is given for a general n-dof system, which ex-
tem toward the negative complex plane. tends the result obtained by Semler 关4兴 Eventually, the proposed
If the mass and/or the stiffness matrices are asymmetric, the method is explained and validated by means of a simple example.
undamped system can be either stable or unstable. This type of
dynamic systems can be found in aeronautics, active control of
large structures, vehicle dynamics and milling machining 关1,2兴. In 2 Theory
these circumstances, neglecting the damping matrix is no longer
possible. 2.1 Sensitivity Eigenvalue Analysis. Consider the system
Because of the asymmetry of the matrices, a destabilizing effect
due to the damping matrix could arise. This atypical phenomenon, Mẍ共 t 兲 ⫹Cẋ共 t 兲 ⫹Kx共 t 兲 ⫽0 (2)
regarded for a long time as a ‘‘paradox’’ 关3兴 has been physically where M and K are real n⫻n asymmetric matrices. The damping
explained by Semler et al. 关4兴. Using a 2 d.o.f. example, Semler matrix C can be decomposed into a product of a matrix U and a
showed that damping introduces two effects: it dissipates energy scalar c, namely C⫽cU, where c⫽ 储 C储 and U⫽C/储C储. C is as-
and it modifies the modes of vibration. The latter could cause the sumed symmetric and U inherits this symmetry.
work done by the external forces to increase. If the work done by Note that the system 共2兲 can be always arranged so that the
the external forces becomes greater than the energy dissipated by damping matrix is symmetric. In fact, if C was not symmetric, it
damping, the system will be unstable. is always possible to decompose the damping matrix using the
Bolotin 关5兴 showed how the destabilizing effect of slight damp- singular value decomposition method, namely C⫽SC̃V, where S
ing is related to the ratio of the damping coefficients. This result and V are square matrices and C̃ is a symmetric matrix. Therefore,
was carried out for a 2 d.o.f. asymmetric system. Studying a lin- the system 共2兲 can be written in the form M̃ẍ(t)⫹C̃ẋ(t)⫹K̃x(t)
earized double pendulum with a follower force, Herrmann et al.
关6,7兴 achieved similar results. Using an approximated method, ⫽0 where M̃⫽S⫺1 MV⫺1 , K̃⫽S⫺1 KV⫺1 and C̃ is symmetric.
Nemat-Nasser 关8兴, showed the destabilizing effect of damping on By means of the decomposition cU, one can calculate the ei-
a cantilevered continuous pipe conveying fluid. The approach he genvalues of the system as a function of c. If c⫽0, Eq. 共2兲 rep-
adopted can be extended to 3 d.o.f systems. Unfortunately, all resents the equation of the undamped system. As c increases, the
those works focus on particular examples with few degrees of damping matrix will affect the dynamic behavior of the system. In
freedom. This paper extends the theory to a general n-dof asym- this way one can study the effect of a slight damping by introduc-
metric linear slightly damped system. If the undamped system is ing a small c. Since 兵 U其 i j ⫽ 兵 C其 i j /c, each component of U repre-
stable, all the eigenvalues are located on the imaginary axis. The sents a damping ratio. Note that, for definition, all the damping
introduction of slight damping modifies the location of the eigen- ratios are constant. For the sake of clarity, we will call U damping
values of the system on the complex plane. By means of an ei- ratio matrix and c damping.
genvalue sensitivity analysis 关9兴, one can foresee which direction In the following, we answer this simple but fundamental ques-
tion: If the undamped system is stable, will the introduction of the
damping destabilize the system? In order to answer the question,
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Sept. 2001; one can find where the eigenvectors move on the complex plane
Revised Jan. 2003. Associate Editor: J. Cusumano. as c changes.
The right eigenvalue problem associated with Eq. 共2兲 can be
represented by
共 M␭ i2 ⫹C␭ i ⫹K兲 ui ⫽0 (3)
where ␭ i , ui are respectively the ith eigenvalue and the ith right
eigenvector of the system. In the same way, the left eigenvector
problem is given by the equation
viT 共 M␭ i2 ⫹C␭ i ⫹K兲 ⫽0 (4)
where vi is the ith left eigenvector of the system. Fig. 1 Mechanical example of a n d.o.f. system. Nonconserva-
Differentiating Eq. 共3兲 with respect to the damping c, one ob- tive forces that cause the mass matrix andÕor the damping ma-
tains trix to be asymmetric are not depicted in figure. A damping
冉冊
element can be inserted between two masses andÕor between
d␭ i d␭ i dui one mass and the frame.
2M␭ i ⫹U␭ i ⫹C ui ⫹ 共 M␭ i2 ⫹C␭ i ⫹K兲 ⫽0 (5)
dc dc dc
Premultiplying by the left eigenvalue viT , it yields
冉
viT 2M␭ i
d␭ i
dc
⫹U␭ i ⫹C
d␭ i
冊
u ⫹vT 共 M␭ i2 ⫹C␭ i ⫹K兲
dc i i
dui
dc
d␭ 1 d␭ 2
dc
⫽
dc
vT1 Uu1
⫽⫺ T
2v1 Mu1
⫽⫺0.5263
冉
⫽viT 2M␭ i
d␭ i
dc
⫹U␭ i ⫹C
d␭ i
dc i
u ⫽0 冊 (6) d␭ 3 d␭ 4
dc
⫽
dc
vT3 Uu3
⫽⫺ T
2v3 Mu3
⫽⫺0.1162 (12)
Therefore the derivative of the ith eigenvalue calculated for c
⫽0 is In conclusion, for a small value of c, the stable undamped system
冏
will remain stable
d␭ i viT 共 U␭ i 兲 ui viT Uui
⫽⫺ T ⫽⫺ (7)
dc vi 共 2M␭ i ⫹C兲 ui c⫽0
2viT Mui 2.2 An Alternative Simpler Way to Express Stability Con-
ditions. Instead of Eq. 共7兲, by expanding matrix U, we can ob-
Eq. 共7兲 is a simple tool for foreseeing the effect of damping. In tain a simplified formulation. For the sake of clarity, we introduce
fact, if Re(d␭i /dc)⬎0, a small c will move the ith eigenvalue ␭ i of Fig. 1, where part of a discrete mass-spring-damped system is
the undamped system (␭ i belongs to the complex axis since the represented. In this way one can visualize the physical meaning of
undamped system is supposed to be stable in a BIBO sense兲 to- the elements of the matrix U. In other words, Fig. 1 shows the
ward the right complex plane, causing the system to become un- damping coefficients one can introduce in the mechanical system.
stable. Conversely, if The forces that cause the stiffness matrix and/or the mass matrix
Re共 d␭ i /dc 兲 ⬍0, (8) to be asymmetric are not depicted in figure.
We can now study the structure of the damping ratio matrix U
a small c will move the ith eigenvalue toward left, causing the and give its physical interpretation. This provides the designer
system to remain stable. Therefore, for small c, the damping ratio with an important tool to select the best combination of damping
matrix will have a ‘‘stabilizing effect’’ only if Re(d␭i /dc)⬍0 ᭙i elements. Basically, one has two different ways to introduce a
⫽1, . . . ,n. Note that the eigenvectors of an undamped system are damping element on a mechanical system:
always real. Therefore, for this reason and for Eq. 共7兲
冉冊
• a damping element can be connected between one mass/
d␭ i d␭ i inertia and the frame;
Re ⫽ (9) • a damping element can be connected between two masses/
dc dc
inertias.
We now apply the condition 共8兲 to the following simple example.
Consider the system Mẍ⫹cUẋ⫹Kx⫽0, with the matrix The former is represented in the figure by a physical damping
冋册冋册冋册
element attached between the mass m l and the frame. Its numeri-
3 2 5 1.5 2 0 cal damping value is c ␰ ll . We call damping element coefficient
M⫽
1 3
K⫽
4 10
U⫽
0 1
(10) the adimensional coefficient ␰ ll .
In this case, if c ␰ ll was the only damping element on the me-
冋册
The Undamped system (c⫽0) is stable in the sense of BIBO chanical system, the matrix U would be
stable. Its eigenvalues are ␭ 1 ⫽1.4684i, ␭ 2 ⫽⫺1.4684i, ␭ 3
⫽1.7074i, ␭ 3 ⫽⫺1.7074i. The right and left eigenvectors of the 0 ¯ ¯ 0
再
undamped system are ] ]
兵 Ull 其 o, p ⫽0
冋册冋册冋册
᭙o,p⫽l
⫺0.7327 ⫺0.7327 0.6526 U⫽Ull ⫽ ⫽ ␰ ll
u1 ⫽ , u2 ⫽ , u3 ⫽ , 兵 Ull 其 o, p ⫽ ␰ ll o⫽p⫽l
0.3825 0.3825 ⫺0.5645 ] ]
u4 ⫽ 冋 0.6526
⫺0.5645 册 (11)
0 ¯ ¯ 0
(13)
v1 ⫽ 冋
⫺0.6466
⫺0.5148
, 册 v2 ⫽冋⫺0.6466
⫺0.5148
, 册 v3 ⫽ 冋
⫺0.2400
⫺0.8289
, 册 where the notation 兵其 o, p indicates the element of the matrix on
the oth row and pth column.
Since the system has n degrees of freedom, it is possible to
v4 ⫽ 冋 ⫺0.2400
⫺0.8289 册 connect up to n damping elements between each mass/inertia and
the frame.
If c ␰ ls was the only damping elements introduced between the
Therefore mass m l and the mass m s the matrix U would be

U⫽Uls ⫽ 再
兵 Uls 其 o,p ⫽0 ᭙o⫽l,o⫽s,p⫽l,p⫽s
兵 Uls 其 l,l ⫽ 兵 Uls 其 s,s ⫽ ␰ ls
In order the derivatives of the eigenvalues to be negative, the
following system of linear inequalities has to be satisfied
冋册
兵 Uls 其 l,s ⫽ 兵 Uls 其 s,l ⫽⫺ ␰ ls A␰⬎0 (21)
0 ¯ ¯ 0 where A苸R n⫻r
is
⫽
]
]
␰ ls
⫺ ␰ ls
0
0
0
⫺ ␰ ls
␰ ls
]
]
(14) A⫽ 冋
sign共 vT1 Mu1 兲 aT1
]
sign共 vTn Mun 兲 aTn
册 (22)
0 ¯ ¯ 0 In conclusion, instead of Eq. 共8兲, the stability condition is now

replaced by the system of inequalities 共21兲. They present the ad-
In a n d.of. system, the number of possible damping elements vantage that damping element coefficients appear explicitly.
attached between one mass and another one is (n(n⫺1))/2. Therefore damping elements can be easily chosen in order main-
Therefore the sum of the number of the possible damping ele- tain the system stable.
ments attached to the frame plus the number of the possible damp- Example:
ing elements attached between two masses is n⫹(n(n⫺1))/2 Considering again the numerical example 共10兲, where ␰ 11⫽2,
冋册
⫽(n(n⫹1))/2. In many cases, the number of damping elements ␰ 22⫽1, one obtains
in a real mechanical system is less than (n(n⫹1))/2.
Using a compact notation and considering all the damping ele- 0.4737 ⫺0.1969
ment coefficients at the same time, one can compose the damping 0.4737 ⫺0.1969
ratio matrix A⫽
⫺0.1566 0.4679
n n n
⫺0.1566 0.4679
U⫽ 兺 U ⫹兺
r⫽1
rr
r,t⫽1
Urt ⫽ 兺
r,t⫽1
Urt (15)
and Eq. 共20兲 becomes
r⬎t r⭓t
A兵 ␰ 11 ␰ 22其 T ⫽ 兵 0.7505 0.7505 0.1547 0.1547其 T ⬎0 (23)
Multiplying by the ith right eigenvalue and premultiplying by the
ith left eigenvalue, it yields Eq. 共23兲, according to Eq. 共12兲 confirms again that the slightly
n n
damped system is stable
viT Uui ⫽ 兺
r⫽1
viT Urr ui ⫹ 兺
r,t⫽1
viT Urt ui
2.3 A Simpler Sufficient Condition for Stability „Solution
r⬎t
Cone…. Although condition 共21兲 is useful to directly foresee the
n n
stabilizing or destabilizing effect of the damping element coeffi-
⫽ 兺兵v 其兵u 其 ␰ i r i r rr ⫹ 兺共兵 vi 其 r 兵 ui 其 r ⫺ 兵 vi 其 r 兵 ui 其 t cients, it is not of practical use for the mechanical designer.
r⫽1 r,t⫽1
r⬎t The main reason is that system 共21兲 has an infinite number of
solutions. A method to choose the ‘‘best solution’’ of 共21兲 among
⫺ 兵 vi 其 t 兵 ui 其 r ⫹ 兵 vi 其 t 兵 ui 其 t 兲 ␰ rt (16) the possible alternatives is desirable.
where the symbol 兵其 r means the rth component of a vector. It is We now introduce the notation ãi ⫽sign(viT Mui )ai . From a
clear from the Eq. 共16兲 that viT Uui is a linear combination of the geometrical point of view, all the infinite solutions of 共21兲 can be
damping element coefficients ␰ r,r , ␰ r,t (r,t⫽1, . . . ,n,r⬎t). obtained in the following way. Each vector ãi can be thought of as
Therefore, one can write the normal vector to an homogeneous hyperplane in R r . The hy-
perplane bounds the halfplane S i ⫽ 兵 ␰兩 ãiT ␰⬎0,␰苸R r 其 . The inter-
viT Uui ⫽aiT ␰ (17) section of the r halfplanes generates the solution set S of system
共21兲 S⫽艚 i⫽1 r
冋册
S i . It is well known that the solution region is a
where
convex polyhedral cone in R r 关11兴. Introducing the definition of
兵 vi 其 1 兵 ui 其 1 solution cone one can find a subset of solutions of system 共21兲.
] A hypercone 关10兴 with axis w and angle ␪ in R r is defined as
兵 vi 其 n 兵 ui 其 n C 共 w, ␪ 兲 ⫽ 兵 ␰兩具 w, ␰典 ⬍ ␪ , ␰苸R r 其 (24)
ai ⫽
兵 vi 其 1 兵 ui 其 2 ⫺ 兵 vi 其 1 兵 ui 其 2 ⫺ 兵 vi 其 2 兵 ui 其 1 ⫹ 兵 vi 其 2 兵 ui 其 2 where 具w,␰典⫽arccos共w"␰/共储w储储␰储兲兲 is the angle between w and ␰.
] A solution cone is a hypercone enclosed by the solution set S. One
兵 vi 其 n⫺1 兵 ui 其 n ⫺ 兵 vi 其 n⫺1 兵 ui 其 n ⫺ 兵 vi 其 n 兵 ui 其 n⫺1 ⫹ 兵 vi 其 n 兵 ui 其 n can find the largest solution cone enclosed by the solution set. The
set of all the angles of the cones enclosed by S is
and the vector ␰苸R r (r⫽((n(n⫹1))/2)) gathers the damping co-
efficients ⌬ sc ⫽ 兵 ␪ 兩 C 共 w, ␪ 兲苸S,w苸R r , ␪ 苸R 其 (25)
␰T ⫽ 关 ␰ 11 , ␰ 22 , . . . , ␰ nn , ␰ 12 , ␰ 13 , . . . ␰ 1n , ␰ 23 , . . . , ␰ 共 n⫺1 兲 n 兴 Therefore, the angle of the largest solution cone is ␪ sc
(18) ⫽max兵⌬sc其 and its axis is
By replacing Eq. 共17兲 into 共7兲, one obtains the derivative of the ith wsc ⫽ 兵 w兩 C 共 w, ␪ sc 兲苸S,w苸R r 其 (26)
eigenvalue as a function of the damping element coefficient vector The largest solution cone can be found using the cone algorithm
proposed by Wan 关10兴. A modified version of the cone algorithm
d␭ i aiT ␰ is proposed in appendix. Since the largest solution cone is the
⫽⫺ (19)
dc 2viT Mui largest cone included in S, it gives a set of solutions of system
共21兲. Of course, it does not represent all the possible solutions; it
Therefore defines a subset of the solutions of the system of inequalities. In
spite of that, one can note that, while system 共21兲 is made up of r
d␭ i inequalities, the largest solution cone is defined by means of only
⬍0⇔sign共 viT Mui 兲 aiT ␰⬎0 (20)
dc one inequality:

arccos 冉 T
wsc
•
␰
储 wsc 储储 ␰储
冊
⬍ ␪ sc ⇔
T
wsc
储 wsc 储
• ␰ˆ ⬎cos共 ␪ sc 兲 (27)
sider only the steady periodic part of the solution. In the following
we calculate the energy the system exchanges during each cycle.
1
where ␰ˆ ⫽␰/储␰储. Moreover, only two parameters are necessary to x⬵ 共 u e i ␻ i t ⫹ūi e ⫺i ␻ i t 兲 (35)
express the condition: wsc , ␪ sc . Equation 共27兲 is a powerful and 2 i
simple tool for designing the damping matrix. The damping ma- The energy dissipated by damping during one cycle is
trix will stabilize the undamped system 共under the hypothesis of
slight damping兲 if Eq. 共27兲 is satisfied. If one could choose the
damping element coefficients, the best solution would be W d ⫽⫺ 冕 0
2␲/␻i
ẋT Cẋdt (36)
wsc while the energy introduced by the external forces is

␰ˆ o pt ⫽ (28)
冕冕
储 wsc 储 2␲/␻i 2␲/␻i
W e ⫽⫺ ẋT ⌬Mẍdt⫺ ẋT ⌬Kxdt (37)
In fact, when ␰ˆ ⫽ ␰ˆ o pt the value of the scalar product wsc
T
/ 储 wsc 储 0 0
ˆ
• ␰⫽1 of Eq. 共27兲 is maximum.
Not only can the designer choose the best solution, but he also Consider two cases:
can analyze the robustness of the solution. 1兲 Undamped system.
In a practical scenario model errors occur. Therefore, the real When c⫽0 共undamped system兲 the system does not dissipate
energy, namely 兩 W dU 兩 ⫽0. Since the undamped system is stable
value of ␰ˆ defers from the theoretical optimum solution ␰ˆ o pt . One 共hypothesis assumed in the sensitivity eigenvalue analysis兲 also
can take into account the uncertainties in the model by means of the work done by external forces is null 兩 W eU 兩 ⫽0. The subscript
the relationship indicates that the energies are related to the undamped system.
␰ˆ ⫽ ␰ˆ o pt ⫾⌬ ␰ˆ (29) 2兲 Unstable slightly damped system (d␭ i /dc⬎0).
When slight damping is introduced, the system starts dissipat-
By replacing Eq. 共29兲 into Eq. 共27兲, we can express the stability ing energy 兩 W dD 兩 ⬎0. This could lead one to believe that, if the
condition as a function of the uncertainty vector ⌬␰ˆ undamped system is stable, all the more so, the slightly damped
system should be stable. This paradox can be explained by means
储 ␰ˆ oT pt ⌬ ␰ˆ 储 ⬍1⫺cos共 ␪ sc 兲 (30) of the following formula presented and proved in the appendix
冏冏冏冏
Using Eq. 共30兲, given the optimum choice for the damping coef-
dW e dW d d␭ i
ficients, one can analyze the effect of the uncertainties as far as ⫽ ⫹ ␲␻ i (38)
robustness of the system is concerned. In other words, as long as dc dc dc
uncertainties satisfy Eq. 共30兲, the damped system is stable. In fact, the work done by the external force 兩 W eD 兩 ⬎0 for small
Resuming the example 共10兲, suppose now that the damping damping ␦ c is
element coefficients ␰ 11 , ␰ 22 have to be set by the designer. The
optimum solution provided by the cone algorithm is
␰ˆ o pt ⫽ 关 0.7262 0.6875兴 T , ␪ sc ⫽24.3° (31)
兩 W eD 兩 ⫽ 兩 W eU 兩 ⫹ 冏冏冏冏
dW e
dc
␦ c⫽
dW e
dc
␦c (39)
Replacing Eq. 共38兲 into 共39兲, it yields

and the damping element uncertainties have to satisfy Eq. 共30兲
0.7262⌬ ␰ 11⫹0.6875⌬ ␰ 22⬍0.0886 (32) 兩 W eD 兩 ⫽ 冉冏冏
dW d
dc
⫹ ␲␻ i
d␭ i
dc 冊
␦ c⫽ 兩 W dD 兩 ⫹ ␲␻ i
d␭ i
dc
␦c (40)
3 A General Physical Explanation of the Destabilizing Eq. 共40兲 states that, when the damped system results unstable
Effect of Damping (d␭ i /dc⬎0)⇒ 兩 W eD 兩 ⬎ 兩 W dD 兩 ⬎0. Therefore, the external forces
In this section we explain in a elegant way the destabilizing introduce an amount of energy at each cycle that is not completely
effect of damping from a physical point of view. We show that dissipated. The net energy at each cycle increases the internal
there exist modes of vibration that extract energy from the reser- energy causing the system to become unstable. Note that dissipa-
voir of the system. In order to be excited these modes need an tion energy is always required in order to have a positive work
energy sink where the system dissipates part of the energy by done by external forces. The physical explanation of the ‘‘paradox
means of damping. We will prove that, when the slightly damped of damping’’ provided in this section has general validity.
system is unstable, the energy acquired by these modes is greater
than the dissipated energy. Conclusions
The mass and stiffness matrices of system 共2兲 can be always This paper deals with asymmetric linear systems. These sys-
decomposed in symmetric positive definite matrices and non sym- tems can be found in many problems of interest in aeronautics and
metric matrices, namely M⫽Ms ⫹⌬M and K⫽Ks ⫹⌬K. There- control. In many cases, stability is studied by neglecting damping.
fore the system 共2兲 can be rewritten in the form Unfortunately, even slight damping can destabilize the system. We
introduced a new criterion for foreseeing the destabilizing effect
Ms ẍ共 t 兲 ⫹cUẋ共 t 兲 ⫹Ks x共 t 兲 ⫽⫺ 共 ⌬Mẍ共 t 兲 ⫹⌬Kx共 t 兲兲 (33) of damping. The criterion asserts that damping does not destabi-
Since Ms and Ks are positive definite, system 共33兲 can be thought lize the system if and only if a system of homogeneous linear
of as a conservative system subjected the external forces inequalities is satisfied. Moreover, we proved that the system of
⫺(⌬Mẍ(t)⫹⌬Kx(t)). inequalities can be replaced with a simpler inequality that gives a
Given the ith eigenvalue and right eigenvector of the system sufficient condition. A simple example shows the theory step by
␭ i , ui , one mode of vibration of the system is given by step. Eventually, a general physical explanation of the destabiliz-
ing effect of damping is given, using energy considerations.
1
共 u e ␭ i t ⫹ūi e ␭ i t 兲
¯
x⫽ (34)
2 i Appendix
where ūi , ¯␭i are the complex conjugated of ui , ␭ i ⫽ ␴ i ⫹i ␻ i . To Description of the Modified Cone Algorithm. The cone al-
gain a better understanding of the energy exchanges, consider gorithm finds a solution cone C(ws , ␪ s ), given an angle ␪ s . It
only the imaginary part of the eigenvalue in order to neglect the starts with an arbitrary axis w0 . The vector is then modified ac-
exponential term of the mode of vibration. In this way we con- cording to the following rule

wi⫹1 ⫽wi ⫹ãi
wi⫹1 ⫽wi
if 具 wi ,ãi 典 ⬎90°⫺ ␪ s
if 具 wi ,ãi 典 ⭐90°⫺ ␪ s
(41) 冉 2M
d␭ i
dc 冊
⫹U 共 i ␻ i ui 兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲
dui
dc
Wan 关10兴 proved that, if there exists a solution cone with angle
␪ s , the algorithm will converge to the solution in a finite number
⫽ 2M 冉 d␭ i
dc 冊
⫹U 共 i ␻ i ui 兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲
dui
dc
of steps. The algorithm can be repeated with different ␪ s . In fact,
one can start from ␪ s ⫽0. Then, by increasing the angle ␪ s in fixed
increments, one can find a series of solution cones. Each time the
⫽ 2M 冉 d␭ i
dc 冊
⫹U 共 ⫺i ␻ i ūi 兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲
dūi
dc
algorithm converges, the angle ␪ s is increased. The algorithm will
not converge when ␪ s overcomes the angle ␪ sc of the largest
solution cone C(wsc , ␪ sc ).
⫽ 2M 冉 d␭ i
dc 冊
⫹U 共 ⫺i ␻ i ui 兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲
dūi
dc
⫽0
(45)
Energy Theorem. The energy introduced by the external
forces in one period of oscillation is Comparing formula 共45兲 and 共44兲, one obtains dūi /dc⫽
冕冕
2␲/␻i 2␲/␻i
⫺dui /dc
W e ⫽⫺ ẋT ⌬Mẍdt⫺ ẋT ⌬Kxdt The increment of the energy given by the external forces for
0 0 each cycle due to an increment of c is
⫽⫺ 冕
0
2␲/␻i
ẋT 共 ⌬Mẍ⫹⌬Kx兲 dt
dW e
dc
⫽⫺ 冕 2␲/␻i
0 4 dc 冉
i ␻ i dui i ␻ t dūi ⫺i ␻ t
e i⫺
dc
e i 冊 T
共 ⫺ ␻ i2 ⌬M⫹⌬K兲
⫽⫺ 冕 2␲/␻i i ␻ i T i␻ t
共 ui e i ⫺ūiT e ⫺i ␻ i t 兲共 ⫺ ␻ i2 ⌬M⫹⌬K兲
⫻共 ui e i ␻ i t ⫹ūi e ⫺i ␻ i t 兲 dt
冕
4
0 2␲/␻i i␻i
i␻it ⫺i ␻ i t ⫺ 共 ui e i ␻ i t ⫺ūi e ⫺i ␻ i t 兲 T 共 ⫺ ␻ i2 ⌬M⫹⌬K兲
⫻ 共 ui e ⫹ūi e 兲 dt (42) 0 4
The increment of the energy dissipated by the damping elements

for each cycle due to an increment of c is ⫻ 冉dui i ␻ t dūi ⫺i ␻ t
dc
e i⫹
dc
e i dt 冊
dW d
dc
冏 c⫽0
⫽⫺ 冕 2␲/␻i
0
dẋT
dc
Cẋdt⫺ 冕 2␲/␻i
0
ẋT Uẋdt ⫽⫺ 冕 2␲/␻i
0
i ␻ i i ␻ t ⫺i ␻ t dui
4
共 e i ⫹e i 兲
dc 冉冊 T
⫺ 冕 2␲/␻i
ẋT C
dẋT
dt
⫻ 共 ⫺ ␻ i2 ⌬M⫹⌬K兲 ui 共 e i ␻ i t ⫹e ⫺i ␻ i t 兲 dt
0 dc
⫺ 冕 2␲/␻i i ␻ i i ␻ t ⫺i ␻ t T
共 e i ⫺e i 兲 u 共 ⫺ ␻ 2 ⌬M⫹⌬K 兲
dui i ␻ t
共e i
冕 2␲/␻i i i
0 4 dc
⫽⫺ ẋT Uẋdt
⫺i ␻ i t
0 ⫺e 兲 dt
⫽ 冕 2␲/␻i
0
␻ i2
4
共 uiT e i ␻ i t ⫺ūiT e ⫺i ␻ i t 兲 U共 ui e i ␻ i t ⫺ūi e ⫺i ␻ i t 兲 dt ⫽⫺ ␲ i 冉冉冊 dui
dc
T
共 ⫺ ␻ i2 ⌬M⫹⌬K兲 ui ⫺uiT
⫽ 冕 2␲/␻i
0
␻ i2 i ␻ t ⫺i ␻ t T
4
共 e i ⫺e i 兲 u Uu 共 e i ␻ i t ⫺e ⫺i ␻ i t 兲 dt
i i
⫻ 共 ⫺ ␻ i2 ⌬M⫹⌬K兲
dui
dc 冊
⫽⫺ ␲␻ i uiT Uui (43) 冉
⫽ ␲ i uiT 共 ⫺ ␻ i2 共 ⌬M⫺⌬MT 兲 ⫹ 共 ⌬K⫺⌬KT 兲兲
dui
dc 冊 (46)
In the last step we used the fact that ūi ⫽ui 苸R when c⫽0 关6兴.
Moreover, differentiating Eq. 共3兲 with respect c and replacing ␭ i Premultiplying Eq. 共3兲 by uiT and differentiating, one gets
⫽i ␻ i
d
关共 M␭ i2 ⫹C␭ i ⫹K兲 ui 兴 ⫽0⇒
冉
ūiT 2M␭ i
d␭ i
dc
⫹U␭ i ⫹C
d␭ i
dc
ui ⫹
dūiT
dc 冊
共 M␭ i2 ⫹C␭ i ⫹K兲 ui
dc
dui
⫹ūiT 共 M␭ i2 ⫹C␭ i ⫹K兲 ⫽0 (47)
d dc
关共 M␭ i2 ⫹cU␭ i ⫹K兲 ui 兴 ⫽0⇒
dc
(44) Using the normalization 2uiT Mui ⫽1 and the fact that C兩 c⫽0 ⫽0
冉冊
one can obtain for the eigenvalue ␭ i ⫽i ␻ i
d␭ i dui
2M ␭ ⫹U␭ i ui ⫹ 共 ␭ i2 M⫹K兲 ⫽0
dc i dc d␭ i dui dūiT
i ␻ i ūiT Uui ⫹i ␻ i ⫹ūiT 共 ⫺ ␻ i2 M⫹K兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲 ui
冉冊
dc dc dc
d␭ i dui
2M ⫹U 共 i ␻ i ui 兲 ⫹ 共 ⫺ ␻ i2 M⫹K兲 ⫽0 d␭ i dui duiT
dc dc
⫽i ␻ i uiT Uui ⫹i ␻ i ⫹uiT 共 ⫺ ␻ i2 M⫹K兲 ⫺
dc dc dc
It is reminded that the derivative 共44兲 is evaluated at c⫽0. Con-
jugating Eq. 共44兲 and considering the fact that d␭ i /dc苸R, and ⫻ 共 ⫺ ␻ i2 M⫹K兲 ui ⫽0 (48)
that the conjugate of the product of two complex numbers equals
the product of their conjugates, it yields Since Ms ⫽MsT and Ks ⫽KsT

uiT 共 ⫺ ␻ i2 M⫹K兲
dui duiT
dc
⫺
dc
共 ⫺ ␻ i2 M⫹K兲 ui 冏冏冏冏
dW e
dc
⫽
dW d
dc
⫹ ␲␻ i
d␭ i
dc
(52)
dui
⫽uiT 共 ⫺ ␻ i2 共 M⫺MT 兲 ⫹ 共 K⫺KT 兲兲
dc
dui References
⫽uiT 共 ⫺ ␻ i2 共 ⌬M⫺⌬MT 兲 ⫹ 共 ⌬K⫺⌬KT 兲兲 (49)
dc 关1兴 Gasparetto, A., 1998, ‘‘A System Theory Approach to Mode Coupling Chatter
in Machining,’’ ASME J. Dyn. Syst., Meas., Control, 120, pp. 545–547.
Replacing Eq. 共48兲 into Eq. 共47兲 one obtains 关2兴 Inman, D. J., 1983, ‘‘Dynamics of Asymmetric Nonconservative Systems,’’
ASME J. Appl. Mech., 50, pp. 199–203.
d␭ i dui 关3兴 Kounadis, A. N., 1992, ‘‘On the Paradox of the Destabilizing Effect of Damp-
i ␻ i uiT Uui ⫹i ␻ i ⫹uiT 共 ⫺ ␻ i2 共 ⌬M⫺⌬MT 兲 ⫹ 共 ⌬K⫺⌬KT 兲兲 ing in Non-Conservative Systems,’’ Int. J. Non-Linear Mech., 27, pp. 597–
dc dc
609.
关4兴 Semler, C., Alighanbari, H., and Paı̈doussis, M. P., 1998, ‘‘A Physical Expla-
⫽0
nation of the Destabilizing Effect of Damping,’’ ASME J. Appl. Mech., 65, pp.
Multiplying by ␲ i one obtains 642– 648.
关5兴 Bolotin, V. V., and Zhinzher, N. I., 1969, ‘‘Effects of Damping on Stability of
d␭ i Elastic Systems Subjected to Non-Conservative Forces,’’ Int. J. Solids Struct.,
⫺ ␲␻ i uiT Uui ⫺ ␲␻ i ⫹ ␲ iuiT 共 ⫺ ␻ i2 共 ⌬M⫺⌬MT 兲 16, pp. 965–989.
dc 关6兴 Herrmann, G., and Jong, I. C., 1966, ‘‘On Nonconservative Stability Problems
of Elastic Systems with Slight Damping,’’ ASME J. Appl. Mech., 32, pp.
dui 125–133.
⫹ 共 ⌬K⫺⌬KT 兲兲 ⫽0 (50) 关7兴 Herrmann, G., and Jong, I. C., 1965, ‘‘On the Destabilizing Effect of Damping
dc
in Nonconservative Elastic Systems,’’ ASME J. Appl. Mech., 32, pp. 592–597.
Therefore, comparing Eqs. 共46兲 and 共43兲 with 共49兲 the substitution 关8兴 Nemat-Nasser, S., Prasad, S. N., and Herrmann, G., 1966, ‘‘Destabilizing Ef-
gives fect of Velocity-Dependent Forces in Nonconservative Continuous Systems,’’
AIAA J., 4, pp. 1276 –1280.
dW d dW e d␭ i 关9兴 Adhikari, S., and Friswell, M. I., 2001, ‘‘Eigenderivative Analysis of Asym-
⫹ ⫽ ␲␻ i (51) metric Non-conservative Systems,’’ Int. J. Numer. Methods Eng., 51, pp. 709–
dc dc dc 733.
关10兴 Wan, J. S., 1994, ‘‘Cone Algorithm: An Extension of the Perceptron Algo-
Since the derivative of the dissipative forces is always negative, rithm,’’ IEEE Trans. Syst. Man Cybern., 24, pp. 1571–1576.
we conclude 关11兴 Murty K., 1976, Linear and Combinatorial Programming, Wiley, New York.

Marty E. Johnson The Effect of Actuator and Sensor
Assistant Professor,
Vibration and Acoustic Labs,
Mechanical Engineering,
Placement on the Active Control
Virginia Tech,
Blacksburg, VA 24061-0238 of Rotor Unbalance
Luiz P. Nascimento This paper investigates both theoretically and experimentally the effect of the location and
São Paulo State University, number of sensors and magnetic bearing actuators on both global and local vibration
São Paulo, Brazil reduction along a rotor using a feedforward control scheme. Theoretical approaches
developed for the active control of beams have been shown to be useful as simplified
Mary Kasarda models for the rotor scenario. This paper also introduces the time-domain LMS feedfor-
Rotor Dynamics Group, ward control strategy, used widely in the active control of sound and vibration, as an
Mechanical Engineering, alternative control methodology to the frequency-domain feedforward approaches com-
Virginia Tech, monly presented in the literature. Results are presented showing that for any case where
Blacksburg, VA 24061-0238 the same number of actuators and error sensors are used there can be frequencies at
which large increases in vibration away from the error sensors can occur. It is also shown
Chris R. Fuller that using a larger number of error sensors than actuators results in better global reduc-
Vibration and Acoustic Labs, tion of vibration but decreased local reduction. Overall, the study demonstrated that an
Mechanical Engineering, analysis of actuator and sensor locations when feedforward control schemes are used is
Virginia Tech, necessary to ensure that harmful increased vibrations do not occur at frequencies away
Blacksburg, VA 24061 from rotor-bearing natural frequencies or at points along the rotor not monitored by error
sensors. 关DOI: 10.1115/1.1569946兴
1 Introduction experimental investigation into the effect of sensor and actuator

location are presented. The effect of the number of sensors and
The Active Magnetic Bearing 共AMB兲 is a feedback mechanism
number of actuators, respectively, used for the Multiple Input
that supports a spinning shaft by levitating it in a magnetic field.
Multiple Output 共MIMO兲 feedforward scheme is also addressed.
Patents associated with passive, active, and hybrid magnetic bear-
The motivation for the work is to examine the optimal set up for
ings go back over 150 years and there are many thousands of
actuators and error sensors on a rotor for achieving local or global
commercial applications of AMBs in the field today 关1兴. Com-
vibration reductions.
pared to conventional rolling element and hydrodynamic bearings,
magnetic bearings have the capability for high surface speeds with
low power losses. The AMB also has the added capability for Review of Open Loop Control Strategies for Rotor Control.
active vibration control allowing for the reduction of rotor A considerable amount of literature exists on control algorithms
vibrations. used for reduction of rotor vibration. Knospe 关2兴 discussed adap-
Reduction in rotor vibration is sometimes obtained by the ap- tive on-line balancing using digital control where algorithms in
plication of an ‘‘open-loop’’ or feedforward control strategy su- the frequency domain were used for the feedforward control
perimposed on the ‘‘closed-loop’’ control strategy necessary for scheme. Knospe 关3兴 used look up tables and recommended the
rotor support. Researchers have successfully implemented these convergence coefficient approach if a ‘‘good off-line’’ model of
‘‘open loop’’ control strategies with active magnetic bearings to the system is available because the computational burden is quite
adaptively reduce synchronous rotor unbalance response and there small. In this study, the authors used three inputs where they
are considerable references in the literature regarding control weighted the rotor midspan input sensor 共the sensor closest to
strategies for achieving this. Many of these control strategies are source of unbalance兲 at 40% and weighted the other two sensors
frequency domain methods and sometime require the use of at 30% each. Mid-span vibration levels were reduced but outboard
‘‘look-up tables’’ in order to track changes in rotor speed. Also, and inboard levels actually increased. Rutland 关4兴 discussed the
there has been little or no discussion on the impact of actuator and transient response of a feedback H-infinity controller in terms of
sensor location on the performance of these active control sys- controlled rotor response. The authors compared this controller
tems. In the work presented here the filtered X-LMS algorithm, system with a synchronous feedback controller and concluded that
which is a time domain based adaptive feedforward approach used the performance of the second strategy was slightly inferior to the
widely in the active control of sound and vibration, will be tested H-infinity controller. Matsumura et al. 关5兴 addressed the reduction
as an alternative control approach for reducing rotor vibration. of vibration due to rotor unbalance using a gain scheduled
The X-LMS is a least mean squares approach where the reference H-infinity robust controller. Hisatani et al. 关6兴 presented a new
signal, typically denoted by ‘‘x,’’ is filtered before the LMS op- method for suppressing the unbalance vibration using two kinds
eration is performed of adaptive filtering techniques. The first filter separated the rotor
The goal of this paper is twofold: 共i兲 to discuss the impact and displacement signal into rotation-synchronous and rotation-
optimization of sensor and actuator location on the reduction of asynchronous components. The synchronized component of the
synchronous vibrations in a high-speed laboratory rotor when ac- rotor system, which is due to the rotor unbalance, can be cancelled
tive control strategies are used and 共ii兲 to present the time domain by the output of the second filter. The second filter was a distur-
filter X-LMS algorithm as an alternative control approach for the bance estimator which used the output of the first filter together
control of rotor vibration. Results of both an analytical and an with the impulse response of the plant measured on-line at high-
speed rotation. Industrial applications also exist. Hope 关7兴 dis-
cussed the use of two types of Active Vibration Control™ 共or
in the Journal of Vibration and Acoustics. Manuscript received August 2001; Revised AVC™兲 on a 6-stage hydrogen process compressor operating at
November 2002. Associate Editor. J. Wickert. 20.6 Mpa 共3000 psi兲. One of the AVC™ modes is an open-loop
rotating magnetic flux that is superimposed on top of the control
fluxes. The rotating AVC™ flux effectively creates a force that is
counter to the rotating unbalance force. The AVC™ flux is adap-
tive and can be applied based on the desired reduction of vibra-
tions at a particular location, either along the machine shaft or on
a base plate to reduce transmitted forces. The second AVC™
mode is a synchronous current reduction approach which makes
the rotor revolve about its inertial center, reducing the control
effort associated with unbalance to nearly zero. The filtered
X-LMS algorithm has been used successfully with magnetic bear-
ings by Piper and Calvert 关8兴 but in their application the control
system was used to actively control fluid borne noise from a cen-
trifugal pump. They used a tachometer as a reference signal and a Fig. 1 Control flow chart showing a filtered X-LMS adaptive
control system
downstream hydrophone to generate an error signal.
Review of Active Vibration Control in Beams. While there
has been considerable work involving control strategies for reduc-
Filtered X-LMS. This section will give a brief review of the
ing unbalance responses in rotors, very little work has been pre-
filtered X-LMS algorithm but the reader should refer to text books
sented addressing the effects of actuator and sensor location rela-
such as that by Widrow and Sterns 关15兴 or by Nelson and Elliott
tive to the force source. There has also been little discussion
关16兴 for a more in depth description.
concerning the number of sensors required to effectively reduce
The filtered X-LMS is a time domain algorithm 共see the control
local and global vibration in a rotating machine. There is however
flow chart Fig. 1兲 that uses a reference signal x, such as a tachom-
a substantial amount of published work on the active control of
eter signal, to drive a set of secondary actuators in order to affect
bending vibration in finite and infinite beams that does address
the system under control. The reference signal is first digitally
these points 关9,10兴. The vibration of a beam can be used as a
sampled and then passed through a finite impulse response 共FIR兲
simplified model of a vibrating rotor and is useful for analyzing
control filter W before being converted back into an analog signal
the potential performance of control systems on rotors.
y and used to drive the control actuators. Another set of sensors,
Fuller et al. 关9兴 discussed both feedback and feedforward con-
called error sensors, are used to monitor the behavior of the sys-
trol strategies for controlling the vibration on beams. While Fuller
tem 共error e兲 and are used to automatically adapt the control FIR
et al. did not specifically discuss actuator and sensor placement,
filter using the LMS algorithm. The ‘‘plant’’ represents the trans-
Nelson and Elliott 关11兴 gave a general discussion about actuator
fer function between the input to the actuators 共magnetic bearing
location in their book on the active control of sound. They showed
currents in this case兲 and the vibration detected at the error sen-
that control is always most successful when the control actuator is
sors 共eddy current proximity probes in this case兲. The disturbance
placed close to the source of the disturbance. They also showed
d is the vibration at the error sensors due to the unbalance in the
that a set of discrete sensors can be used to approximate the total
system. This control architecture differs from the LMS in that the
vibrational energy 共sound in their case兲 in a system with a larger
reference signal needs to be first filtered by a model of the plant G
number of sensors leading to a better approximation. Fuller et al.
共i.e. filtered ‘‘X’’兲 before being used by the LMS algorithm. The
关12兴 looked at simultaneous control of flexural and extensional
model of the plant is usually stored as an FIR filter 共see below兲
waves in a beam and considered the control of noise radiation
and is measured in a system identification stage before the control
from beams 关13兴. The most complete discussion of the active con-
system is turned on.
trol of vibration in beams is given in a publication by Brennan
If a signal, such as x, has been sampled at discrete time inter-
et al. 关10兴 where the authors compared different wave control and
vals then it can be considered as a sequence x(n) where n can
vibrational power minimization strategies for the control of vibra-
only take integer values. The current output of an FIR filter, such
tion on both finite and infinite beams. They showed that in finite
a W, is the weighted sum of the previous inputs. For example the
beams, global control 共i.e. reduction achieved everywhere on the
current output from the control filter y(n) can be expressed as a
beam兲 is relatively easy to achieve at beam resonances 共or critical
series of previous inputs of the reference signal y(n)
speeds兲 but difficult when away from resonance conditions. This
point has considerable relevance to the control of rotor vibrations ⫽ 兺 i⫽0
I⫺1
w i x(n⫺i)where the control filter has I filter weights (w i ).
since rotors are normally run at speeds away from resonance con- The LMS algorithm updates the filter coefficients in W using the
ditions. The paper also showed that active control using a single most recent error signal e(n) and the past I filtered reference
secondary actuator resulted in vibration reduction downstream of signals r(n⫺i).
the secondary actuator but is not very effective upstream of the w i 共 n⫹1 兲 ⫽w i 共 n 兲 ⫺ ␣ e 共 n 兲 r 共 n⫺i 兲 (1)
actuator 共often making the vibration larger兲. Post and Silcox 关14兴
showed that the minimization of vibration over a section of a All I filter coefficients can be updated this way. The coefficient ␣
beam can actually lead to significant increases in vibration away is the convergence coefficient and determines how rapidly the
from the error sensor. This has implications for error sensor control system converges. ␣ must be large enough such that the
location. convergence time is small but cannot be too large since this can
cause instability. Ideally, this algorithm converges to a solution
2 Theory where the time averaged sum of the squared error signals are
minimized. In principle, only 2 coefficients are necessary to
In this section the filtered X-LMS algorithm is briefly described achieve good control if the disturbance is at a single frequency. If
and a simple, one dimensional model of a rotor is used to inves- multiple frequencies need to be controlled, for example harmonics
tigate the performance of an active control system. The model of the rotor speed, then more than two coefficients are necessary.
developed in this section is a beam model 共i.e. does not include
gyroscopic effects兲 and is intended to act as a tool through which 1-D Model of Beam for Active Control Analysis. In this
the actuator and sensor locations can be investigated and is not section a simplified model of the rotor vibration 共or velocity兲 in
intended to accurately predict the behavior of the test rotor. In the the vertical direction will be described. The rotor will be consid-
following section a rotor finite element package that includes gy- ered as a free beam 共or rod兲 with gyroscopics neglected with
roscopic effects is used to more accurately model the behavior of masses attached and supported by bearings modeled as a pair of
the test rotor and to validate the results of the simplified model springs 共with damping included兲 as shown in Fig. 2. Both unbal-
presented here. ance forces and control forces have been introduced onto the

At a single frequency the velocities at a number of locations can
be described in matrix form as,
uiÄTijfj (5)
where the column vector ui describes the velocities at positions xi
共also a column vector兲 due to a number of forces fj acting at
locations xj . Each element in the matrix Tij is calculated using
Eq. 共4兲.
Matrix Impedance Method for Modeling Bearings. The
above theory and Eq. 共5兲 can only be used to model the motion of
the rotor itself but does not include any bearing stiffness, damping
or mass that may be supporting or supported by the rotor. These
elements will be included in the model using a matrix impedance
method 关18兴. In this paper this method will be used to describe the
addition of stiffness, damping and mass at three discrete locations
along the rotor but can be extended to include any number of
mass, stiffness and damping locations. As shown in Fig. 2 the
rotor interacts with external loads at three distinct points along the
beam, namely, at the two bearing positions where two small
masses m 1 and m 2 are attached and the rotor is supported by two
Fig. 2 Schematic of rotor system used for experiments and springs of stiffness k 1 and k 2 and in the middle of the beam where
beam used in the theoretical model. The bearing supports and a mass m 3 , containing a slight unbalance, is applied. The bearings
masses are attached at three locations along the beam. Sen- are also considered to have viscous dampers c 1 and c 2 included in
sors and actuator axes for the experimental system are also them. The mass and stiffness attachments create reaction forces
shown. when the rotor is moved and these forces can be described using
an impedance matrix Z by,
beam to determine the optimal active control performance. The

f r1
fr⫽⫺Zum⇔ f r2 ⫽ 冋册
冋册
velocity of the rotor in the horizontal direction will not be consid- f r3
ered in this model and will be assumed to be independent of the
vertical velocity. Any velocity in the horizontal direction can be k1
冋册
j ␻ m 1 ⫹c 1 ⫹ 0 0
controlled in the same way as the vertical velocity using another j␻ u m1
uncoupled active control system producing similar results.
The velocity, u, of a beam or rod can be described as a series or ⫺ k2 u m2
0 j ␻ m 2 ⫹c 2 ⫹ 0
summation of modes 关17兴. j␻ u m3
⬁ 0 0 j␻m3
u 共 x, ␻ 兲 ⫽ 兺 a 共 ␻ 兲␺ 共 x 兲
n⫽0
n n (2) (6)
Where um is the vector of velocities at the three mass locations
where the nth mode shape is a function of position x and is given
and the reaction force is given by the vector fr . The velocity
by ␺ n (x) and the complex mode amplitude of the nth mode is
vector um can be considered as the combination of the velocity
given by a n ( ␻ ). If the frequency range of interest is limited then
umr due to the reaction forces fr and the velocity umj due to ex-
it is possible to accurately describe the beam’s behavior using a
ternal input forces to the rotor fj . The external input forces in this
finite set of modes. The mode shapes are given in Appendix A.
model are created by the mass unbalance and by the active control
The mode amplitude a n ( ␻ ) is a function of angular frequency ␻
forces. Using Eq. 共5兲 and Eq. 共6兲 leads to an expression for the
and can be calculated by considering all of the forces acting on the
reaction force in terms of the external forces,
rotor. The mode amplitude due to a single force F( ␻ ) acting at a
single point x j on the rotor is given by, um⫽umj⫹umr⫽Tmjfj⫹Tmmfr⫽Tmjfj⫺TmmZum
a n共 ␻ 兲 ⫽
2
冉 j␻
m r 共 ␻ n ⫺ ␻ 兲 ⫹2 j ␨ n ␻ n ␻
2 2 冊
␺共 x j兲 F共 ␻ 兲 (3) ⇒um⫽ 关 I⫹TmmZ兴 ⫺1 Tmjfj
⫺1
(7)
⇒fr⫽⫺Zum⫽⫺Z关 I⫹TmmZ兴 Tmjfj (8)

where the natural frequency and damping ratio for the nth mode
are given by ␻ n and ␨ n respectively. The damping ratio is typi- The two matrices Tmm and Tmj contain rotor mobilities describing
cally low for a steel rotor, on the order of 0.005. Most of the the velocity at the three mass locations due to forces acting at the
damping in this system will be provided by the bearings. The total mass locations and the locations of the external forces respec-
mass of the rotor is m r ⫽lS ␳ where S is the cross sectional area of tively. The elements in these matrices can be calculated using Eq.
the rotor, ␳ is the density of the rotor material and l is the rotor 共4兲. Equations 共7兲 and 共8兲 can be used to calculate a new set of
length. The natural frequencies of the modes can be calculated augmented rotor mobilities T̂ij that include the reaction force fr of
using the equations presented in Appendix B. the masses, dampers and stiffnesses. From this augmented set of
By combining Eq. 共2兲 and Eq. 共3兲, point and transfer mobility equations the velocity at any point on the rotor due to a force
terms can be calculated. The transfer mobility T i j is defined as the applied at any other point on the rotor can be calculated. We can
velocity of the beam at position x i due to an input force at position now re-write the mobility equation 共Eq. 共5兲兲 to include reaction
x j . The point mobility is simply the velocity due to a force oper- forces.
ating at the same position on the beam i.e. x i ⫽x j .
⬁ ui⫽Tijfj⫹Timfr⫽T̂ijfj
2 j␻
T i j共 ␻ 兲⫽ 兺
m r n⫽0 共 ␻ n ⫺ ␻ 兲 ⫹2 j ␨ n ␻ n ␻
2 2 ␺共 xi兲␺共 x j 兲 (4)
T̂ij⫽ 关 Tij⫺TimZ关 I⫹TmmZ兴 ⫺1 Tmj兴 (9)

Table 1 Physical properties of rotor and bearings used for the simulation
Physical Properties
Rotor Bearing 1 Bearing 2 Unbalance mass
Legth ᐉ 18 in or Mass 0.25 kg Mass 0.25 kg Mass 0.8 kg
457 mm
Diameter d 0.375 in or Stiffness 16000 Nm⫺1 Stiffness 16000 Nm⫺1 - -
9.52 mm k1 k2
Modulus of 207 GNm⫺2 Damper 50 Nm ⫺1
s Damper 50 Nm ⫺1
s - -
elasticity E c1 c2
⫺3
Density ␳ 7800 kgm Position 4.5 in or Position 14.5 in or 368 Position 6.5 in or
Poisson’s 0.3 x1 114 mm x2 mm xm 165 mm
ratio ␯
The two matrices Tim and Tij contain rotor mobilities describing can in principle perfectly control the vibration at a single position
the velocity at the observation locations xi due to forces acting at on the rotor 共or beam兲 but it may cause the vibration levels at
the mass locations and the locations of the external forces. Using other locations to be greatly increased.
this augmented set of equations the performance of an active con- In this section the results for a single representative configuration
trol system can be evaluated. of mounts and unbalance mass will be presented. The physical
parameters used for these simulations are given in Table 1 and are
Active Control Performance. In this section a method for similar to the known values of the experimental system 共see sec-
calculating the control performance of an active control system tion 4兲. The unbalance in the system will be generated at the mass
will be presented. An active control system using the filtered mounted on the rotor shaft between the active magnetic bearings
X-LMS system will attempt to minimize the sum of the squared but closer to the inboard 共left兲 bearing x⫽6.5 in or 0.165 m. The
outputs from the error sensors. If a control system has J control magnetic bearings act as both mounts and actuators and are lo-
actuators and L error sensors then at a single frequency the L cated at x⫽4.5 in or 0.114 m and at x⫽14.5 in or 0.368 m. x
length vector of errors e can be written in terms of the vector of ⫽0 refers to the beginning of the rotor at the motor end 共see Fig.
primary disturbance forces fp 共i.e. rotor unbalance兲 and the J 2 and Fig. 3兲.
length vector of secondary control forces fc . The results in this section 共Fig. 4 and Fig. 5兲 will be presented
e⫽T̂epfp⫹T̂ecfc (10) for excitation frequencies between 0 and 200 Hz such that the
vibration level plotted at each frequency represents the vibration
The two matrices T̂ep and T̂ec contain augmented rotor mobilities level as if the rotor was run at that frequency and the excitation
describing the velocity at the error locations xe due to forces act- was caused by a unit unbalance force. It is not intended to show
ing at the locations xp of the unbalance forces 共primary distur- the control at all of these frequencies simultaneously since this is
bance兲 and the locations xc of the control forces. The sum of the not how the rotor would be excited in practice.
squared error signals can be calculated using the Hermitian trans-
pose 共or conjugate transpose denoted by H) as eHe and is mini- Mode Shapes and Natural Frequencies. The augmented
mized when the secondary control forces fc are given by 关9,11兴, mobility of the rotor given in Eq. 共9兲 can be considered to contain
augmented natural frequencies and augmented mode shapes. The
⫺1 H
fc ⫽ 关 T̂H
ecT̂ec兴 T̂ecT̂epfp (11) mode shapes and natural frequencies of the first three modes are
– opt shown in Fig. 3. It can be seen that all three of these modes have
Using these optimal forces the velocity at any set of observation components that are due to bending in the rotor and components
locations xi along the rotor can then be calculated before ui b and due to bouncing or rocking on the two mounts.
–
after ui a optimal control as, Limitations Due to Actuator Location. In this section the
–
limitation due to actuator location will be investigated. To achieve
ui b⫽T̂ipfp (12) this it will be assumed that the control system is designed to
–
minimize the global vibration level. For the purposes of investi-
ui a⫽T̂ipfp⫹T̂icfc (13)
– – opt
The two matrices T̂ip and T̂ic contain augmented rotor mobilities
describing the velocity at the observation locations xi due to
forces acting at the locations xp of the unbalance forces 共primary
disturbance兲 and the locations xc of the control forces. This pro-
cess can be repeated for a range of frequencies.
3 Theoretical Analysis
The performance of a control system implemented on a rotor
will be limited by: 共i兲 the control architecture, 共ii兲 the actuator
placement and 共iii兲 the sensor placement. In this section the limi-
tations caused by actuator placement and sensor placement will be
investigated using the theory presented in section 2. The term
‘‘global vibration level’’ will be used in this section and refers to
the mean velocity squared level averaged over the entire length of
the rotor 共at 20 equally spaced points兲. Although it is often more
important to reduce the vibration levels at particular positions
along the rotor, the global vibration level will provide some indi-
cation as to the overall performance of the system and highlight Fig. 3 The mode shapes and natural frequencies of the first
potential problems. Since the excitation is tonal, a single actuator three modes of the rotor predicted using a beam model.

Fig. 4 The global vibration level, when the global vibration
level is the cost function, using actuator 1, actuator 2 and then
both actuators. Fig. 6 Velocity along the rotor at 100 Hz with and without con-
trol using actuator 1 and a single error sensor e1 , actuator 1
and both error sensors and then both actuators and both error
sensors.
gating the effect of actuator location alone it is assumed that a
‘‘perfect’’ error sensor, that can directly measure the global vibra-
tion level, is used 共not realizable in practice兲. deliberately run at speeds away from resonance there are likely to
In the first example the magnetic bearing to the left of the be control problems using single actuator systems. These difficul-
unbalance mass is used as a control actuator and is driven such ties can be overcome if two actuators are used. Near resonance
that the vibration averaged over the entire shaft 共at twenty points conditions reductions are possible using a single actuator but the
for this simulation兲 is minimized. In this example the unbalance location is very important. For example near the first resonant
mass is located closer to the left mount at x⫽6.5 in or 0.165 m. mode, actuator 2 has very poor performance.
Figure 4 shows the global vibration level 共averaged over twenty The performance of an active control system is often dependent
points兲 over the entire shaft with and without 共thick solid line兲 on the number of actuators vs. the number of degrees of freedom
control. Three cases are considered, 共i兲 when actuator 1 close to that are important in the system under control 共or complexity of
the unbalance mass is used 共dashed line兲, 共ii兲 when actuator 2 is the system兲. This complexity can be roughly approximated by the
used 共dotted line兲 and 共iii兲 when both actuators are used simulta- modal overlap factor 关11兴 that describes the number of modes
neously 共thin solid line兲. The results are plotted for all frequencies acting within a given bandwidth. For beams this term does not
between 0 and 200 Hz and this represents the best control possible increase very rapidly with frequency 共as compared to two and
at each of these frequencies 共i.e. limitations in performance are three dimensional structures兲 and even while coupled to the
due purely to the actuator location and unbalance location兲. mounts the rotor seems to maintain this modal simplicity 共i.e. the
The first thing to be noted from these results is that the actuator modal overlap factor remains low兲. It is therefore likely that two
near to the unbalance mass is much more effective than the actua- well positioned actuators will be sufficient to achieve good vibra-
tor farther from the unbalance mass. This demonstrates that it is tion control over a fairly large frequency band.
important to locate the actuator as close to the source of distur-
bance as is possible. It can also be seen that at frequencies away Using Discrete Error Sensors. In any realistic control sys-
from resonance or critical frequencies, for example at 80 Hz using tem only a limited number of sensors can be used and it is there-
actuator 1, performance can be very poor. Since rotors are often fore important to investigate the use of discrete error sensors on
the performance of the control system. For this study, sensor lo-
cations were determined based on accessibility to the shaft as
dictated by physical constraints. Figure 5 shows the global vibra-
tion level when the vibration is cancelled at a single point (x
⫽11 in or 0.279 m兲 downstream of the unbalance mass using a
single upstream actuator 共dashed line兲. Unless the system is ex-
cited close to a resonance frequency the global vibration level can
be substantially increased after control. For example the vibration
at 120 Hz increases globally by 6 dB. If a second error sensor is
introduced at the end of the rotor (x⫽18 in or 0.457 m兲 then the
control system minimizes the average squared vibration level at
the two error sensor locations 共dotted line兲 and this helps to con-
dition the control system and reduces the likelihood of increased
vibration. The introduction of the second error sensor greatly im-
proves the performance of the system, and global performance is
similar to the case with the ideal global error sensor 共see Fig. 4兲.
As will be shown later this type of system can also be used to
reduce the vibration over an area on the rotor.
Figure 6 shows the vibration along the rotor with and without
control at 100 Hz using both a single error sensor and then both
Fig. 5 The global vibration level when actuator 1 is used. The sensors. Since this vibration has a complex value the absolute
two cases shown are with error sensor 1 and then with both value of the velocity is plotted. Before control 共heavy solid line兲
error sensors used as part of the cost function. the vibration is dominated by the second mode that is a rocking

Rotor Dynamic Analysis Using a Transfer-Matrix Analysis.
In order to more accurately predict the experimental rotor’s dy-
namic behavior, the rotor-bearing system was modeled by using a
transfer-matrix method based rotor analysis code 关19兴. Figure 2
shows the schematic of the experimental rotor. In the model the
shaft was considered to be a flexible beam divided into elements
having distributed mass and elasticity. The unbalance mass 共or
disk兲 was modeled as a lumped mass, and its properties were
added to the element where it is attached. The equivalent stiffness
and damping of the magnetic bearings is dependent on the
‘‘closed loop’’ or feedback control system 共PID controller兲 used to
position the rotor in the magnetic field. These stiffness and damp-
ing values can be calculated as a function of the overall system
transfer function G(i ␻ ), and the position stiffness K x and the
current stiffness K i of the bearings. The overall controller transfer
function expresses the relationship between the output control cur-
rent and input shaft position, and can be written as,
G 共 i ␻ 兲 ⫽a G 共 ␻ 兲 ⫹ib G 共 ␻ 兲 (14)
This transfer function is composed of components from the posi-

tion sensor, low pass filter, PID filter and power amplifier. The
equivalent stiffness and damping are then given by,
K eq ⫽K x ⫹K i a G 共 ␻ 兲 ; (15)
C eq ⫽K i b G 共 ␻ 兲 / ␻ (16)
The position stiffness and current stiffness are given as a function

of the physical characteristics of the magnetic bearing 关20兴. Since
the transfer function of the control circuit has a frequency depen-
dence, the equivalent stiffness and damping is also given as a
function of the frequency, so that the average values of these
parameters can be calculated in the frequency band of interest,
using the procedure presented by Clements 关21兴.
Finding the modal characteristics of the system provides useful
information that can help optimize the active control configura-
tion. There is great interest in being able to control the vibration of
the rotor near critical speeds 共or resonances兲 and the vibration
mode shapes may also indicate the best location for placement of
the errors sensors. Figure 7 shows the first three vibration mode
shapes of the rotor with the unbalance mass placed at the three
experimental locations along the length of the rotor: at midspan,
close to the inboard bearing, and close to the outboard bearing.
Also these simulations 共note: results presented in Hz兲 show that
considerable changes in the natural frequencies and vibration
Fig. 7 The mode shapes and natural frequencies for the first mode shapes of the rotor occur as a function of the unbalance
three modes of the rotor for the three different experimental mass location. These simulation results validate the theory pre-
mass configurations. sented in section 2 since there is a very close match in natural
frequencies and mode shapes with the Transfer-Matrix model
共note: results presented in rpm兲.
type mode 共note that the absolute value is plotted兲. If actuator 1 is
used to cancel the vibration at a single error sensor 共1兲 the overall
vibration level is not significantly controlled and vibration prob-
lems can still occur at positions away from the error sensor
共dashed line兲. If two downstream error sensors are used then glo-
bal vibration reduction is still not large 共dotted line兲 but the down-
stream vibration section, where the error sensors are located, has
the lowest vibration levels. This implies that the use of multiple
sensors can allow vibration reduction over a region even if global
control is not possible. When both actuators are used to cancel the
vibration at the pair of downstream error sensors 共light solid line兲
very effective isolation of this section of the rotor is achieved. It
has been shown 关10兴 that for infinite beams 共using wave theory兲
all downstream wave propagation can be removed if the velocity
and gradient of velocity in the x-direction are simultaneously re-
duced to zero at one point along the beam. These results imply
that even in circumstances where global control is difficult to
achieve, a pair of actuators can be effective in isolating sections of Fig. 8 Picture of experimental rig used in this study. The two
a rotor. magnetic bearings and unbalance mass are shown.

Table 2 Different Error Sensors and Actuators Locations for Unbalance Control
Cases Error Sensors Actuators Unbalance Mass Position

1 S 1 and S 2 V 1 and W 1
3 S 1 , S 2 , S 3 and S 4 V 1 and W 1
4 S 1 , S 2 , S 3 and S 4 V 1 , W 1 , V 2 and W 2
8 S 1 , S 2 and SV 1 V 1 and W 1
9 S 1 , S 2 and SV 1 V 2 and W 2
4 Experimental Results two independent actuators. A summary of the different control

configurations used for the nine cases is shown in Table 2. It
Experimental Set-Up. Figure 8 shows a picture of the small should be noted that using two actuators 共i.e. one magnetic bear-
high-speed test rotor used in this study. The rotor consists of a 457 ing acting in one plane兲 and two error sensors 共also in one plane兲
mm 共18 in.兲 long shaft that is 9.52 mm 共0.375 in.兲 in diameter and
in the experimental case would be similar to using a single sensor
is driven by an electric motor through a flexible coupling. The
actuator pair in the theoretical analysis presented in section 3 be-
rotor is capable of operating at rotating speeds as high as 10,000
rpm and its rotor dynamic characteristics mimic that of large-scale cause of the added degree of freedom in the real rotor 共i.e., both
turbomachinery. A steel disk with holes for the placement of un- horizontal and vertical motion is permitted兲.
balance weights has been attached to the shaft at various locations Prior to control, the model of the plant G was measured by
to vary the amount and location of rotor unbalance forces. The injecting white noise into the actuators and measuring the output
disk is 76.2 mm 共3 in兲 in diameter, 25.4 mm 共1 in兲 thick and of the error sensors. This allowed the system to operate, if neces-
weighs 811 gr. 共1.78 lbs兲. In addition to the large unbalance disk sary, over a range of frequencies. The experimental tests for un-
there are two additional masses 共250 gr.兲 which are the bearing balance control were carried out at the rotating speed of 3420 rpm
ferromagnetic rotors that are also attached to the shaft. The rotor or 57 Hz, which is close to the first critical speed of the rotor, as
is supported in two 8-pole heterepolar Active Magnetic Bearings calculated in the rotor dynamic analysis. Figure 9 shows the au-
共AMBs兲 with a 12 lbs force capacity. Each of the AMBs consists tospectrum of the displacement measured at the four sensor loca-
of a radial rotor, a stator, a position sensor ring and a backup tions with and without control for Case 1. As shown in Fig. 9,
bearing. The radial rotor and the stator are used to levitate the large reductions in unbalance vibration at the error sensor posi-
rotor while the sensor ring measures radial position of the shaft
tions is achieved. The 48 dB in attenuation at these positions
for feedback. The backup bearing provides support to the rotor
shows that the filtered X-LMS system is extremely effective 共a
when the bearing is turned off or in case of a power loss. Each
Radial AMB has two axes of control, each utilizing a PID control 250 fold decrease in vibration level兲. To analyze the effect of
algorithm to maintain support of the shaft. The control axes are control, the resulting vibration reduction for all 9 cases at all
oriented at 45 degrees to the horizontal 共denoted V and W兲 as measurement positions is summarized in Table 3. For Cases 1– 6
shown in Fig. 2. all eight sensor signals were recorded and for cases 7–9 three
The sensors available for use as error sensors are the four eddy sensor signals were recorded. Also presented in Table 3 is an
current sensors located at two points of the shaft listed as S 1 , S 2 , estimate of the ‘‘global’’ reductions in vibration that have been
S 3 and S 4 as well as the displacement sensors already used for calculated by considering the sum of the squared outputs from all
feedback control of each of the support bearings shown as V 1 , eight proximity probes.
W 1 , V 2 , and W 2 in Fig. 2. All sensors, even if they are not used Included in Table 3 is a list of the control performance pre-
as error sensors, are used for rotor displacement monitoring pur- dicted using the computational model presented in section 2 and 3
poses. A single eddy current sensor mounted at the drive end of for Cases 1 to 6. A number of control trends match those of the
the shaft is used as a key phase reference signal to drive the theoretical model.
control system 共denoted K p ). The positions of the sensors were to
some extent limited by physical constraints. i. If a pair of error sensors is positioned at a single location
The multi-channel filtered X-LMS active control algorithm was along the rotor then very good local control can be achieved. This
implemented on a quad TMS C40 board made by Spectrum Sig- is clearly illustrated by cases 1, 2, 5 and 7 where typically ⬎40 dB
nal, running five C40 DSP chips in parallel. This system was reduction in vibration level was measured.
developed at Virginia Tech and can accommodate a wide variety ii. Good local control does not correspond to global reduc-
of control configurations i.e. various numbers of reference signals, tions. This is well illustrated by case 6 and especially case 7 where
actuators and errors signals. a large reduction at the error sensor location is achieved while
Experimental Results of Unbalance Control. In this work other sections of the rotor undergo an increase in vibration 共i.e.
nine different control configurations were used where different shown as negative reduction in Table 3兲.
numbers and locations of sensors and actuators were used with iii. Actuators positioned close to the disturbance achieve better
different locations of the unbalanced disk. Both inboard and out- control. This is clearly illustrated by comparing the results of case
board magnetic bearings were used as actuators. Each bearing had 5 to case 6. In case 5, where the actuator closest to the unbalance
two uncoupled active perpendicular axes and were considered as mass is used, large reductions are achieved across the rotor 共15.1

Conclusions
A theoretical and experimental investigation into the effect of
the location and number of sensors and actuators on both global
and local vibration reduction along a rotor using a feedforward
control scheme with magnetic bearing actuators has been com-
pleted. Theoretical approaches developed for the active control of
beams have been shown to be useful as simplified models for the
rotor scenario for optimization studies involving sensor and actua-
tor location and number. Also presented is the time-domain fil-
tered X LMS feedforward control strategy, used widely in the
active control of sound and vibration, as an alternative control
methodology to the frequency-domain open loop approaches com-
monly presented in the literature. This control scheme performed
extremely well with reductions in vibration amplitudes as high as
51.7 dB 共a 380 fold decrease兲.
The rotor under analysis consists of a small 共9.52 mm diameter兲
high-speed laboratory rotor supported in active magnetic bearings
with one large disk that can be placed in different locations. Both
experimental results and simulation results are presented in Table
3. In general,
1. A reduced model simulation method was demonstrated to be
successful at predicting overall behavior trends on a rotor when
examining the impact of sensor and actuator configuration
schemes.
2. Very good reduction of local vibrations can be achieved at
the error sensor when a single error sensor is used 共over 50 dB兲
but not without risk of increased vibration at other locations and
Fig. 9 The measured vibration level at the four measurement hence poor global attenuation.
sensors S1, S2, S3 and S4 for case 1. Huge reductions in the 3. Using a larger number of error sensors than actuators results
vibration level at the error sensors „S1 and S2… were achieved in better global reduction of vibration but decreased local reduc-
„48 dB or 250 fold reduction…. tion. For example, by increasing the number of error sensors from
2 to 4 共cases 1 and 3兲 the global attenuation increased from 16.8
dB to 17.8 dB. Although some local reductions were not as large
dB global兲 whereas in case 6 the reductions are much lower 共only the lower global attenuation reflects the fact that the control is
3.5 dB global兲 and increases in vibration were measured at one more evenly spread along the rotor.
point along the rotor. 4. For any case where the same number of actuators as error
iv. Using a larger number of error sensors than actuators results sensors are used there can be frequencies at which large increases
in better global reduction of vibration but worse local reduction. in vibration away from the error sensors can occur.
This is illustrated by comparing the results from Case 1 and Case 5. There may be advantages for hardware reduction and/or re-
3. Local reduction at the error sensor drops from 48 dB to 28 dB duced control burden to go with a scheme of more error sensors
but the global attenuation increases from 16.8 dB to 17.8 dB im- and reduced actuators with little reduction in overall performance
plying a more even reduction in vibration along the rotor. at frequencies near resonance.
v. Using a larger number of actuators improves performance:
This is illustrated comparing Case 4 to Case 3 where improved Overall, the study demonstrated that an analysis of actuator and
vibration reductions were achieved 共19.1 dB as compared to 17.8 sensor location when active control schemes are used to reduce
dB兲. However, it should be cautioned that in any case where the rotor vibration is necessary at the design phase to insure that
same number of actuators as error sensors are used 共‘‘square sys- harmful increased vibrations do not occur at points along the rotor
tem’’兲 there may be frequencies at which large increases in vibra- not monitored by error sensors or unexpectedly at frequencies that
tion away from the error sensors can occur. In addition, note that do not correspond to rotor-bearing natural frequencies.
there is not a dramatic improvement between Case 3 and Case 4
indicating that a configuration with more sensors and a reduced Acknowledgments
number of actuators may be advantageous from a reduced hard- The authors would like to acknowledge the National Science
ware or reduced control burden standpoint. Foundation 共NSF兲 for supporting the work presented in this paper
Table 3 Unbalance Frequency Reduction „dB…-3420 RPM „57 Hz…
Simulation results: Attenuation

Experimental results: Attenuation 共dB兲共dB兲
Cases SV 1 SW 1 S1 S2 S3 S4 SV 2 SW 2 Global SV 1 , SW 1 S1 , S2 S3 , S4 SV 1 , W 1
1 13.2 15.7 48.8 48.1 17.8 22 6.8 14.3 16.8 7.4 ⬁ 16.8 11.1
2 3.5 11.7 49.4 51.7 11.3 15.8 3.4 3.4 9.8 16.1 ⬁ 16.1 4.8
3 10.4 14.5 23.2 28.5 20 26.6 8.9 15.1 17.8 13.0 22.1 23.3 6.8
4 15.2 19.8 34.8 30.5 29.2 29.3 7.8 9.4 19.1 16.7 ⬁ ⬁ 17.6
5 8.7 14.1 41.6 50.8 18.9 28.1 11.2 20.3 15.1 18.0 ⬁ 19.9 7.2
6 3.5 6.5 8.1 16.4 6.2 4.4 ⫺9.3 ⫺8.2 3.4 4.5 ⬁ ⫺3.0 ⫺20.8
7 ⫺3.8 41.2 41.7
8 2.2 14.1 11.4
9 25.6 30.6 29.6

under Grant No. DMI-9733124. The authors would also like to where k n is the wavenumber of the n th mode, E is the modulus of
thank Dr. Gordon Kirk and the Virginia Tech Rotordynamics elasticity, I is the second moment of area and d is the rotor diam-
Laboratory for the use of equipment and rotor analysis codes. eter. Small correction factors can be added to equation 23, for the
Besides, the authors would like to acknowledge the FAPESP— lower order modes, to improve the accuracy of the calculation
Fundação de Amparo á Pesquisa do Estado de São Paulo, Brazil, 关17兴.
which granted funds to Luiz P. Nascimento for a Visiting Scholar
Program at Virginia Tech.
Appendix A: Mode Shapes for a Free-Free Beam References

关1兴 Kasarda, M. E. F., 2000, ‘‘An Overview of Active Magnetic Bearing Technol-
The first two modes mode of a free-free beam are rigid body ogy and Applications,’’ invited review paper, Public Admin. Rev., 32共2兲, pp.
modes with the higher modes given as combinations of sine, co- 91–99.
sine and hyperbolic functions 关17兴. 关2兴 Knospe, et al., 1993, ‘‘Adaptive On-Line Rotor Balancing Using Digital Con-
trol,’’ Proceedings of ’93 ROMAG, pp. 153–164.
␺ 0 共 x 兲 ⫽1 (17) 关3兴 Knospe et al., 1998, ‘‘Synthesis Techniques for Robust Adaptive Vibration
Control,’’ 6th International Symposium on Magnetic Bearings, Cambridge,
2x Mass, p. 183.
␺ 1 共 x 兲 ⫽1⫺ (18) 关4兴 Rutland et al., 1994, ‘‘Comparison of Controller Designs for Attenuation of
ᐉ vibration in a Rotor-Bearing System Under synchronous and Transient condi-
冉冊冉冊
tions,’’ 4th International Symposium on Magnetic Bearings, Zurich, Switzer-
x x land, pp. 107–112.
␺ n 共 x 兲 ⫽cos ␥ ⫺0.5 ⫹k cosh ␥ ⫺0.5 关5兴 Matsumura et al., 1994, ‘‘Elimination of Unbalance Vibration in AMB Sys-
ᐉ ᐉ tems Using Gain Scheduled H ⬁ Robust Controllers,’’ 4th International Sym-
posium on Magnetic Bearings, Zurich, Switzerland, pp. 113–118.
for n⫽2,4,6,¯ (19) 关6兴 Hisatani et al., 1997, ‘‘Adaptive Filtering for Unbalance Vibration Suppre-
sion,’’ Proceedings of MAG’97, Alexandria, VA, pp. 125–130.
where 关7兴 Hope, R. W., Tessier, L. P., Knospe, C., and Miyaji, T., 1998, ‘‘Adaptive
␥ Vibration Control of Industrial Turbomachinery, 98-GT-405,’’ International
Gas Turbine & Aeroengine Congress & Exhibition.
sin
2 ␥ ␥ 关8兴 Piper, G. E., and Calvert, T. E., 1995, ‘‘Active Fluidborne Noise Control of a
k⫽⫺ and tan ⫹tanh ⫽0 Magnetic Bearing Pump,’’ NCA-Vol. 21, IMECE, Proceedings of the ASME
␥ 2 2 Noise Control and Acoustics Division, pp. 55–76.
sinh 关9兴 Fuller, C. R., Elliott, S. J., and Nelson, P. A., 1996, Active Control of Vibration,
2 Academic Press, London.
冉冊冉冊
关10兴 Brennan, M. J., Elliott, S. J., and Pinnington, R. J., 1995, ‘‘Strategies for the
x x
␺ n 共 x 兲 ⫽sin ␥ ⬘ ⫺0.5 ⫹k ⬘ sinh ␥ ⬘ ⫺0.5 Active Control of Flexural Vibration on a Beam,’’ J. Sound Vib., 186共4兲, pp.
ᐉ ᐉ 657– 688.
关11兴 Nelson, P. A., and Elliott, S. J., 1993, Active Control of Sound, Academic
for n⫽3,5,7,¯ (20) Press.
关12兴 Fuller, C. R., Gibbs, G. P., and Silcox, R. J., 1990, ‘‘Simultaneous Active
where Control of Flexural and Extensional Waves in Beams,’’ J. Intell. Mater. Syst.
Struct., 1, pp. 235–247.
␥⬘ 关13兴 Guigou, C., and Fuller, C. R., 1993, ‘‘Active Control of Sound Radiation from
sin a Simply Supported Beam: Influence of Bending Near-field Waves,’’ J. Acoust.
2 ␥⬘ ␥⬘ Soc. Am., 93共5兲, pp. 2716 –2725.
k ⬘⫽ and tan ⫺tanh ⫽0 关14兴 Post, J. T., and Silcox, R. J., 1990, ‘‘Active Control of the Forced Response of
␥⬘ 2 2
a Finite Beam,’’ Proc. Of Noise-Con’90, Austin, Texas, pp. 197–202.
sinh
2 关15兴 Widrow, B., and Sterns, S. D., 1985, Adaptive Signal Processing, Prentice
Hall, Englewood Cliffs, New Jersey.
关16兴 Elliott, S. J., Stothers, I. M., and Nelson, P. A., 1987, ‘‘A Multiple Error LMS
Appendix B: Natural Frequencies of the Modes of a Algorithm and Its Application to the Active Control of Sound and Vibration,’’
Free-Free Beam IEEE Trans. Acoust., Speech, Signal Process., 35, pp. 1423–1434.
关17兴 Warburton, G. B., 1954, ‘‘The Vibration of Rectangular Plates,’’ Proc. Inst.
For a free-free rod the natural frequencies of the modes are Mech. Eng., 168, p. 371–383.
given by 关17兴, 关18兴 Bishop, R. E. D., and Johnson, D. C., 1960, The Mechanisms of Vibration,
Cambridge University Press.
␻ 0 ⫽0 (21) 关19兴 Childs, D., 1993, Turbomachinery Rotordynamics Phenomena, Modeling, and
Analysis, John Wiley & Sons, Inc., New York.
␻ 1 ⫽0 (22) 关20兴 Allaire, P. E., et al., 1993, Magnetic Bearings, STLE Handbook of Tribology
and Lubrication, Vol. III.
␻ n ⫽k 2n 冑 EI 共 n⫺0.5兲 2 ␲ 2 d
S␳
⫽
4ᐉ 2
冑 E
␳
for n⭓2 (23)
关21兴 Clements, J. R., 2000, ‘‘The Experimental Testing of an Active Magnetic
Bearing/Rotor System Undergoing Base Excitation,’’ Master Thesis, Virginia
Tech.

Hemanshu R. Pota
e-mail: [email protected] Robust Control of a 2D Acoustic
Ian R. Petersen Enclosure*
School of Electrical Engineering,
UNSW at the Australian Defense Force Academy, This paper reports experimental results in the application of feedback control of acoustic
Canberra ACT, 2600, Australia noise in a 2D enclosure. It is shown that the feedback control of 2D enclosures poses
peculiar problems compared to simple 1D ducts. The chief among them are the need for
explicit budgeting of uncertainties and dealing with high model orders. This paper dem-
Atul G. Kelkar onstrates a practical way to design controllers for such systems using minimax LQG
Department of Mechanical Engineering, methods. It is shown that an important step in the controller design is the proper choice
Iowa State University, of a weighting function. The results presented in this paper are impressive and they can be
Ames, IA 50011 further improved by proper choice of actuator and sensor placements.
e-mail: [email protected] 关DOI: 10.1115/1.1570447兴
1 Introduction hinges which provide access to the inside of the enclosure to

position the microphones. The experiment is so set that there are
Active reduction or control of acoustic noise has significant
two speakers and one microphone. One speaker is used as a dis-
practical applications 关1兴. Successful active noise control schemes
turbance and the other speaker is the control speaker. There is one
mostly use adaptive feedforward control 关1–3兴. Feedforward con-
microphone to sense the acoustic disturbance and provide feed-
trol is ideal in the situations where a signal strongly correlated
with the noise can be directly measured. There are many practical back to the controller. The feedback control system is a SISO
applications, e.g., structure induced vibrations, where it’s difficult system with the disturbance coming from an independent speaker
to obtain a signal correlated with acoustic noise which can be used located on one end of the enclosure.
to effect a feedforward control scheme. In these situations it is The 2D enclosure is 1850 mm⫻320 mm⫻1240 mm. Let the
fruitful to apply feedback control for active noise reduction. origin 共0, 0, 0兲 be at the bottom left corner when one is facing the
The essentials of modeling and control of a 1D duct have been door side of the enclosure; z-axis is up, y-axis is into the enclo-
discussed in 关4,5兴. Experimental results using feedback have been sure, and the x-axis is along the length of the enclosure as seen in
obtained for 1D ducts; see 关6,7兴 for results without robust control Fig. 2. The disturbance speaker is located at one end of the enclo-
approach while the work in 关5,8 –10兴 uses robust control theory to sure with the center at 共0, 127, 406兲 mm; the control speaker is
design controllers. It is well-known 关11兴 that for feedforward con- located at 共600, 320, 430兲 mm; and the microphone is located at
trol, perfect model matching is essential to provide satisfactory 共588, 200, 420兲 mm.
noise cancellation. In feedback control, a mismatch in the model 2.2 System Identification and Nominal Modeling. Experi-
can lead to an unstable system. This is the main reason for the mentally identified model is used in this paper to design the con-
reluctance to use feedback control for acoustic noise reduction. troller. The difficulty in obtaining an analytical model for enclo-
This means that the proper use of robust control theory is essential sures with high modal densities is well-known 关14兴. The use of
to obtain results which have practical applications. For example, finite element and other numerical methods is an active area of
the significant improvement of feedback controller performance in research 关15,16兴. From the results in 关15,16兴 it can be clearly seen
关9兴 over the other reported feedback controllers is due to its ex- that the accuracy of these models have a long way to go before
plicit accounting of modelling uncertainties. they can be used for feedback controller design. The error is so
All the results in the literature using feedback control 关5–7,9兴 large that robust controllers designed to maintain closed-loop sta-
are reported for 1D ducts only. In principle, control of 1D ducts is bility will be ineffective. Most analytical and numerical methods
no different from 2D enclosures. But in practice there is much predict the modes quite accurately. The main difficulty in model-
difference. Firstly it’s a lot easier to get an excellent match be- ling is the uncertainty in the damping terms which makes accurate
tween identified models and experimental data for 1D ducts. Sec- modelling for high modal density enclosures very difficult. In 1D
ondly the model order of 1D ducts is significantly lower. In this enclosures like acoustic ducts or enclosures with low modal den-
paper the practical aspects of noise reduction in 2D enclosures are sity, analytical models can be used to design very effective con-
highlighted based on experimental results. The minimax LQG trollers 关4兴. This is the reason why many experimental setups to
feedback control 关12,13兴 is used to design practical controllers for demonstrate active control for 3D enclosures choose small enclo-
a 2D acoustic enclosure at the Iowa State University 共ISU兲 shown sures 关17兴, this keeps the modal density low hence the modes well
in Fig. 1. apart. The problem of obtaining accurate acoustical models with
unknown dampings and high modal densities is a very difficult
2 Experimental Setup and Modeling one which is currently an active research area.
Significant advance has been made in active cancellation of
2.1 Experimental Setup. A picture of the experimental acoustic noise using feedforward techniques 关18,19兴. These meth-
acoustic enclosure to be considered in this paper is shown in Fig. ods need an accurate estimate of the error path transfer function.
1. A schematic of the enclosure is shown in Fig. 2. This is the path from the secondary 共control兲 sources to the error
The enclosure is constructed from aluminum sheets. One side microphones or the locations where the acoustic noise has to be
of the enclosure 共with the largest cross-section兲 has two doors on reduced. There are two major differences between the use of the
error path transfer function and the acoustical model for feedback
*This work was supported by the Australian Research Council 共H. R. Pota & I. R. control design. Firstly, most practical implementations of feedfor-
Petersen兲, the NSF through Grant No. CMS:9713846, and NASA through Grant No. ward noise cancellations work best for tonal noise and this means
NAG-1-01039 共A. G. Kelkar兲
that the model needs to be accurately known only at a few par-
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received December ticular frequencies. Most of the digital filters implementing these
2001; Revised February 2002. Associate Editor: R. F. Keltie. models are adaptively tuned too. The second difference is that the
Fig. 1 The experimental 2D enclosure
order of these error path transfer functions can be chosen to be

very high. It’s common to have an order of 256 and up. It must be Fig. 3 Uncertain system representation
realised that the design of feedback controllers with such high-
order models is nearly impossible. For these practical reasons ex-
perimentally identified models are used in this paper.
resents the noise input and it is assumed that the effects of the
To identify the system model, two separate frequency responses
uncertainty show up at the sensor output through the disturbance
are recorded using Stanford Research’s SR785 spectrum analyzer.
channel. The minimax LQG method requires that the inputs due to
One response is taken from the disturbance speaker to the micro-
uncertainty 共i.e., the output of the ⌬(s) block兲 and the noise enter
phone and the other from the control speaker to the same micro-
through the same channel. Figure 3 shows one particular uncer-
phone. The subspace identification technique 关20兴 is then used
tainty model but the minimax LQG method is not restricted to this
to get a state-space representation of this two-input-one-output
model alone and alternative representations are possible.
system.
The block ⌬(s) in Fig. 3 is chosen such that
After the identification process, we end up with a nominal
transfer function matrix of the form
兩 ⌬ 共 j ␻ 兲兩 ⭐1᭙ ␻ , (1)
P 共 s 兲 ⫽ 关 P 1 共 s 兲 P 2 共 s 兲兴 and W(s) is a stable frequency weighting transfer function. The
controller design section will discuss the importance of choosing
where P 1 (s) represents the transfer function from the disturbance
this weighting function properly.
speaker input to the microphone output and P 2 (s) represents the
Let the true transfer function from the control speaker to the
transfer function from the control speaker input to the microphone
output. microphone be given by P̃ 2 (s). From the block diagram in Fig. 3
it can be seen that
2.3 Uncertainty Modeling. The system representation
which forms the starting point in the minimax LQG controller 共 P̃ 2 共 s 兲 ⫺ P 2 共 s 兲兲 / P 1 共 s 兲 ⫽⌬ 共 s 兲 W 共 s 兲 .
design is shown in Fig. 3. In the figure y is the microphone output,
u is the control speaker input, z is the uncertainty output, w̃ rep- To restrict 兩 ⌬( j ␻ ) 兩 ⭐1᭙ ␻ , the weighting function needs to be
chosen such that
冏 P 1共 j ␻ 兲
冏
P̃ 2 共 j ␻ 兲 ⫺ P 2 共 j ␻ 兲
⭐ 兩 W共 j ␻ 兲兩 ᭙ ␻ . (2)
The bound 共2兲 is an inequality bound on the magnitude of W(s)

and there are several functions which will satisfy this bound. In
this paper the function on the left-hand-side of the bound 共2兲 is
computed in the frequency range of interest from the experimental
measurements and the identified system models. From these func-
tions a magnitude envelope is constructed and finally that magni-
tude envelope is matched by a transfer function obtained using the
Yule-Walker method 关21,22兴.
The weighting function W(s) in the experiments reported in
this paper is mainly used to account for the modelling uncertain-
ties. In general it can include parameter uncertainties too. In the
case of acoustical system this would mean the change in tempera-
ture and the enclosure geometry. From the experimental and iden-
tified models in Figs. 7 and 8 it can be seen that the modelling
uncertainty is much larger than the uncertainty due to parameter
variations. As a matter of fact these experiments were run over
Fig. 2 Schematic of the experimental 2D enclosure several days. It is common for the microphones to be slightly

shifted during the experiments. The speakers and the enclosure
geometry was fixed though. The experiments were conducted un-
der realistic conditions giving confidence in the robustness of the
controller to uncertainties.
3 Minimax LQG Control

This section presents a brief description of the minimax LQG
robust controller synthesis method. A more complete and rigorous
description of this method can be found in the references 关12,13兴.
The minimax LQG method is applied to uncertain systems of the
form shown in Fig. 4. In this figure, the nominal system is de-
scribed by the following stochastic state equations:
ẋ⫽ 共 Ax⫹B 1 u⫹B 2 ␾ 兲 ⫹B 2 w̃,
Fig. 5 The scaled H ⴥ control problem
z⫽C 1 x⫹D 1 u,
y⫽C 2 x⫹D 2 ␾ ⫹D 2 w̃, y 共 0 兲 ⫽0, (3)
It is well known that the solution to this H ⬁ control problem can
n m
In the above equations, x(t)苸R is the state, u(t)苸R is the
control input, w̃(t) is a unity covariance white noise input, z(t) be obtained in terms of the following pair algebraic Riccati equa-
苸Rq is the uncertainty output, ␾ (t)苸Rp is the uncertainty input tions 共e.g., see 关23兴兲:
and y(t)苸Rl is the measured output.
The uncertainty block can be any dynamical system satisfying a 共 A⫺B 2 D ⬘2 共 D 2 D 2⬘ 兲 ⫺1 C 2 兲 Y ⬁ ⫹Y ⬁ 共 A⫺B 2 D ⬘2 共 D 2 D ⬘2 兲 ⫺1 C 2 兲 ⬘
general uncertainty constraint; see 关12,13兴. In particular, this un-
certainty constraint is satisfied by the uncertainty block ⌬(s) in
equation 共1兲.
冉
⫺Y ⬁ C ⬘2 共 D 2 D ⬘2 兲 ⫺1 C 2 ⫺
1
R Y
␶ ␶ ⬁ 冊
It is assumed that the cost function under consideration is of the ⫹B 2 共 I⫺D 2⬘ 共 D 2 D 2⬘ 兲 ⫺1 D 2 兲 B 2⬘ ⫽0 (6)
form 共E共.兲 is the expected value兲
冕 T
and
1
J⫽ lim E 共 x 共 t 兲 ⬘ Rx 共 t 兲 ⫹u 共 t 兲 ⬘ Gu 共 t 兲兲 dt, (4) X ⬁ 共 A⫺B 1 G ⫺1 ⫺1 ⫺1
␶ ⌼ ␶⬘ 兲 ⫹ 共 A⫺B 1 G ␶ ⌼ ␶⬘ 兲 ⬘ X ⬁ ⫹ 共 R ␶ ⫺⌼ ␶ G ␶ ⌼ ␶⬘ 兲
T→⬁ 2T
冉冊
0
1
where R⭓0 and G⬎0. The minimax LQG control problem in- ⫺X ⬁ B 1 G ⫺1
␶ B 1⬘ ⫺ B B ⬘ X ⫽0, (7)
volves finding a controller which minimizes the maximum of this ␶ 2 2 ⬁
cost function where the maximum is taken over all uncertainties where the solutions are required to satisfy the conditions Y ⬁ ⬎0,
satisfying the uncertainty constraint 共1兲. If we define a variable
X ⬁ ⬎0, I⫺1/␶ Y ⬁ X ⬁ ⬎0 and R ␶ ⫺⌼ ⬘␶ G ⫺1 ␶ ⌼ ␶ ⭓0. Here R ␶ ,R
␨⫽ 冋册
R 1/2x
G 1/2u
, (5)
⫹ ␶ C 1⬘ C 1 , G ␶ ,G⫹ ␶ D 1⬘ D 1 and ⌼ ␶ , ␶ C 1⬘ D 1 . In order to solve
the minimax LQG control problem, the parameter ␶ ⬎0 is chosen
to minimize the cost bound W ␶ 共an upper-bound on J in 共4兲兲
then the minimax LQG control problem can be solved by solving defined by
冋册
the scaled H ⬁ control problem represented in Fig. 5; see 关12,13兴.
In this H ⬁ control problem, the nominal system is described by 共 ␶ Y C T2 ⫹B 2 D T2 兲共 D 2 D T2 兲 ⫺1
Eq. 共3兲 and 共5兲 and the controller is to be constructed such that the W ␶ ,tr ⫻ 共 ␶ C 2 Y ⫹D 2 B 2 兲 X 共 I⫺Y X 兲 ⫺1 .
T
(8)
closed loop system is stable and the transfer function from w̃(t) to ⫹␶Y R␶
␰ (t) satisfies the H ⬁ norm bound
Then, the minimax LQG controller is defined by the state
储 T w̃ ␰ 共 j ␻ 兲储 ⭐1 ᭙ ␻ . equations
x̂˙ ⫽ 共 A⫺B 1 G ⫺1 冉 ⫺1
␶ ⌼ ⬘␶ 兲 x̂⫺ B 1 G ␶ B ⬘1 ⫺
1
B B ⬘ X x̂
␶ 2 2 ⬁ 冊
⫹ I⫺ 冉冊
1
Y X
␶ ⬁ ⬁
⫺1
共 Y ⬁ C 2⬘ ⫹B 2 D 2⬘ 兲 ⫻ 共 D 2 D 2⬘ 兲 ⫺1
冉冉
⫻ y⫺ C 2 ⫹
1
D B ⬘ X x̂
␶ 2 2 ⬁ 冊冊
u ␶ ⫽⫺G ⫺1
␶ 共 B ⬘1 X ⬁ ⫹⌼ ⬘␶ 兲 x̂. (9)
4 Controller Design
The appropriate state-space representation in Eq. 共3兲 are arrived
at from P 1 (s), P 2 (s), and W(s) as discussed previously. Note
that the theory of 关12,13兴 requires that D 2 D 2⬘ ⬎0 in 共3兲. This is
achieved by adding a small measurement noise to the system in
addition to the process noise w̃(t). We choose the matrix R in the
cost function 共4兲 as R⫽C 2⬘ C 2 . That is, the term x(t) ⬘ Rx(t) in the
cost function 共4兲 corresponds to the norm squared value of the
Fig. 4 Stochastic uncertain system nominal system output. The term u ⬘ Gu in the cost function 共4兲 is

treated as a design parameter affecting controller gain. However,
in all cases it was found that setting G to the small value of G
⫽10⫺8 did not lead to excessive controller gains.
Note that with the above choice of plant model 共3兲 and cost
function 共4兲, the nominal LQG problem essentially amounts to the
problem of minimizing the noise energy at the microphone posi-
tion when the system is subject to a white noise disturbance en-
tering the system through the control input channel.
The minimax LQG controller is synthesised by first choosing
the constant ␶⬎0 to minimize the quantity W ␶ defined in Eq. 共8兲.
With this value of the parameter ␶, the controller is constructed
according to the formula 共9兲. The order of this controller will be
the sum of the order of the nominal plant model P(s) together
Fig. 6 Two dimensional duct feedback controller setup
Fig. 7 Experimental and identified disturbance speaker response
Fig. 8 Experimental and identified control speaker response

Fig. 9 Uncertainty bound „dashed-20 log 10円 W „ j ␻ … 円 ,solid-20 log 10円 P̃ 2 „ j ␻ …
À P 2„ j ␻ …円Õ 储 P 1„ j ␻ …円…
with the order of the weighting filter W(s). Typical order of con- 5 Experimental Results
trollers designed in this paper is around eighty. Such a high order
controller may lead to problems of numerical error and excessive The reduced dimension controller designed in Section 4 was
computational load when implemented. Hence, the balanced con- implemented on a dSPACE DS1103 system as shown in Fig. 6.
troller reduction method described in Section 19.1.1 of 关23兴 is The spectrum analyser is used to measure the closed loop
applied to reduce the controller order to around forty. It was found frequency response from the disturbance speaker input to the
that with this level approximation, there was very little degrada- microphone output. In order to implement each controller, it
tion in the predicted closed loop performance. was first discretized using the FOH method with a sample period
Fig. 10 LQG minimax controller

Fig. 11 Nyquist plot of the loop-gain
of 0.5⫻10⫺3 seconds. The resulting discrete time controller From the figures it can be seen that even a 40th order model is
was then implemented on the dSPACE system with this sample unable to match the data exactly.
period. The uncertainty envelope W(s) and the magnitude of the 40th
5.1 Control of Low Frequency Modes. Experimental fre- order transfer function, for the identified models in Figs. 7 and 8,
quency response data was collected from 20–500 Hz. It is not derived using Yule-Walker method is shown in Fig. 9.
practical to fit a model over the entire frequency range. A 40th For the case under discussion the controller order is 80 and
order model was fitted in the 20–100 Hz frequency range. Figures balanced model reduction technique 关23兴 was applied in order to
7 and 8 show the experimental response and the identified re- obtain a 40th order approximation to the 80th order controller. The
sponse for the disturbance and control speakers, respectively. designed controller is shown in Fig. 10.
Fig. 12 Simulated closed loop and open loop response

Fig. 13 Experimental closed loop and open loop response
The predicted stability and robustness of the closed loop system method is used to demonstrate the controller performance between
can be seen by the Nyquist plot based on the final 共reduced di- 250– 450 Hz. The identified and experimental responses are
mension兲 controller transfer function and the measured plant shown in Figs. 14 and 15. The choice of uncertainty envelope and
transfer function data P̃ 2 ( j ␻ ). This Nyquist plot is shown in Fig. the experimental closed loop and open loop performance is shown
11. The disturbance attenuation performance of the controller is in Figs. 16 and 17, respectively. The frequency response outside
illustrated in Fig. 12. of the range shown in these figures remained unchanged. There is
Figure 13 shows the resulting measured frequency response for a clear improvement in the damping of the resonant peaks in the
the open loop and closed loop system. frequency range of interest.
5.2 Control of Higher Frequency Modes. The experimen- 5.3 Proper Choice of the Uncertainty Model. It is shown
tal results presented thus far were restricted to the lower frequency in Section 3 that the minimax LQG control problem is equivalent
range of 20–100 Hz. In this section the minimax LQG control to the minimization of the H ⬁ norm of the closed-loop transfer
Fig. 14 Experimental and identified control speaker response

Fig. 15 Experimental and identified disturbance speaker response
function between the noise input w̃(t) and output ␰ (t) as defined actuators and sensors can be used to get rid of deep nulls. In other
in Fig. 5. The output ␰ (t) consists of two terms, one of them is the words optimal actuator-sensor placement methods should use
uncertainty output z(t) which is also known as performance out- modelling error as a measure of optimality.
put. To minimize the H ⬁ norm the controller will attempt to keep The influence of the choice of W(s) on the controller can be
the output z(t) small. From this we realize that W(s) 共whose used to shape the controller. For example, the frequency band
output is z(t)) should be kept small for high authority control. In where 兩 W( ␻ ) 兩 is large, the controller gain will be low. In the
obtaining models for acoustic enclosures it is seen that the mod- closed-loop response in Fig. 13 it can be seen that there is an
elling error is the largest at deep nulls. A proper placement of increase in the gain near the deep null around 70 Hz. This is due

À P 2„ j ␻ …円Õ 储 P 1„ j ␻ …円…

Fig. 17 Experimental open loop and closed loop response
to the extreme difficulty in finding a good match between the 6 Conclusions

identified model and the experimental data. To shape the closed
loop response at 70 Hz the choice of frequency weighting function This paper successfully demonstrates that feedback control can
W(s) can be such that it is very conservative above 70 Hz as be applied to very high order acoustical systems for active noise
shown in Fig. 18. The conservative choice of the envelope at control. Due to the practical difficulty in implementing very high
frequencies greater than 70 Hz ensures that the controller doesn’t order controllers it is essential that a small frequency band be
have high gain in that region. The experimental response of the selected for noise attenuation. The chosen minimax LQG control
controller designed with this method is shown in Fig. 19. From method gives the flexibility to choose the frequency weighting
the figure it’s clear that a proper selection of the weighting func- function W(s) such that the controller targets a specified fre-
tion helps in shaping the closed loop response. quency band. Overall the experimental results in this work are

À P 2„ j ␻ …円Õ 储 P 1„ j ␻ …円…

Fig. 19 Experimental closed loop and open loop response
very impressive and further work in the selection of actuator- 关12兴 Petersen, I. R., Ugrinovski, V., and Savkin, A. V., 2000, Robust Control Design
sensor placement 共to obtain a tighter fit between the model and using H ⬁ Methods, Springer-Verlag, London.
关13兴 Ugrinovskii, V. A., and Petersen, I. R., 1998, ‘‘Time-averaged Robust Control
data兲 and a choice of frequency weighting function can improve
of Stochastic Partially Observed Uncertain Systems,’’ Proceedings of the IEEE
the results significantly. Conference on Decision and Control, IEEE, Tampa, FL.
关14兴 Fang, B., Kelkar, A. G., and Joshi, S. M., 2002, ‘‘Modelling and Control of
References Acoustic-Structure Interaction in 3-D Enclosures,’’ IEEE 2002 Conference on
Decision and Control, Las Vegas, Nevada.
关1兴 Elliott, S. J., 1999, ‘‘Down with Noise,’’ IEEE Spectrum, June, pp. 54 – 61. 关15兴 Ding, W. P., and Chen, H. L., 2001, ‘‘A Symmetrical Finite Element Model for
关2兴 Hu, J.-S., Yu, S.-H., and Hsieh, C.-S., 1998, ‘‘Application of Model-Matching Structure-Acoustic Coupling of an Elastic, Thin-Walled Cavity,’’ J. Sound
Techniques to Feedforward Active Noise Control Design,’’ IEEE Trans. Con- Vib., 243共3兲, pp. 547–559.
trol Syst. Technol., 6共1兲, pp. 33– 42. 关16兴 Marburg, S., Beer, H.-J., Gier, J., Hardtke, H.-J., Rennert, R., and Perret, F.,
关3兴 Omoto, A., and Elliott, S. J., 1999, ‘‘The Effect of Structured Uncertainty in 2002, ‘‘Experimental Verification of Structural-Acoustic Modelling and De-
the Acoustic Plant on Multichannel Feedforward Control Systems,’’ IEEE sign Optimization,’’ J. Sound Vib., 252共4兲, pp. 591– 615.
Trans. Speech Audio Process., 7共2兲, pp. 204 –212. 关17兴 Banks, H. T., Brown, D. E., Smith, R. C., Metcalf, V. L., Wang, Y., and Silcox,
关4兴 Pota, H. R., and Kelkar, A. G., 2001, ‘‘Modelling and Control of Acoustic
R. J., 1994, ‘‘Noise Control in a 3-D Structural Acoustic System: Numerical
Ducts,’’ ASME J. Vibr. Acoust., 123共1兲, pp. 2–10.
and Experimental Implementation of a PDE-Based Methodology,’’ IEEE Pro-
关5兴 Kelkar, A. G., and Pota, H. R., 2000, ‘‘Robust Broadband Control of Acoustic
ceedings of the 33rd Conference on Decision and Control, Lake Buena Vista,
Duct,’’ Proceedings of the 39th IEEE Conference on Decision and Control,
Sydney, Australia, pp. 4485– 4490. FL, pp. 305–310.
关6兴 Hong, J., Akers, J. C., Venugopal, R., Lee, M.-N., Sparks, A. G., Washabaugh, 关18兴 Nelson, P. A., and Elliott, S. J., 1992, Active Control of Sound, Academic
P. D., and Bernstein, D. S., 1996, ‘‘Modeling, Identification, and Feedback Press, London.
Control of Noise in an Acoustic Duct,’’ IEEE Trans. Control Syst. Technol., 关19兴 Elliott, S., 2001, Signal Processing for Active Control, Academic Press, Syd-
4共3兲, pp. 283–291. ney.
关7兴 Clark, R. L., and Cole, D. G., 1995, ‘‘Active Damping of Enclosed Sound 关20兴 McKelvey, T., Akçay, H., and Ljung, L., 1996, ‘‘Subspace-based Multivariable
Fields Through Direct Rate Feedback Control,’’ J. Acoust. Soc. Am., 97共3兲, pp. System Identification from Frequency Response Data,’’ IEEE Trans. Autom.
1710–1716. Control, 41共7兲, pp. 960–979.
关8兴 Erwin, R. S., and Bernstein, D. S., 1997, ‘‘Discrete-time H 2 /H ⬁ Control of an 关21兴 Friedlander, B., and Porat, B., 1984, ‘‘The Modified Yule-Walker Method of
Acoustic Duct: Delta-domain Design and Experimental Results,’’ Conference ARMA Spectral Estimation,’’ IEEE Trans. Aerosp. Electron. Syst., 20共2兲, pp.
on Decision and Control, IEEE, San Deigo, CA, pp. 281–282. 158 –173.
关9兴 Petersen, I. R., and Pota, H. R., 2000, ‘‘Minimax LQG Optimal Control of an 关22兴 Petersen, I. R., Pota, H. R., and Jahromi, M. R. S., 2002, ‘‘System Identifica-
Experimental Acoustic Duct,’’ IEE Control 2000 Conference, Cambridge, UK. tion, Uncertainty Modelling and Actuator Placement in the Robust Control of
关10兴 Petersen, I. R., 2001, ‘‘Multivariable Control of Noise in an Acoustic Duct,’’ an Acoustic Duct,’’ Proceedings of the Conference on Information, Decision,
Proceedings of the European Control Conference ECC2001, Porto, Portugal. and Control, Adelaide, Australia.
关11兴 Pota, H. R., and Kelkar, A. G., 2001, ‘‘On Perfect Acoustic Noise Cancelling 关23兴 Zhou, K., Doyle, J. C., and Glover, K., 1996, Robust and Optimal Control,
Control,’’ Control and Intelligent Systems, 29共2兲, pp. 48 –54. Prentice Hall, NJ.

Active Vibration Suppression With
Time Delayed Feedback
Various active vibration suppression techniques, which use feedback control, are imple-
mented on the structures. In real application, time delay can not be avoided especially in
Rifat Sipahi the feedback line of the actively controlled systems. The effects of the delay have to be
thoroughly understood from the perspective of system stability and the performance of the
Nejat Olgac* controlled system. Often used control laws are developed without taking the delay into
account. They fulfill the design requirements when free of delay. As unavoidable delay
Mechanical Engineering Department,
appears, however, the performance of the control changes. This work addresses the sta-
University of Connecticut,
bility analysis of such dynamics as the control law remains unchanged but carries the
Storrs, CT 06269
effect of feedback time-delay, which can be varied. For this stability analysis along the
delay axis, we follow up a recent methodology of the authors, the Direct Method (DM),
which offers a unique and unprecedented treatment of a general class of linear time
invariant time delayed systems (LTI-TDS). We discuss the underlying features and the
highlights of the method briefly. Over an example vibration suppression setting we declare
the stability intervals of the dynamics in time delay space using the DM. Having assessed
the stability, we then look at the frequency response characteristics of the system as
performance indications. 关DOI: 10.1115/1.1569942兴
1 Introduction In short, the dynamics of a typical active vibration suppression

system needs to be carefully evaluated with special consideration
Active vibration suppression has long been implemented on
of delay. The unforced dynamics of the system is:
structures under oscillatory excitations in various ways 关1,3,4兴. In
general, a partial or full state feedback control law is utilized in
order to achieve good suppression performance. Most generally
ẋ⫽Ax⫹Bx共 t⫺ ␶ 兲 ␶ 苸R⫹ (5)
this is represented as a multi input-multi output 共MIMO兲 state
and it has to be studied from the stability perspective. Equation 共5兲
space form of:
represents a widely utilized class of linear time invariant retarded
ẋ⫽Ax⫹B̄u⫹F (1) time delayed systems 共LTI-RTDS兲. This general class of systems
has been extensively investigated for over four decades 关9–19兴.
where u is the full state feedback, given by: The deployment of a recently developed Direct Method 共DM兲
关2,22兴 which yields unique observations on the system stability
u⫽Kx (2)
and the ensuing vibration suppression performance of Eq. 共4兲
⇒ẋ⫽ 共 A⫹B…x⫹F (3) form the main theme of the present text.
Until the Direct Method 共DM兲 the stability posture of the LTI-
Here x( n⫻1 ) is the state vector, A( n⫻n ) , B̄( n⫻m ) are the system and TDS along the axis of time delay was not completely resolved.
the control matrices, respectively, K( m⫻n ) is the full state feedback That is, a complete set of values of delay which cause stable or
gain matrix, F( n⫻1 ) is the excitation 共disturbance兲 force acting on unstable behavior were not available through a systematic and
the structure, and B⫽B̄K. Utilizing the full state feedback law Eq. practical methodology. 关18兴 gives an extensive review of the state-
共2兲, vibration suppression performance can easily be improved of-the art in this venue, including the techniques developed up to
within the frequency range of the excitation forces F. 2001. Most of the later methods convert the LTI-TDS into a dis-
It is very well known, however, that the control systems are tributed time delay equation via an integral transformation as ex-
commonly subject to time delay, which converts Eq. 共3兲 into a plained in 关20兴. They indicate that there are some additional dy-
retarded time delayed system 共RTDS兲 which is given as: namics appearing due to this transformation. Consequently, the
new dynamics invites additional characteristic roots, which im-
ẋ⫽Ax⫹Bx共 t⫺ ␶ 兲 ⫹F ␶ 苸R⫹ (4) pose adverse restrictions on the range of time delay. Other meth-
ods, which are mainly based on various Lyapunov-Krasovski
functional 关18,20–21兴, bring conservative solutions for the stable
where ␶ is the time delay influencing the system dynamics. regions of time delay. A recent publication 关19兴 compares five
It is critical to understand that the time delay may appear due to different approaches 共all developed in the 90s兲 based on Lyapunov
the inherent nature of the core dynamics 关5– 6兴 as well as the and LMI 共linear matrix inequality兲 techniques. It is rather intrigu-
feedback control 关1–2,7– 8兴. Regardless of the reason of the delay ing to observe that even the best stability margin that these meth-
in the dynamics ods can provide is more than 50% conservative with respect to the
共a兲 the underlying mathematics 共i.e., retarded time delayed sys- true margin 共which is detected by following a tedious Nyquist
tems兲 remains the same. The novel treatment presented in this text method兲. In contrast, the DM 关22兴 coincides with the Nyquist
is applied to vibration suppression problems, although it is generic results precisely, furthermore it is numerically very efficient.
enough to be considered for a wide variety of TDS’s. These features make the Direct Method, extremely compelling.
共b兲 it may deteriorate the quality of vibration suppression and Another shortfall of the existing methods is that, they all start
in the worst case it can even drive the dynamics into instability. with a stable dynamics for nondelayed 共i.e., delay⫽0兲 case. And
they try to assess the first positive delay value, which brings in-
*Author of correspondence. E-mail: [email protected]
stability. This value forms the so-called, ‘‘delay margin’’ of the
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Feb. 2002; system and the stability analysis stops at that point. However, as
Revised Nov. 2002. Associate Editor: J. Wickert. realized by many investigators 关23–26兴, there can be more than
one such stable pocket in delay domain. This knowledge is ex- nary axis. The corresponding ␶ values are 兵 ␶ kᐉ 其 , k⫽1, . . . m,
tremely important, as one can imagine. If there are so called, the ᐉ⫽0, . . . ⬁ which are distributed periodically 共by 2 ␲ / ␻ ck .兲 This
secondary stability pockets one may opt to operate within one of proposition implies that the infinitely many characteristic roots of
these, still performing a stable control despite the larger time de- Eq. 共6兲 can only move across the imaginary axis through these m
lay in the picture. No existing methodology, however, could offer ‘pinholes’ ( ␻ ck i,k⫽1, . . . m).
a procedure to detect these additional pockets. The Direct Method Proposition II: As ␶ transits through one of the 兵 ␶ kᐉ 其 , k
共DM兲关22兴 suggests a unique construct ultimately yielding an ex- ⫽1, . . . m, ᐉ⫽0, . . . ⬁ values, the respective root can pass
plicit function of delay, which declares completely the stability through the pinhole ␻ ck i only in one direction. That is, the root
pockets. The end results clearly offer much stronger revelations tendency 共RT兲 which is defined as
than simply stating the stability margin. The method is the first of
its kind as it handles the general n-dimensional LTI-RTDS 共with
commensurate time delays兲 and uncovers its unique properties.
Some of these properties are realized in 关23兴 for scalar dynamics
(n⫽1), however the approach therein is quite restrictive to imple-
RT 兩 ␻␶ ⫽
⫽ ␻ ck
␶kᐉ
⫽sgn Re
冋冉 ds
d␶
冏 s⫽ ␻ ck i
␶ ⫽ ␶ kᐉ
冊册 k⫽1, . . . m
ᐉ⫽0, . . . ⬁
(7)
ment on a general LTI-RTDS 共i.e. n⬎1). For these reasons, the

is invariant of the particular delay ␶ ⫽ ␶ k ⫹p2 ␲ / ␻ ck , p
Direct Method is unique in the dynamic systems and controls
⫽0, . . . ⬁ which generates the crossing ␻ ck i.
community, opening number of new directions for future research.
What this proposition conceptually implies is that all of the m
We wish to summarize the strong features of the Direct Method
pinholes for the possible passages of the characteristic roots
here to better prepare the reader:
through the imaginary axis are guarded by ‘‘one-way check
i. It does not have to start with a stable system for zero delay. valves.’’
ii. It is exact, i.e. the limits of the stability pockets are precise. In Section 2, we introduce the steps of DM, and two proposi-
They are not conservative results, which is a concern for the tions, as well as the novel stability analysis, the ‘‘Direct Method,’’
peer methodologies 关9–10,18 –21,23兴. are followed by an application on a realistic vibration control
iii. All stable pockets are declared exclusively, in one explicit problem, in Section 3.
function of time delay.
In what follows we briefly review the procedure suggested by
the DM leaving the details to 关22兴. The system in Eq. 共5兲 has a 2 The Direct Method
transcendental characteristic equation: For completeness of the treatment, we wish to review the Direct
n Method 共DM兲 briefly here. The stability assessment of the system
CE 共 s, ␶ 兲 ⫽det共 sI⫺A⫺Be ⫺ ␶ s 兲 ⫽ 兺 a 共 s 兲e
j⫽0
j
⫺ j␶s
⫽0 with ␶ ⬎0 共5兲 is needed in the domain of ␶ 共the only free parameter in 共5兲兲.
DM starts with detecting all the crossing frequencies ␻ ck (k
(6) ⫽1, . . . m) as described in Proposition I and the corresponding
where a j (s)’s are polynomials in ‘‘s’’ of degree n⫺ j for retarded time delays
time delayed dynamics. Clearly CE(s, ␶ ) has infinitely many fi-
nite characteristic roots, which move on the complex plane as ␶ 2␲
varies. As indicated by the feature 共i兲 above, this system could be ␶ kᐉ ⫽ ␶ k0 ⫹ ᐉ ᐉ⫽0, . . . ⬁ (8)
␻ ck
unstable when free of delay. The number of unstable roots for this
case, can be determined using, for instance, Routh’s array. As a which create these crossings.
matter of fact, these roots are nothing but the n eigenvalues of The following structured steps are performed next for the DM
A⫹B and they can also be calculated numerically to display the stability analysis:
natural response characteristics of the system, for ␶⫽0. The inter- 共1兲 All imaginary roots and generator ␶’s are assembled as:
esting outcome of this work could be that, if a system is unstable
for ␶⫽0, it may return to stability as ␶ increases. This would be a
关 ␻ ck , 兵 ␶ kᐉ 其兴 , k⫽1, . . . m, ᐉ⫽0, . . . ⬁
beneficial effect of the time delay on the dynamics.
As ␶ increases these n characteristic roots start moving, along
with infinitely many other roots 共because of the transcendentality 共2兲 The set of common root tendencies RT at each one of these
of Eq. 共6兲兲, what we call the ‘‘secondary’’ roots. These secondary m crossing frequencies are evaluated, RT 兩 ␻ ck . As per Proposition
roots are dormant, stable roots for ␶⫽0 共i.e. they are all in the II, this is invariant of the value of 兵 ␶ kᐉ 其 , k⫽1, . . . m.
form of ⫺⬁⫿ai 共a finite and real兲, and they do not contribute to 共3兲 All 兵 ␶ kᐉ 其 , k⫽1, . . . m, ᐉ⫽0, . . . ⬁ are tabulated in an ar-
the natural response of the system兲. But they play a very critical ray, in ascending order, along with the ␻ ck ’s and RT 兩 ␻ ck ’s.
role when ␶⬎0, because they creep into the finite domain and 共4兲 The number of unstable roots (NU) is determined for ␶⫽0
become influential. They may even destroy the stability of the 共non-delayed case兲. Call this number NU(0).
dynamics. 共5兲 Take the next ␶ kᐉ in ascending order and check RT 兩 ␻ ck . It is
Obviously for certain values of ␶⬎0 some of the infinitely obvious that, for RT⫽⫹1, NU increases by 2, and for RT
many characteristic roots cross the imaginary axis 共excluding ⫽⫺1, NU decreases by 2. Repeat this step for the following ␶’s
those dynamics that are stable independent of time delay兲, causing until the particular target value of delay ␶ is reached.
a transition from stable to unstable half of the complex plane 共or 共6兲 Identify those regions in ␶ where NU( ␶ )⫽0 as ‘stable’ and
vice versa兲. These crossing frequencies (⫿ ␻ i) and the ␶ values others as ‘unstable’. Also, state the NU( ␶ ) as the number of un-
generating them must all be detected exhaustively for a complete stable roots in this interval of ␶.
stability analysis. This procedure, indeed, lends itself to a novel explicit expres-
At this point, we state two very critical propositions from 关22兴, sion of NU( ␶ ), owing to the Propositions I and II:
which eventually yield the three conclusions marked i, ii, iii
兺 ⌫冉冊
above. We will recite them here for clarity, leaving their proofs m
to 关2,22兴. ␶ ⫺ ␶ k0
NU 共 ␶ 兲 ⫽NU 共 0 兲 ⫹ •U 共 ␶ , ␶ k0 兲 •RT k (9)
Proposition I: The time delayed system in Eq. 共5兲 can have k⫽1 ⌬␶k
only a finite number of imaginary characteristic roots 共call these
⫿ ␻ ck i, k⫽1, . . . m), for all values of ␶ 苸R⫹ . Note that, the where NU(0) is the number of unstable roots when ␶⫽0,
subscript ‘c’ refers to the ‘crossing’ of the roots over the imagi- U( ␶ , ␶ k0 )⫽A step function in ␶ with the step taking place at ␶ k0

Table 1 The stability table
␶关s兴 ␻ 关rad/s兴 RT NU
0 0
0.3071 7.5032 ⫹1 2
0.5441 4.3864 ⫹1 4
1.1444 7.5032 ⫹1 6
1.9766 4.3864 ⫹1 8
1.9818 7.5032 ⫹1 10
2.0028 3.0446 ⫺1 8
2.0052 2.98 ⫺1 6
] ] ] ]
the Direct Method of stability for LTI-RTDS. Now, we demon-

Fig. 1 Dynamic model strate how to deploy the DM for the active vibration suppression
applications by way of example.
3 Active Vibration Suppression Using Time Delayed
Full State Feedback
再 0⬍ ␶ ⬍ ␶ k0
0 Let’s take a realistic vibration suppression setting given in Fig.
1 from 关27兴 where the mass m 11 is excited harmonically and the
U 共 ␶ , ␶ k0 兲 ⫽ 1 for ␶ ⭓ ␶ k0 , ␻ ck ⫽0 response of m 12 , x 12 , is desired to be minimized. This model is a
2 ␶ ⭓ ␶ k0 , ␻ ck ⫽0 simplified display of two high-rise buildings (m 11 and m 21) inter-
connected via two passageways 共m 12 and m 22 兲. Two actuators
⌫(x)⫽Ceiling function of x, ⌫ returns the smallest integer greater deploy the feedback control, which is determined using pole
than or equal to x placement techniques.
The expression NU( ␶ ) of Eq. 共9兲 requires the knowledge of We alter the system parameters from the original example in
four things, all of which are known: order to make the numerical results compact enough to include
共i兲 NU(0) here:
共ii兲 ␶ k0 , k⫽1, . . . m, the smallest ␶⬎0 values corresponding to
each one of the ␻ ck ’s m 11⫽0.2, m 12⫽0.15, m 21⫽0.2, m 22⫽0.15 kg
共iii兲 ⌬ ␶ k ⫽2 ␲ / ␻ ck , k⫽1, . . . m c 1 ⫽2.2, c 2 ⫽1.9 N•s/m
共iv兲 RT 兩 k ⫽RT 兩 ␻ ck , k⫽1, . . . m
The stable regions of ␶ 共i.e. the pockets兲 are those which dis- k 10⫽2, k 11⫽4, k 12⫽2, k 20⫽4, k 21⫽2, k 22⫽2 N/m
play NU( ␶ )⫽0. That is ␶s⫽ 兵 ␶ 苸R⫹ ,NU( ␶ )⫽0 其 . This completes The governing equation of this controlled system is obtained as:
冢冣冢冣冉冊
0 1 0 0 0 0 0 0 0 0
⫺30 ⫺11 10 0 0 0 0 11 ⫺5 0 0
f 共t兲
0 0 0 1 0 0 0 0 0 0
0
13.33 0 ⫺26.67 ⫺12.67 0 12.67 0 0 0 ⫺6.67 0
ẋ⫽Ax⫹B̄u⫹F⫽ x⫹ u⫹
0 0 0 0 0 1 0 0 0 0 0
0 0 0 12.67 ⫺40 ⫺12.67 13.33 0 0 6.67 0
0
0 0 0 0 0 0 0 1 0 0 0 8⫻1
0 11 0 0 10 0 ⫺20 ⫺11 8⫻8
5 0 8⫻2
where x⫽ 关 x 11 ,ẋ 11 ,x 12 ,ẋ 12 ,x 21 ,ẋ 21 ,x 22 ,ẋ 22兴 T is the state vector

with x i j representing the motion of m i j , A, B̄ and F are self
evident terms, f ⫽ f 0 sin(␻ t), and the full state feedback control is
K⫽ 冉 14.20 2.48 ⫺8.24 ⫺2.69 ⫺11.41 ⫺3.12 13.05 4.55
8.25 2.52 ⫺10.37 ⫺4.45 ⫺10.17 ⫺2.65 5.33 2.19
冊 2⫻8
u⫽Kx. Ultimately the solution to this equation in Laplace domain such that the controlled systems characteristic roots, i.e.
is
det共 sI⫺A⫺B兲 ⫽0
x共 s 兲 ⫽ 共 sI⫺A⫺B兲 ⫺1 F共 s 兲 (10)
are ⫺2, ⫺4, ⫺5, ⫺6, ⫺1⫿3i, ⫺3⫿i. This is achieved using
where B⫽B̄K, F(s)⫽L关 F(t) 兴 , F(s)⫽L关 f (t) 兴 MATLAB/Control Toolbox. When a delay appears in the feedback
The characteristic roots of the uncontrolled structure is calcu- line; i.e. u(t)⫽Kx(t⫺ ␶ ), the characteristic equation of the sys-
lated from tem becomes
det共 sI⫺A兲 ⫽0 det共 sI⫺A⫺Be ⫺ ␶ s 兲 ⫽0 (11)

which is identical to that given in Eq. 共6兲. As we expand Eq. 共11兲
and they are ⫺24.02, ⫺20.69, ⫺0.72, ⫺1.79, ⫺0.018⫿6.44i, it takes the form of:
⫺0.038⫿4.10i.
We select a 0 共 s 兲 ⫹a 1 共 s 兲 e ⫺ ␶ s ⫹a 2 共 s 兲 e ⫺2 ␶ s ⫽0 (12)

Fig. 2 Number of unstable roots „ NU … plot in ␶ domain Fig. 3 Stability posture of the system
where a p (s) is a polynomial in s of degree 8⫺p, p⫽0, 1, 2.

Notice that the largest commensurate delay is 2␶ in Eq. 共12兲 be- That is, only 4 sets of ␶ kᐉ ’s 共each with dimension of infinity兲
cause of the specific formation of A and B matrices in this case. cause the roots pass through the pinholes of ␻ c1 , ␻ c2 , ␻ c3 , ␻ c4
The stability analysis follows as described in Section 2. in the direction given by root tendencies 共for ⫹1 from stable to
All imaginary roots are calculated using the methodology pre- unstable complex plane, and for ⫺1 vice versa兲.
sented in 关22兴 and they are tabulated along with the corresponding Continuing the listings of Eq. 共13兲 one can simply form the
delays and root tendencies 共Proposition II兲 below. Table 1, which we call the ‘‘stability table’’, in ascending order of
␶ kᐉ , k⫽1, . . . m, ⫽0, . . . ⬁.
冦冧
2␲p The corresponding NU( ␶ ) variation is like in Fig. 2. The sta-
␶ 1k ⫽0.544⫹
再冎
␻ c1 bility posture of the system is clear: it’s stable only for the time
delay 0⬍␶⬍307 ms. Afterwards it never returns to stability. What
␻ c1 ⫽4.3864 2␲p
␶ 2k ⫽0.307⫹ this implies in practice is that the delay on the feedback can only
␻ c2 ⫽7.5032 ␻ c2 be within this range for stability. Beyond 307 ms, the given feed-
→ from 共8兲
␻ c3 ⫽3.0446 2␲p back structure renders instability by enforcing unstable right-
␻ c4 ⫽2.98 ␶ 3k ⫽2.003⫹ hand-side roots numbered by NU( ␶ ), as per Fig 2. A numerical
␻ c3
simulation study validates this finding as given in Fig. 3.
2␲p
␶ 4k ⫽2.005⫹ The end result of this effort is not the declaration of stable or
␻ c4
再冎
unstable regions only, but also what happens to vibration suppres-
sion in the stable regions when the delay exists. We take ␶⫽250
RT 兩 1 ⫽⫹1
ms 共knowing that it is harmless from stability perspective兲 and
RT 兩 2 ⫽⫹1 study the frequency response of the mass m 12 under f 0 sin(␻t)
→ p⫽0, . . . ⬁ (13)
RT 兩 3 ⫽⫺1 harmonic excitation on m 11 . When Eq. 共10兲 is solved for x 12 it
RT 兩 4 ⫽⫺1 results in:
x 12共 s 兲 ⫺16.78 s5 ⫺186.52 s4 ⫺887.89 s3 ⫺2930.75 s2 ⫺6516.5 s⫺6155.36

⫽ 8 (14)
F共 s 兲 s ⫹25 s7 ⫹272 s6 ⫹1724 s5 ⫹7172 s4 ⫹20516 s3 ⫹39520 s2 ⫹46000 s⫹24000
Fig. 4 Frequency responses of the dynamics

for non-delayed feedback u⫽Kx(t) and 关5兴 Olgac, N., and Hosek, M., 1998, ‘‘A New Perspective and Analysis for Re-
generative Machine Tool Chatter,’’ Int. J. Mach. Tools Manuf., 38共7兲, pp. 783–
x 12共 s 兲 p 0 共 s 兲 ⫹p 1 共 s 兲 e ⫺ ␶ s ⫹p 2 共 s 兲 e ⫺2 ␶ s 798.
⫽ (15) 关6兴 Tlusty, J, 1985, ‘‘Machine Dynamics,’’ R. I. King, ed., Handbook of High
F共 s 兲 q 0 共 s 兲 ⫹q 1 共 s 兲 e ⫺ ␶ s ⫹q 2 共 s 兲 e ⫺2 ␶ s Speed Machining Technology, Chapman and Hall, New York.
关7兴 Olgac, N., and Holm-Hansen, B., 1994, ‘‘A Novel Active Vibration Absorption
for delayed feedback u⫽Kx(t⫺ ␶ ), where p i (s), q i (s) are self- Technique: Delayed Resonator,’’ J. Sound Vib., 176, pp. 93–104.
evident polynomials of s. 关8兴 Olgac, N., Elmali, H., Hosek, M., and Renzulli, M., 1997, ‘‘Active Vibration
The frequency response features of this system are given in Fig. Control of Disturbed Systems Using Delayed Resonator with Acceleration
Feedback,’’ ASME J. Dyn. Syst., Meas., Control, 119, pp. 380–388.
4 for various cases. It is clear that the feedback controlled case Eq. 关9兴 Chen, J., Gu, G., and Nett, C. N., 1994, ‘‘A New Method for Computing Delay
共14兲 registers substantial improvement over the uncontrolled case Margins for Stability of Linear Delay Systems,’’ Syst. Control Lett., 26, pp.
for ␻⬎2.7 rad/s. And the delayed case Eq. 共15兲 is bringing further 107–117.
improvement except in the frequency range of 7.1⬍␻⬍8.1 rad/s. 关10兴 Chen, J., 1995, ‘‘On Computing the Maximal Delay Intervals for Stability of
Linear Delay Systems,’’ IEEE Trans. Autom. Control, 40„6…, pp. 1087–1092.
In any event, it offers much better vibration suppression than the
关11兴 Hale, J. K., and Verduyn Lunel, S. M., 2001, ‘‘Effects of Small Delays on
uncontrolled 共passive兲 case. Stability and Control,’’ Operator Theory; Advances and Applications, 122, pp.
It is obvious that, Fig. 4 has meaning if and only if the system 275–301.
is stable, which can only be assured using the Direct Method 关12兴 Hale, J. K., and Verduyn Lunel, S. M., 2001, ‘‘Strong Stabilization of Neutral
presented. Therefore the novel method plays a very critical role in Functional Differential Equations,’’ IMA J. Math. Control Inform., 19, pp.
1–19.
this operation. 关13兴 Hale, J. K., and Verduyn Lunel, S. M., 1993, Introduction to Functional Dif-
ferential Equations, Springer-Verlag.
4 Conclusions 关14兴 Hale, J. K., Infante, E. F., and Tsen, F.-S. P., 1985, ‘‘Stability in Linear Delay
Equations,’’ J. Math. Anal. Appl., 105, pp. 533–555.
A general active vibration suppression problem under delayed 关15兴 Hertz, D., Jury, E. I., and Zeheb, E., 1984, ‘‘Simplified Analytic Stability Test
feedback is investigated in this paper. It is well known that for for Systems with Commensurate Time Delays,’’ IEE Proc., 131共1兲, Pt共D兲, pp.
such control systems, time delay plays a very important role on 52–56.
the vibration suppression quality and more importantly on the 关16兴 Jalili, N., and Olgac, N., 1999, ‘‘Multiple Delayed Resonator Vibration Ab-
sorber for MDOF Mechanical Structures,’’ J. Sound Vib., 223共4兲, pp. 567–585.
stability of the system. In this phase, we need to fully understand
关17兴 Kolmanovski, V. B., and Nosov, V. R., 1986, Stability of Functional Differen-
the stability picture of the dynamics under time delay influence, tial Equations, Academic Press, London, Great Britain.
first. For this we deploy a unique procedure, the Direct Method. 关18兴 Niculescu, S-I., 2001, Delay Effects on Stability, Springer-Verlag.
The method offers certain structured steps for the stability analysis 关19兴 Zhang, J., Knospe, C. R., and Tsiotras, P., 2001, ‘‘Stability of Time-delay
and uncovers very interesting features of linear time invariant Systems: Equivalence Between Lyapunov and Scaled Small-gain Conditions,’’
IEEE Trans. Autom. Control, 46共3兲, pp. 482– 486.
time delayed systems 共LTI-TDS兲, which were not recognized ear- 关20兴 Gu, K., and Niculescu, S.-I., 2000, ‘‘Additional Dynamics in Transformed
lier. Using the DM, one can determine the exact boundaries of Time-Delay Systems,’’ IEEE Trans. Autom. Control, 45共3兲, pp. 572–575.
stable and unstable regions in ␶ 共time delay兲 domain. It is then 关21兴 Park, P., 1999, ‘‘A Delay-Dependent Stability Criterion for Systems with Un-
possible to investigate the efficiency of a vibration suppression certain Time-Invariant Delays,’’ IEEE Trans. Autom. Control, 44, pp. 876 –
application, maintaining the stability of the system under time 877.
关22兴 Olgac, N., and Sipahi, R., 2002, ‘‘An Exact Method for the Stability Analysis
delay effects. And we include a comparison of controlled vibra- of Time Delayed LTI Systems,’’ IEEE Trans. Autom. Control, 47共5兲, pp. 793–
tion suppression for various frequency zones between the time 797.
delayed and non-delayed cases. 关23兴 Cooke, K. L., and van den Driessche, P., 1986, ‘‘On Zeroes of Some Transcen-
dental Equations,’’ Funkcialaj Ekvacioj, 29, pp. 77–90.
关24兴 Thowsen, A., 1981, ‘‘The Routh-Hurwitz Method for Stability Determination
References of Linear Differential-Difference Systems,’’ Int. J. Control, 33共5兲, pp. 991–
关1兴 Olgac, N., 1995, ‘‘Delayed Resonators as Active Dynamic Absorbers,’’ United 995.
States Patent 5,431,261. 关25兴 Thowsen, A., 1981, ‘‘An Analytic Stability Test for a Class of Time-Delay
关2兴 Olgac, N., Special Lecture at the 3rd IFAC Workshop on TDS 2001, http:// Systems,’’ IEEE Trans. Autom. Control, 26共3兲, pp. 735–736.
www.siue.edu/ifacdelay/ 关26兴 Thowsen, A., 1982, ‘‘Delay-independent Asymptotic Stability of Linear Sys-
关3兴 Seto, K., and Furuishi, Y., 1991, ‘‘A Study on Active Dynamic Absorber,’’ tems,’’ IEE Proc., 29, pp. 73–75.
ASME Proceedings, Paper DE, Vol. 38, pp. 263–270. 关27兴 Seto, K., 1995, ‘‘Structural Modeling and Vibration Control.’’ Internal Report,
关4兴 Seto, K., and Yamashita, S., 1991, ‘‘Simultaneous Optimum Design Method Nihon University. Department of Mechanical Engineering, College of Science
for Multiple Dynamic Absorbers to Control Multiple Resonance Peaks,’’ SAE and Technology, Nihon University, 1-8-14 Kanda Surugadai Chiyoda-ku, To-
Transactions, 100, pp. 1481–1489. kyo, 101-8308, Japan.

Bayesian Model Screening for the
Gaëtan Kerschen
e-mail: [email protected] Identification of Nonlinear
Jean-Claude Golinval Mechanical Structures
Vibrations & Identification des Structures,
Department of Aerospace, The development of techniques for identification and updating of nonlinear mechanical
Mechanics and Materials, structures has received increasing attention in recent years. In practical situations, there
University of Liège, is not necessarily a priori knowledge about the nonlinearity. This suggests the need for
Chemin des Chevreuils 1 (B52), strategies that allow inference of useful information from the data. The present study
B-4000 Liege, Belgium proposes an algorithm based on a Bayesian inference approach for giving insight into the
form of the nonlinearity. A family of parametric models is defined to represent the non-
linear response of a system and the selection algorithm estimates the likelihood that each
François M. Hemez member of the family is appropriate. The (unknown) probability density function of the
Engineering Science & Applications Division, family of models is explored using a simple variant of the Markov Chain Monte Carlo
ESA-WR, Mail Stop P946, sampling technique. This technique offers the advantage that the nature of the underlying
Los Alamos National Laboratory, statistical distribution need not be assumed a priori. Enough samples are drawn to guar-
Los Alamos, New Mexico 87545 antee that the empirical distribution approximates the true but unknown distribution to
e-mail: [email protected] the desired level of accuracy. It provides an indication of which models are the most
appropriate to represent the nonlinearity and their respective goodness-of-fit to the data.
The methodology is illustrated using two examples, one of which comes from experimental
data. 关DOI: 10.1115/1.1569947兴
1 Introduction growth in interest in a particular class of identification techniques

based on a finite element model and referred to as finite element
The importance of diagnosing, identifying and modelling non-
model updating techniques 关20–22兴.
linearity has been recognized for a long time, e.g., for the design
The problem of variable selection is one of the common issues
of shock absorbers and engine mounts. The identification of non-
in the field of identification of nonlinear systems. The purpose is
linear systems began in 1979 with the introduction of the restoring
to model the relationship between the response variable of interest
force surface 共RFS兲 method by Masri and Caughey 关1兴. An
and a subset of predictor variables, possibly with interactions be-
equivalent method, referred to as force-state mapping, was pro-
tween these latter variables. Generally speaking, there is uncer-
posed independently by Crawley, Aubert and O’Donnel 关2,3兴.
tainty about which subset to use.
Since then, numerous methods were proposed. It is not our inten-
A possible means of determining which variables should be
tion to review all the methods available but rather to cite the most
included in the model is through least-squares parameter estima-
popular techniques that have been considered during the last
tion and the use of the significance factor 关12兴. Cumulative and
twenty years.
multiple coherence functions may also be used in conjunction
The first application of the Hilbert transform was made in the
with the conditioned reverse path formulation 关23兴. The present
frequency domain 关4兴. The time-domain Hilbert transform was
study investigates an inference technique based on the Bayesian
also utilized to solve an inverse problem 关5,6兴. The use of the
definition of probability—as opposed to the frequentist’s point-of-
Volterra series in the field of structural dynamics began in the late
view—for identifying promising subsets of predictors 关24 –27兴.
1980s 关7兴. NARMAX models consist of polynomials that include
While the frequentist interpretation defines probability strictly as
various linear and nonlinear terms combining the inputs, outputs
the number of occurrences of an event among a collective of
and past errors and were introduced by Leontaritis and Billings
possibilities, the Bayesian approach defines probability as the sub-
关8,9兴. Another area of signal processing that has gained impor-
jective opinion of the analyst or expert. To stress the difference
tance in studying nonlinear systems deals with higher-order spec-
between the two approaches, consider the simple question ‘‘What
tra 关10,11兴. These are a natural extension of the ordinary linear
is the probability of life on the planet Mars?’’ Such question
spectral analysis. For a detailed description of all these tech-
makes no sense in the frequentist framework because observations
niques, the reader is referred to reference 关12兴.
can obviously not be obtained from a collective of planets similar
The development of frequency response function-based ap-
to the planet Mars. Similarly and even though we might not al-
proaches has received increasing attention in recent years. The
ways be aware of it, many problems occur in structural dynamics
reverse path technique has been proposed by Rice and Fitzpatrick
that require probability to be defined in terms of our a priori
关13兴 and applied to simulated and experimental data 关14,15兴. The
knowledge of the phenomenon studied. The identification of the
conditioned reverse path formulation 关16兴 extends the application
form of a model is one such problem addressed in this work.
of the reverse path algorithm to systems characterized by nonlin-
The procedure developed in this work exploits ‘‘priors’’—that
earities away from the location of the applied force. This method
is, a probability structure that reflects the analyst’s a priori opinion
exploits the spectral conditioning techniques introduced by Ben-
about the phenomenon investigated—on the variables of the re-
dat 关17兴. A related series of papers by Adams and Allemang also
gression model in order to give the list of all visited models to-
develop the frequency response function-based approaches
关18,19兴. gether with their relative posterior probabilities. Models are vis-
Finally, it is worth pointing out that there has also been a ited according to their goodness-of-fit to the data, which, in the
Bayesian framework, represents the likelihood of predicting the
observed response. This implies that models well fitted to the
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002; data—that is, more likely models—are visited more often. The
Revised January 2003. Associate Editor: M. I. Friswell. marginal probabilities of inclusion of single variables are also
computed. To avoid the overwhelming burden of calculating the error and minimize it. The most straightforward choice is to adopt
posterior probabilities of all models, a Gibbs sampler is consid- the Euclidean norm of the prediction error e k ⫽y k ⫺x Tk ␤ :
ered to perform an efficient stochastic search of the model space.
It is emphasized that the main difficulty of this inference problem
is that the multi-dimensional probability density function 共PDF兲 of
J共 ␤ 兲⫽ 兺
k⫽1 . . . N
e Tk e k ⫽e T e (5)
the family of models must be sampled. However, this PDF is
unknown, making random walk techniques such as the Markov in which case the best, linear, unbiased estimator of the coeffi-
Chain Monte Carlo 共MCMC兲 sampling the only possible choice. cients ␤ is provided by:
Gibbs sampling has been proposed as a computationally attractive
␤ˆ ⫽ 共 X T X 兲 ⫺1 X T y (6)
alternative to MCMC, yet, it can explore an unknown PDF
关24,26兴. where the column-vector y collects N observations and the N rows
While the principle of Bayesian inference has previously been by m columns matrix X evaluates the m effects for each of the N
冋册
applied to various problems in structural dynamics 共e.g., Refer- observations:
再冎
ences 关28,29兴兲, no attempt has been made, to the best of the au-
thors’ knowledge, to adapt the Bayes updating rule to the screen- x 1,1 x 1,2 ¯ x 1,m
y1
ing of model form during nonlinear system identification. After a y2 x 2,1 x 2,2 ¯ x 2,m
brief discussion of model fitting in Section 2, the Bayesian screen- y⫽ ;X⫽ (7)
] ] ] ]
ing algorithm for model selection is outlined in Section 3. The
yN
methodology is illustrated using two examples. Section 4 dis- x N,1 x N,2 ¯ x N,m
cusses a numerical simulation intended at demonstrating the over-
Clearly, other objective functions yield different estimators. The
all performance of the screening method. The second example
generalization of the objective function 共5兲 is commonly referred
involves experimental data sets collected during the European
to as the generalized least-squares 共GLS兲 estimator 关31兴. Weight-
COST-F3 program 共Section 5兲. The numerical predictability of the
ing matrices are introduced and a regularization term penalizes
identified model is finally assessed in Section 6.
solutions too distant from the user-defined starting point ␤ o . Eqs.
共8兲 and 共9兲 show the GLS objective and the corresponding GLS
2 Model Fitting estimator, respectively:
Model fitting generally refers to the calibration of model coef- J 共 ␤ 兲 ⫽e T W ⫺1 T ⫺1
ee e⫹ 共 ␤ ⫺ ␤ o 兲 W bb 共 ␤ ⫺ ␤ o 兲 (8)
ficients ␤ given a sequence of points (t k ;y k ) in the design space.
It is assumed that a model is available: ␤ˆ ⫽ 共 X T W ⫺1 ⫺1 ⫺1 T ⫺1
ee X⫹W bb 兲 X W ee y (9)
y⫽M 共 ␤ ;t 兲 (1) In general, weighting matrices are chosen arbitrarily or based
on experience, for example, to weight the importance of some
where t denotes the input variables, y denotes the output variables observations more than others. When covariance matrices are
and ␤ denotes the model’s coefficients. used, the GLS estimator becomes similar to the Bayesian estima-
For clarity, the discussion will assume that the model form is tor. Rigorously speaking, other factors should appear in the defi-
polynomial-like. Nevertheless, nothing prevents the Bayesian nition of the Bayesian objective function. Because these addi-
model screening proposed in Section 3 to be applied to other tional factors are constant, however, the same estimator as the one
functional forms. Fractional models could be considered, for ex- shown in Eq. 共9兲 is obtained. An important benefit of Bayesian
ample, to fit the poles and zeros of frequency response functions. inference is that it provides a posterior estimate of the covariance
Exponential models could be considered to represent the decay of matrix:
propagating waves as a function of time or distance. Artificial
neural networks are increasingly used in a variety of applications Ŵ 共bbposterior兲 ⫽ 共 W ⫺1 T ⫺1
bb ⫹X W ee X 兲
⫺1
(10)
in structural dynamics because they can, depending on their form,
approximate any non-linear function 关30兴. Correlation coefficients of the posterior covariance matrix 共10兲
Another notion that must be clarified before proceeding with provide insight into the quality of the estimator. Reference 关32兴
the discussion is the notion of ‘‘effect.’’ The model shown in Eq. discusses a shock propagation application where significant pos-
共1兲 depends on inputs t where t does not necessarily refer to time. terior correlation is obtained between coefficients that have no
Functions of the input variables t can be defined that will be physical reason to be correlated. The authors conclude that the
referred to as effects and denoted by x in the following. Such form of the model is inappropriate. They further demonstrate that
functions can assume any form, linear or non-linear. For example, it is indeed the case when improved goodness-of-fit and posterior
the 2-input, 1-output nonlinear model: correlation indicators are obtained with a different model.
With the exception of investigating the posterior correlation,
y⫽0.3t 1 ⫹2.0 sin共 t 2 兲 ⫺1.5e ⫺t 1 t 2 (2) however, no practical tool is available to select the appropriate
can equivalently be defined through the three effects x 1 ⫽t 1 , x 2 form of a nonlinear model, which is the process we refer to a
⫽sin(t2) and x 3 ⫽e ⫺t 1 t 2 as: model screening. Model form—for example, replacing a linear
contribution by a cubic stiffness—is usually selected based on
y⫽0.3x 1 ⫹2.0x 2 ⫺1.5x 3 (3) experience or empirical observation. Sometimes, several choices
seem equally likely and the analyst has to go through the pains-
While the input variables t 1 and t 2 might be independent, note
taking process of fitting each model and assessing their goodness-
that the effects x 1 , x 2 and x 3 are neither independent nor uncor-
of-fit. Because it is based on the concept of goodness-of-fit, such
related. The Bayesian model screening discussed in Section 3
approach leads to over-fitting.
does not require the effects to be independent or uncorrelated.
Another subtle but important issue is to estimate the posterior
With the definition of effects x that can be functions of the input
probability of a particular model as opposed to simply relying on
variables t, the polynomial-like model can be simply represented
the goodness-of-fit. By definition, the posterior probability is con-
as:
ditioned on the evidence available—that is, experimental observa-
tions. Posterior probability and goodness-of-fit complement each
y⫽ 兺
k⫽1 . . . m
x k ␤ k ⫽x T ␤ (4) other because the former indicates if the analyst’s prior opinion of
the form of the model is consistent with the evidence. In Section
The commonly encountered method of fitting the coefficients ␤ 3, a practical tool is proposed for model screening based on the
is to define an objective function that represents the prediction concept of posterior probability.

3 Bayesian Model Screening
In the previous section, the state of the practice in model fitting
has been briefly overviewed. Polynomial models have been con-
sidered for simplicity. It is emphasized that the Bayesian model
screening technique proposed here applies to any model, no mat-
ter which functional form it takes. Essentially, the only two gen-
eral assumptions made are as follows. First, a model y
⫽M ( ␤ ;x) must be available. As previously mentioned, the effects
x can be linear or nonlinear functions of the input variables t.
Second, an inference is available for calibrating the coefficients ␤.
The inference procedure is usually referred to as ‘‘best-fitting’’
with polynomials and ‘‘training’’ with neural networks.
Model screening consists in identifying the most probable mod-
els based on a family of models defined by the user and reference
data that the model’s predictions must reproduce with the highest
possible fidelity. It is emphasized that model screening does not
necessarily identify the best model but rather ranks potential mod-
els according to their posterior probability of occurrence.
The procedure starts by, first, defining a family of models. This
is achieved by defining various effects x i and how these effects Fig. 1 Concept of ‘‘family’’ of models
are allowed to interact to form the population of potential models.
Figure 1 illustrates the concept of a family of models by showing
two effects x 1 and x 2 that interact with each other. The model- posteriors of the nth iteration become the priors of the (n⫹1)th
forming rule illustrated in Fig. 1 is that linear and quadratic inter- iteration. All models visited are kept in memory and, once enough
actions are allowed between the effects x 1 and x 2 . The horizontal samples have been drawn, the probability of occurrence of each
plane represents the family of all potential models that must be model is estimated by the frequency of occurrence—that is, the
explored. The vertical dimension represents the likelihood that a ratio between the number of times each model is visited and the
particular model is appropriate to represent the data. It is this total number of models visited. The iterative procedure is summa-
notion of likelihood that will be employed to guide the search for rized in Fig. 2.
the most appropriate models. Figure 1 illustrates a hypothetical In summary, Bayesian model screening provides the probabili-
situation where the model shown with a star symbol, y⫽ ␤ o ties of occurrence of the most appropriate members of a user-
⫹ ␤ 1 x 2 ⫹ ␤ 2 x 21 ⫹ ␤ 3 x 2 x 21 , is the maximum likelihood model.
The second step of the procedure is to assign the prior prob-
ability of occurrence of each effect x i . The priors can reflect
empirical observations, experience or the analyst’s knowledge of
the system investigated. In the application discussed in Section 5,
for example, no specific knowledge of the system can be used to
guide a pertinent choice of priors. Probabilities of occurrence are
therefore set to a uniform 25% level for all effects.
The next step is to let the Bayesian screening method find the
most appropriate models among all possible combinations of ef-
fects. To identify the most probable models a measure of
goodness-of-fit to the reference data must be defined. This can be
assessed using a conventional root mean square 共RMS兲 error be-
tween data and predictions. Assuming Gaussian distributions, the
RMS error becomes proportional to the likelihood function
L(y 兩 ␤ ) that estimates the likelihood that the model is appropriate
given the available data:
L共 y 兩␤ 兲⫽ 兺
k⫽1 . . . N
共 y k ⫺x Tk ␤ 兲 2 (11)
Note that the likelihood function 共11兲 is similar to Eq. 共5兲 previ-
ously discussed. Other functions can be used, in particular the
Bayesian objective 共8兲, as well as the many objective functions
commonly used in test-analysis correlation and model updating
关33兴.
Once the likelihood of a particular model has been estimated,
the posterior probabilities of the model’s effects can be updated
according to the Bayes Theorem that states that the posterior prob-
ability PDF( ␤ 兩 y) is equal to the likelihood function L(y 兩 ␤ ) mul-
tiplied by the prior probability PDF( ␤ ) and divided by the prob-
ability of the data PDF(y):
L 共 y 兩 ␤ 兲 PDF共 ␤ 兲
PDF共 ␤ 兩 y 兲 ⫽ (12)
PDF共 y 兲
The probability of the observed data PDF(y) is generally kept
constant and omitted in the updating Eq. 共12兲. Because the proce- Fig. 2 Simplified flow chart of the Bayesian model screening
dure is iterative in nature, the Bayes update 共12兲 is repeated and algorithm

The main difference between the two is that the Gibbs algorithm
samples one direction of the design space at a time, which makes
for simpler numerical implementation. Figure 4 illustrates the dif-
ference between MCMC and Gibbs sampling. It pictures two ran-
dom walks from the lower left corner (x⫽0;y⫽0) to the upper
right corner (x⫽1;y⫽1). A constraint is enforced that prevents
the 30 points drawn in both sequences from being repeated and
from moving backwards. Pentagram symbols show a sequence of
Gibbs samples while hexagram symbols picture a realization of
the MCMC chain. In the former case, the solution is advanced in
one direction at a time whereas the MCMC chain randomly ad-
vances the solution in the two dimensions simultaneously.
4 Numerical Application
Fig. 3 Concept of random walk optimization
The first application presented is extremely simple and aims at
illustrating the overall performance of the model screening proce-
dure. Consider an output variable y defined by the following
input-output model:
y⫽2 sin共 2t 兲 ⫹3 cos共 t 兲 ⫺1.5 sin共 3t 兲 cos共 2t 兲 (13)
where t is an input variable that varies from zero to fifty with
increments of ⌬t⫽0.05. It is assumed that the model form shown
in Eq. 共13兲 is unknown. Instead, observations y k ⫽y(k⌬t), for k
⫽0 . . . 100, are obtained and the problem consists in identifying
the numerical model that best matches the observed data. It is
emphasized that, in this numerical simulation, no actual experi-
ment is performed. The continuous solution 共13兲 is shown in Fig.
5 with a solid line. The hexagram symbols represent the discrete
samples assumed to be collected.
Next, consider a set of candidate predictors:
x 1 ⫽sin共 t 兲
冦
x 2 ⫽cos共 t 兲
x 3 ⫽sin共 2t 兲
(14)
x 4 ⫽cos共 2t 兲
x 5 ⫽sin共 3t 兲
x 6 ⫽cos共 3t 兲
Fig. 4 Illustration of MCMC and Gibbs random walk sampling
In addition to the six predictors of Eq. 共14兲, six other predictors
strategies
labeled x 7 , x 8 , x 9 , x 10 , x 11 and x 12 are defined as random func-
tions. It can be observed that, if the functional form of the output
variable y were known, it could be written as:
defined family of models, their goodness-of-fit indicators and the
posterior probabilities PDF( ␤ 兩 y) of effects involved in the most y⫽3x 2 ⫹2x 3 ⫺1.5x 4 x 5 (15)
likely models.
To do so, however, the unknown posterior probability function
must be sampled. The problem of exploring an unknown PDF is
solved with the Markov Chain Monte Carlo algorithm. The
MCMC sampling is advantageous in this situation because it can
sample any distribution, whether it is Gaussian or not. The
MCMC sampling can be viewed conceptually as an optimization
solver that performs a random walk through the optimization
space. This concept is illustrated in Fig. 3 where points in the
optimization space are sequentially visited. More appropriate so-
lutions are guaranteed more frequent visits because the acceptance
criterion of a given solution is based on its likelihood function.
Each candidate point in the design space—here, the design
space is the horizontal plane of potential models illustrated in
Figs. 1 and 3—is accepted or rejected based on its value of the
likelihood function 共11兲 and a Chi-square test. This particular ac-
ceptance criterion implies that inappropriate models have a small
chance of being accepted just like appropriate models have a
small chance of rejection. If rejected, a new point is randomly
selected in the neighborhood of the last accepted point. The se-
quence of points accepted is stored to estimate, once the process
has been completed, the probability of occurrence of each model.
The sampling procedure used in this work is the Gibbs sam-
pling, the simplest of the many variants of the MCMC algorithm. Fig. 5 Simulated non-linear function „13…

Table 1 Top five models and number of appearances These uniform probabilities reflect the fact that little is known
about the form of the model before starting the analysis. It can be
Model Effects Posterior Probability
observed that effects 2, 3 and 43, namely x 2 , x 3 and x 4 x 5 , are
1 x 2 , x 3 , x 4x 5 52.0% associated with a probability of 100% while the other effects may
2 x 2 , x 3 , x 4 x 5 , x 10 3.0% be ignored because their posterior probabilities are reduced to
3 x 2 , x 3 , x 4x 5 , x 3x 4 2.0% insignificant levels.
4 x 2 , x 3 , x 4 x 5 , x 2 x 10 2.0%
5 x 2 , x 3 , x 4x 5 , x 2x 5 2.0% In conclusion, the Bayesian model screening clearly suggests a
model that includes the three effects x 2 , x 3 and x 4 x 5 . The iden-
tified coefficients corresponding to these effects are equal to 2.99,
2.02 and ⫺1.52, respectively, and they are in good agreement with
Clearly, y does not depend on predictors x 1 , x 6 , x 7 , x 8 , x 9 , x 10 , the actual coefficients shown in Eq. 共15兲. The algorithm is imple-
x 11 and x 12 . The objective of model screening is to identify the mented as interpreted Matlab™ functions and it performs the
model form 共15兲. Equivalently, it can be stated that the objective analysis in a few seconds of CPU time with a typical desktop
of model screening is to identify the linear effects x 2 , x 3 and the personal computer.
linear interaction effect x 4 x 5 from all the potential combinations
defined by the family of models considered. 5 Experimental Application
The family of models defined for this illustration is composed In this Section, Bayesian model screening is applied to the
of the linear models that include the twelve linear effects x i and problem of identifying the form of a nonlinear model using real,
the linear interaction models, defined as the previous models aug- experimental data. The analyzed data sets are chosen from those
mented with the 66 interaction effects x i x j . The total number of proposed by the VTT Technical Research Center of Finland within
different effects x i and x i x j with twelve predictors is therefore the framework of the European COST action F3 working group
equal to 78. The total number of different models that can be on ‘‘Identification of Nonlinear Systems’’ 关34兴.
defined belonging to this family by combining the 78 effects is in The structure investigated consists of wire rope isolators
excess of 3.02 10⫹23 models, a number that approaches the num- mounted between the load mass and the base mass, as shown in
ber of atoms in the known Universe. Clearly, exploring such a Fig. 7. The load mass acts like a free inertial mass. The motion
large number of combinations without focusing on the models of and forces experienced by the isolators are measured. In particu-
highest likelihood would not be feasible. lar, the acceleration responses ẍ 2 and ẍ 1b of the load mass and
The procedure described in the foregoing section is applied to bottom plate, the applied force f and the relative displacement x 12
the data using 50 samples dedicated to the initialization of the between the top and bottom plates are measured. The excitation
Gibbs sampler and 100 samples for the computation. Initializing produced by an electro-dynamic shaker corresponds to a white
the Markov chain is referred to as ‘‘burn-in’’ and guarantees that noise sequence, low-pass filtered at 400 Hertz. What makes this
the remainder of the chain is not biased due to a particular choice system interesting for identification is that the attenuation of the
of starting point. The samples drawn during burn-in are disre- vibration across the interface is difficult to characterize because
garded and only the 100 samples drawn during the optimization the mechanics of the isolators is unknown to a large extent. Sig-
itself are kept to estimate the final probability of occurrence of nificant nonlinear dynamics are expected due to the geometrical
each model in the family. The top five models are listed in Table nonlinearity—pre-loading in the wire rope isolators changes with
1. It can be observed that the best model in terms of posterior the load mass.
model probability is the actual model. The mean-square error for
the top five models is about 0.003%. This means that it is not
necessary to include other terms than the ones present in the best
model.
Figure 6 represents the marginal posterior probability of each
effect being in a particular model. The prior probabilities—that
reflect the prior knowledge—are set to 25% for each linear effect
x i ; 10% for the interaction effects x i x j if one of the parent effect
x i or x j is selected in the model; and 1% only for the interaction
effects x i x j when neither x i nor x j are considered in the model.
Fig. 6 Marginal posterior probability of each effect included in

the family of models Fig. 7 Wire rope isolators

Table 2 Testing matrix of the VTT benchmark The next step of the RFS method is to describe the restoring
force by means of a mathematical model. This is achieved through
Forcing Level Mass 1 共2.2 kg兲 Mass 2 共5.8 kg兲
model fitting such as described in Section 2. The generic form of
Level 1 共0.5 volt兲 Test 1 — models sought is usually given by:
Level 2 共2.0 volt兲 Test 2 — m n
Level 3 共4.0 volt兲 Test 3 Test 5
Level 4 共8.0 volt兲 Test 4 — f 共 x,ẋ 兲 ⫽ 兺兺␣
i⫽0 j⫽0
i jx
i j
ẋ (18)
To resolve the problem of order determination, which refers to the

identification of the most appropriate dimensions m, n in Eq. 共18兲,
an over-determined system of linear equations is formed with the
Four excitation levels are considered ranging from 0.5 Volt up available restoring force data. The singular value decomposition is
to 8.0 Volt. A nominal series of four tests are performed with a then used to select the appropriate order. Reference 关35兴 details
load mass of 2.2 kg. A fifth test is also carried out with the heavier the identification procedure and shows that the final model in-
load mass of 5.8 kg. Table 2 defines the testing matrix from which cludes a linear stiffness term, a viscous damping term and a non-
data sets have been collected. linear stiffness contribution:
Reference 关35兴 discusses the identification of the VTT bench-
mark structure using the RFS. The main idea behind the RFS f NL 共 x 12 ,ẋ 12兲 ⫽k l x 12⫹c l ẋ 12⫹k nl 兩 x 12兩 ␣ sign共 x 12兲 (19)
method is briefly overviewed to explain the system identification where the coefficients k l , c l , k nl and ␣ identified with the RFS
approach and the reader is referred to Reference 关35兴 for more method and singular value decomposition are listed in Table 3.
details. These results are used in the remainder as the reference through
The derivation of the main equations of the RFS method starts which the performance of the Bayesian model screening is
by writing Newton’s second law for the load mass m 2 , which assessed.
yields: The final model features a mean square error 共MSE兲 equal to
m 2 ẍ 2 ⫹ f NL 共 x 2 ⫺x 1b ,ẋ 2 ⫺ẋ 1b 兲 ⫽0 (16) 2.11%, which indicates an excellent correlation to test data. The
MSE indicator is a normalized metric that measures the goodness-
where f NL denotes the nonlinear internal force. Clearly, the force of-fit between model predictions and physical observations. It is
f NL is unknown but it can be ascertained, as shown in Eq. 共16兲, defined as:
that its value depends on the displacement and velocity of the load
100
mass relative to those of the bottom plate. Introducing the relative
displacement, x 12⫽x 2 ⫺x 1b , Eq. 共16兲 becomes:
MSE⫽ 兺
N ␴ 2y k⫽1 . . . N
共 y k ⫺x Tk ␤ˆ 兲 2 (20)
f NL 共 x 12 ,ẋ 12兲 ⫽⫺m 2 ẍ 1b ⫺m 2 ẍ 12 (17) where, to comply with notations introduced in Section 2, y k rep-
resents the available restoring force data and ␴ y is the standard
Equation 共17兲 can be viewed as describing the response of a
deviation of data y k . The vector ␤ collects the coefficients k l , c l
SDOF system subjected to a base acceleration. Because the accel-
and k nl , assuming that the exponent ␣ is known and equal to 1.5,
eration signals shown in the right-hand side of Eq. 共17兲 are mea-
and the vector x k collects the corresponding effects in Eq. 共19兲.
sured and the mass is known, it is possible to compute the restor-
The exercise of identifying the most appropriate model form is
ing force f NL at each instant from Eq. 共17兲.
now repeated with the Bayesian model screening. First, three ef-
The value of the restoring force is shown in Fig. 8 in the four
fects are defined in agreement with Eq. 共19兲. They are the linear
cases where the load mass is equal to 2.2 kg 共Tests 1– 4兲. At low
stiffness x 12 , linear damping ẋ 12 and nonlinear stiffness
excitation level, the system’s behavior is predominantly linear be-
cause the restoring force varies linearly with the displacement, as 兩 x 12兩 ␣ sign(x 12). The corresponding coefficients are denoted by k l ,
can be observed for the 0.5 Volt and 2.0 Volt levels. As the exci- c l and k nl , as before. Second, model-forming rules are defined
tation level is increased, a softening stiffness nonlinearity appears, which are that main effects and linear interactions between the
as can be observed from the 4.0 Volt and 8.0 Volt levels. main effects are allowed. This means that a total of six effects
leading to sixty two different model forms are allowed. Such com-
binatorial complexity is trivial compared to the example discussed
in Section 4. The complexity here stems from the fact that real
data sets are analyzed with all the risk of erroneous identification
caused by ‘‘noisy’’ measurements and signal conditioning issues.
Because the exponent ␣ is unknown, the Bayesian model
screening is repeated for several assumed values of ␣. The value
that leads to the smallest MSE is retained. Repeating model
screening could become CPU-time intensive if long MCMC
chains are requested for each analysis. For this application, an
initial chain of length 50 is dedicated to burn-in and a chain of
length 300 is requested for the optimization. It has been verified
that requesting more samples does not improve the quality of the
final results. Figure 9 shows the evolution of the MSE as a func-
tion of the exponent ␣. The minimum value is obtained for ␣
Table 3 RFS identification of Eq. „19…
Coefficient Value Units

kl ⫹6 N/m
1.09 10
cl 183.44 N.sec/m
k nl ⫺8.52 10⫹7 N/m1.5
Fig. 8 Estimation of the restoring force at the four levels 0.5 ␣ 1.5 Unitless
Volt, 2.0 Volt, 4.0 Volt and 8.0 Volt

Table 5 Bayesian identification of Eq. „19….
Coefficient Value Difference共*兲

kl ⫹6 2.75%
1.12 10 N/m
cl 198.19 N.sec/m 8.04%
k nl ⫺9.07 10⫹7 N/m1.5 6.46%
␣ 1.5 0.00%
共兲
* Difference relative to coefficients in Table 3.
in Table 4 includes only the main effects x 12 , ẋ 12 and

兩 x 12兩 1.5sign(x 12) and appears 86% of the time in the Markov chain.
The main effects, labeled 1–3 in Figure 10, have a posterior
probability of 100% while interaction effects, labeled 4 – 6, may
be considered negligible because their posterior probability is be-
low 10%. For this application, the prior probabilities were set to a
uniform 20% for the main effects; 10% for an interaction effect
when at least one of the ‘‘parent’’ main effect is selected in the
Fig. 9 Evolution of the MSE as a function of the non-linearity
exponent ␣ model; and 1% only for an interaction effect when none of the
parents are selected. The increase in probability for effects 1–3 in
Figure 10 and the reduction for effects 4 – 6 are therefore signifi-
cant. From these results it can be concluded that a suitable model
⫽1.5, the same value as the one identified in Reference 关35兴 with for the restoring force is given by Eq. 共19兲 with an exponent equal
the RFS method and singular value decomposition. It may seem to ␣ ⫽1.5.
paradoxical that the MSE greatly increases for ␣ ⫽1, i.e., for a The coefficients k l , c l , k nl and ␣ identified with the Bayesian
linear model while its value remains low in the neighborhood of model screening are listed in Table 5. The last column in Table 5
␣ ⫽1 共e.g., ␣ ⫽0.99). The reason is that there is still a slight compares the identification results to those of the RFS method in
curvature for values of ␣ different from 1 that can be enhanced by Reference 关35兴. To calibrate the model’s coefficients, the Bayesian
taking high values of the corresponding nonlinear parameter k nl . model screening currently relies on the least-squares estimator 共6兲
Table 4 and Figure 10 display the top five models and the even though other solvers could be implemented. Although the
marginal posterior probability of each effect, respectively. The ‘‘true’’ solution is unknown, it can be stated that both methods
mean square error for each of the top five models is around provide consistent results because the maximum difference is less
2.37%, very similar to the RFS results previously reported. Such than 10%.
low MSE values indicate that the agreement with experimental The small differences witnessed between the RFS identification
data meets the expected level of accuracy. The most likely model and Bayesian model screening may be attributed to the different
data sets used. Referring to Table 2, the RFS identification was
conducted using the five combinations of input levels and load
Table 4 Top five models and number of appearances masses 共Tests 1–5兲. The Bayesian model screening is restricted to
four of the five cases, as discussed in Section 6, to provide a
Posterior validation of the model’s predictive accuracy.
Model Effects Probability
1 x 12 ,ẋ 12 , 兩 x 12兩 1.5 sign(x 12) 86.0% 6 Validation of the Identified Model
2 Model 1⫹x 12兩 x 12兩 1.5 sign(x 12) 7.0%
3 Model 1⫹ẋ 12兩 x 12兩 1.5 sign(x 12) 3.0% It was pointed out previously that only four of the five data sets
4 Model 1⫹x 12ẋ 12 2.6% are considered during the identification. These are Tests 1, 2, 4
5 Model 1⫹x 12兩 x 12兩 1.5 sign(x 12)⫹ 0.3% and 5. The remaining data set 共Test 3, 2.2 kg load mass, 4.0 Volt
ẋ 12兩 x 12兩 1.5 sign(x 12) level兲 is exploited to assess the predictive accuracy of the identi-
fied model.
Although it might not yet be the state of the practice in struc-
tural system identification, many authors, among whom we cite a
recent discussion of model validation in Reference 关36兴, have em-
phasized that identified models should be independently validated.
It essentially means that independent experiments or data sets
should be used for model screening and parametric calibration, on
one hand, and model validation and predictive accuracy assess-
ment, on the other hand. The predictive accuracy of a model can-
not be objectively assessed over the operational range of interest
as long as the independence between training data and validation
data is not met.
Here, data sets collected during Tests 1, 2, 4 and 5 are used for
model screening and system identification while the data collected
during Test 4 are used for model validation. Essentially, Eq. 共19兲
is evaluated with the coefficients of Table 5 to predict the restor-
ing force. Displacement and velocity time series in Eq. 共19兲 are
estimated from numerical integration of the measured acceleration
signals. The evolution of the predicted restoring force versus time
is then compared to the ‘‘true’’ restoring force measured during
Fig. 10 Marginal posterior probability of each effect included Test 3. The true restoring force is estimated directly from accel-
in the family of models eration measurements, as shown in Eq. 共17兲. In Fig. 11, the two

References
关1兴 Masri, S. F., and Caughey, T. K., 1979, ‘‘A Non-parametric Identification Tech-
nique for Non-linear Dynamic Problems,’’ ASME J. Appl. Mech., 46, pp.
433– 447.
关2兴 Crawley, E. F., and Aubert, A. C., 1986, ‘‘Identification of Nonlinear Structural
Elements by Force-state Mapping,’’ AIAA J., 24, pp. 155–162.
关3兴 Crawley, E. F., and O’Donnel, 1986, ‘‘Identification of Nonlinear System Pa-
rameters in Joints using the Force-state Mapping Technique,’’ AIAA Paper
86 –1013, pp. 659– 667.
关4兴 Simon, M., and Tomlinson, G. R., 1984, ‘‘Use of the Hilbert Transform in
Modal Analysis of Linear and Non-linear Structures,’’ J. Sound Vib., 96, pp.
421– 436.
关5兴 Feldman, M., 1994, ‘‘Nonlinear System Vibration Analysis using the Hilbert
Transform-I. Free Vibration Analysis Method ‘FREEVIB,’ ’’ Mech. Syst. Sig-
nal Process., 8, pp. 119–127.
关6兴 Feldman, M., 1994, ‘‘Nonlinear System Vibration Analysis using the Hilbert
Transform-II. Forced Vibration Analysis Method ‘FORCEVIB,’ ’’ Mech. Syst.
Signal Process., 8, pp. 309–318.
关7兴 Gifford, S. J., and Tomlinson, G. R., 1989, ‘‘Recent Advances in the Applica-
tion of Functional Series to Non-linear Structures,’’ J. Sound Vib., 135, pp.
289–317.
关8兴 Leontaritis, I. J., and Billings, S. A., 1985, ‘‘Input-Output Parametric Models
for Non-linear Systems: Part I—Deterministic Non-linear Systems,’’ Int. J.
Fig. 11 Comparison between measured and predicted restor- Control, 41, pp. 303–328.
ing forces „4 Volt level… 关9兴 Leontaritis, I. J., and Billings, S. A., 1985, ‘‘Input-Output Parametric Models
for Non-linear Systems: Part II—Stochastic Non-linear Systems,’’ Int. J. Con-
trol, 41, pp. 329–344.
关10兴 Bendat, J. S., and Piersol, A. G., 1980, Engineering Applications of Correla-
time series are compared. It can be observed that the measured tion and Spectral Analysis, John Wiley & Sons, New-York 共U.S.A.兲.
关11兴 Choi, D., Chang, J., Stearman, R. O., and Powers, E. J., 1984, ‘‘Bispectral
and predicted forces agree to the point where the difference be- Interaction of Nonlinear Mode Interactions,’’ Proceedings of the 2nd Interna-
tween signals is not visible. tional Modal Analysis Conference, Orlando, FL, pp. 602– 609.
The MSE, that assesses the overall fidelity between measure- 关12兴 Worden, K., and Tomlinson, G. R., 2001. Non-linearity in Structural Dynam-
ments and model predictions, is equal to 0.77% when the restoring ics: Detection, Identification and Modelling, Institute of Physics Publishing,
Bristol, Philadelphia 共PA兲.
force of Test 3 is predicted based on the identified Eq. 共19兲. Based 关13兴 Rice, H. J., and Fitzpatrick, J. A., 1991, ‘‘A Procedure for the Identification of
on the low MSE values obtained during model identification Linear and Non-linear Multi-degree-of-freedom Systems,’’ J. Sound Vib., 149,
共2.37%兲 and validation 共0.77%兲, it can be stated that the predic- pp. 397– 411.
tion accuracy seems to be within 97%. This means that a predic- 关14兴 Rice, H. J., and Fitzpatrick, J. A., 1991, ‘‘The Measurement of Non-linear
tion should be within 3% of a measurement, should a hypothetical Damping in Single-degree-of-freedom Systems,’’ ASME J. Vibr. Acoust., 113,
pp. 132–140.
experiment be conducted. Clearly, the main drawback of our as- 关15兴 Esmonde, H., Fitzpatrick, J. A., Rice, H. J., and Axisa, F., 1992, ‘‘Modelling
sessment of predictive accuracy is that it is based on a single and Identification of Non-linear Squeeze Film Dynamics,’’ J. Fluids Struct., 6,
analysis. Other validation data sets would be required to reach a pp. 223–248.
more quantitative statement of accuracy. Nevertheless, this analy- 关16兴 Richards, C. M., and Singh, R., 1998, ‘‘Identification of Multi-Degree-of-
Freedom Non-linear Systems Under Random Excitations by the Reverse-Path
sis leads to the conclusion that a reliable identification has been Spectral Method,’’ J. Sound Vib., 213, pp. 673–708.
performed over the operational range of interest, that is, within the 关17兴 Bendat, J. S., 1990, Nonlinear System Analysis and Identification from Ran-
ranges of 2.2-to-5.8 kg load mass and 0.5-to-8.0 Volt excitation dom Data, John Wiley & Sons, New-York 共U.S.A.兲.
level. 关18兴 Adams, D. E., and Allemang, R. J., 1999, ‘‘Characterization of Nonlinear
Vibrating Systems using Internal Feeback and Frequency Response Modula-
tion,’’ ASME J. Vibr. Acoust., 121, pp. 495–500.
关19兴 Adams, D. E., and Allemang, R. J., 2000, ‘‘A Frequency Domain Method for
7 Conclusion Estimating the Parameters of a Non-Linear Structural Dynamic Model through
Feedback,’’ Mech. Syst. Signal Process., 14, pp. 637– 656.
Model selection is one of the common issues in the field of 关20兴 Hemez, F. M., and Doebling, S. W., 2001, ‘‘Review and Assessment of Model
identification of nonlinear systems. A Bayesian inference ap- Updating for Non-linear, Transient Dynamics,’’ Mech. Syst. Signal Process.,
proach for giving insight into the form of the nonlinearity has 15, pp. 45–74.
关21兴 Lenaerts, V., Kerschen, G., and Golinval, J.-C., 2001, ‘‘Proper Orthogonal
been proposed in this paper. The key advantage of the method is Decomposition for Model Updating of Non-linear Mechanical Systems,’’
that a collection of potential models together with their posterior Mech. Syst. Signal Process., 15, pp. 31– 43.
probability is obtained instead of the single best model. It allows 关22兴 Dippery, K. D., and Smith, S. W., 1998, ‘‘An Optimal Control Approach to
for more flexibility in deciding the most appropriate model of the Nonlinear System Identification,’’ Proceedings of the 16th International Modal
Analysis Conference, Santa Barbara CA, pp. 637– 643.
non-linearity. In addition, the marginal posterior probability of 关23兴 Bendat, J. S., 1986, Random Data: Analysis and Measurement Procedures,
each effect being in a particular model may also be evaluated. John Wiley-Interscience, New York 共NY兲, second edition.
The results obtained using two illustrative examples enable us 关24兴 Chipman, H., Hamada, M., and Wu, C. F. J., 1997, ‘‘A Bayesian Variable
to confirm the performance and the utility of the proposed tech- Selection Approach for Analyzing Designed Experiments With Complex Alias-
ing,’’ Technometrics, 39, pp. 372–381.
nique. We believe that Bayesian model screening will become an 关25兴 Wu, C. F. J., and Hamada, M., 2000, Experiments: Planning, Analysis, and
important addition to the structural dynamicist’s toolbox. Parameter Design Optimization, Wiley, New-York, NY.
关26兴 George, E. I., and McCulloch, R. E., 1993, ‘‘Variable Selection via Gibbs
Sampling,’’ Journal of the American Statistical Society, 88, pp. 881– 889.
关27兴 Carlin, B. P., and Chib, S., 1995, ‘‘Bayesian Model Choice via Markov Chain
Acknowledgments Monte Carlo,’’ Journal of the Royal Statistical Society Series B, 77, pp. 473–
The authors acknowledge the VTT Technical Research Center 484.
关28兴 Kulczycki, P., 2001, ‘‘An Algorithm for Bayes Parameter Estimation,’’ ASME
of Finland for sharing the ‘‘Dynamic properties of resilient J. Dyn. Syst., Meas., Control, 123, pp. 611– 614.
mounts’’ benchmark data. The authors are very grateful to Mandy 关29兴 Yuen, K. V., Katafygiotis, L. S., Papadimitriou, C., and Mickleborough, N. C.,
Cundy, technical staff member at Los Alamos National Labora- 2001, ‘‘Optimal Sensor Placement Methodology for Identification With Un-
tory, for helping them with the numerical implementation. Dr. measured Excitation,’’ ASME J. Dyn. Syst., Meas., Control, 123, pp. 677– 686.
关30兴 Cybenko, G., 1989, ‘‘Approximation by Superpositions of a Sigmoidal Func-
Kerschen is supported by a grant from the Belgian National Fund tion,’’ Mathematics of Control, Signals, and Systems, 2, pp. 303–314.
for Scientific Research 共FNRS兲, which is gratefully acknowl- 关31兴 Walter, E., and Pronzato, L., 1997, Identification of Parametric Models From
edged. Experimental Data, Springer-Verlag, Berlin 共Germany兲.

关32兴 Hasselman, T. K., Anderson, M. C., and Wenshui, G., 1998, ‘‘Principal Com- 关35兴 Kerschen, G., Lenaerts, V., Marchesiello, S., and Fasana, A., 2001, ‘‘A Fre-
ponents Analysis for Nonlinear Model Correlation, Updating and Uncertainty quency Domain vs. a Time Domain Identification Technique for Non-linear
Evaluation,’’ Proceedings of the 16th International Modal Analysis Confer- Parameters Applied to Wire Rope Isolators,’’ ASME J. Dyn. Syst., Meas.,
ence, pp. 664 – 651. Control, 123, pp. 645– 650.
关33兴 Mottershead, J. E., and Friswell, M. I., 1993, ‘‘Model Updating in Structural 关36兴 Doebling, S. W., 2002, ‘‘Structural Dynamics Model Validation: Pushing the
Dynamics: A Survey,’’ J. Sound Vib., 162„2…, pp. 347–375. Envelope,’’ Proceedings of the International Conference on Structural Dynam-
关34兴 http://www.ulg.ac.be/ltas-vis/costf3/costf3.html, portal Web site of the ics Modelling—Test, Analysis, Correlation and Validation, Madeira Island
COST-F3 action. 共Portugal兲.

Journal of
Vibration
and Acoustics Technical Briefs
Closed-Form Exact Solution to Hⴥ standard model of the DVA. In 1928, Ormondroyd and Den Har-
tog 关2兴 pointed out the damping of the DVA has an optimum value
Optimization of Dynamic so as to minimize the maximum amplitude response of the pri-
Vibration Absorbers „Application to mary system. Such optimization criterion is now known as H ⬁
optimization. In 1932, Hahnkamm 关3兴 derived the optimum tuning
Different Transfer Functions parameter ␯ opt of the DVA using special points in which all re-
and Damping Systems…
Toshihiko Asami
Mem. ASME
Professor, Department of Mechanical Engineering,
Himeji Institute of Technology, 2167 Shosha, Himeji,
Hyogo 671-2201, Japan
Osamu Nishihara
Mem. ASME
Associate Professor, Department of Systems Science,
Kyoto University, Yoshida-Honmachi, Sakyo-ku, Kyoto
606-8501, Japan
Fig. 1 Systems with viscous damped DVA „Voigt type DVA… „a…
e-mail: [email protected] Force excitation system „b… Motion excitation system
Table 1 Transfer functions of the systems shown in Figs. 1

H ⬁ optimization of the dynamic vibration absorbers is a classical and 2
optimization problem, and has been already solved more than 50 Motion excitation system shown in Fig. 1共b兲
years ago. It is a well-known solution, but we know this solution is Force excitation system
only an approximate one. Recently, one of the authors has pro- shown in Fig. 1共a兲 Absolute response Relative response
冏冏冏冏
posed a new method for attaining the H ⬁ optimization of the
absorber in linear systems. The new method enables us to obtain
the exact algebraic solution of the H ⬁ optimization problem of the
1f
冏冏冏冏
x1
xst
⫽
x1
f 0 /k 1
1
冏冏 x1
x0
4
␻ 21 y 1
ẍ 0
1 y1
⫽ 2
␭ x0
absorber. In this paper, we first apply this method to the design
optimization of a viscous damped (Voigt type) absorber and a
hysteretic damped absorber attached to undamped primary sys-
2f
冏冏冏冏
ẋ1
␻1xst
⫽␭
x1
f 0 /k 1
2
冏冏冏冏
ẋ 1
␻ 1x 0
⫽␭
x1
x0
5
冏冏冏冏
␻1y 1 1 y 1
ẋ 0
⫽
␭ x0
tems. For each absorber, six different transfer functions are taken
here as performance indices to vibration suppression or isolation.
As a result, we found the closed-form exact solutions to all trans-
3f 冏冏冏冏
ẍ1
␻21xst
⫽␭ 2
x1
f 0 /k 1
3 冏冏冏冏
ẍ 1
␻ 21 x 0
⫽␭ 2
x1
x0
6 冏冏 y1
x0
fer functions. The solutions obtained here are then compared with
those of the approximate ones. Finally, we present the closed-form
exact solutions to the hysteretic damped absorber attached to
damped primary systems. 关DOI: 10.1115/1.1569514兴
1 Introduction
The dynamic vibration absorber 共DVA兲, first invented by Frahm
关1兴 in 1909, is a passive vibration control device. As shown in Fig.
1, it essentially consists of a mass, a spring and a dashpot, which
is attached to a primary system subjected to an exciting force or
motion. This type of DVA is called the Voigt type DVA and it is a

in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received April 2000; Fig. 2 Systems with hysteretic damped DVA „a… Force excita-
Revised Nov. 2002. Associate Editor: R. L. Clark. tion system „b… Motion excitation system
Table 2 Expressions for the transfer functions and definition of the symbols of the viscous
damped systems shown in Fig. 1
No. Transfer function Expressions
冏冏冑
N f (␭)⫽( ␯ 2 ⫺␭ 2 ) 2 ⫹(2 ␨ 2 ␯ ␭) 2
1f x1 N f共␭兲
⫽ N a (␭)⫽ 关 1⫹(2 ␨ 1 ␭) 2 兴关 ( ␯ 2 ⫺␭ 2 ) 2 ⫹(2 ␨ 2 ␯ ␭) 2 兴
f 0 /k1 D共 ␭ 兲
1
冏冏冑
x1
x0
⫽
N a共 ␭ 兲
D共 ␭ 兲
N r (␭)⫽ 关 (1⫹ ␮ ) ␯ 2 ⫺␭ 2 兴 2 ␭ 4 ⫹(1⫹ ␮ ) 2 (2 ␨ 2 ␯ ␭) 2 ␭ 4
D(␭)⫽ 关 (1⫺␭ 2 )( ␯ 2 ⫺␭ 2 )⫺ ␮ ␯ 2 ␭ 2 ⫺4 ␨ 1 ␨ 2 ␯ ␭ 2 兴 2
6
冏冏冑
y1
x0
⫽
N r共 ␭ 兲
D共 ␭ 兲
⫹(2␭) 2 兵 ( ␯ 2 ⫺␭ 2 ) ␨ 1 ⫹ 关 1⫺(1⫹ ␮ )␭ 2 兴 ␨ 2 ␯ 其 2
natural frequency ␭⫽ ␻ / ␻ 1 forced frequency ratio

␻ 1 ⫽ 冑k 1 /m 1
of primary system ␮ ⫽m 2 /m 1 mass ratio
natural frequency natural frequency

␻ 2 ⫽ 冑k 2 /m 2 ␯⫽␻2 /␻1
of absorber ratio
absolute displacement c1 damping ratio

x1 ␨1⫽
of primary system 2m1␻1 of primary system
relative displacement c2 damping ratio

y 1 ⫽x 1 ⫺x 0 between primary system ␨2⫽
and foundation 2m2␻2 of absorber
sponse curves pass through independent of the absorber damping

c 2 or ␨ 2 when the primary system has no damping (c 1 or ␨ 1
equals zero兲. Finally, Brock 关4兴 derived the optimum absorber
damping ␨ 2opt in 1946. And now, we all know these optimum
tuning and damping parameters through the textbook written by
Den Hartog 关5兴. Their method utilizing the special points of the
curves is now called the fixed-points method.
Although the fixed-points method gives us a simple and conve-
nient way for solving the H ⬁ optimization problem of the DVA, it
is only an approximate method. Recently, we proposed a new
method for finding the exact solution to the H ⬁ optimization 关6兴.
Using this method, we solved a standard H ⬁ optimization prob-
lem, that is, the minimization of the compliance transfer function
of the system with Voigt type DVA. In this paper, we will apply
this method to different transfer functions listed in Table 1. We
also apply this method to optimizing another DVA system shown
in Fig. 2. This type of DVA is called the hysteretic damped DVA.
The principal difference between the viscous and hysteretic damp-
ing is that for the viscous damping the energy dissipated per cycle
depends linearly on the frequency of vibration; whereas for the
hysteretic damping it is independent of the frequency.
Table 1 shows the transfer functions taken here as performance
indices to vibration suppression or isolation: No. 1 f through No.
3 f are transfer functions in the force excitation systems shown in
Figs. 1共a兲 and 2共a兲, and No. 1 through No. 6 are those in the
motion excitation systems shown in Figs. 1共b兲 and 2共b兲. Transfer
functions No. 1 through No. 3 represent absolute responses of the
primary system and No. 4 through No. 6 represent relative re-
sponses between the primary system and the base. In this table,
dot over a symbol indicates the derivative with respect to time.
Other symbols are explained in Table 2.
The exact solutions obtained here are compared with the ap-
proximate ones reported by Korenev and Reznikov 关7兴.
Fig. 3 Frequency response curves of the undamped primary 2 Fixed-Points Approach as an Approximate Method
system with a viscous damped absorber „graphical representa-
tion of transfer function No. 1… „a… General view of the curves Figure 3共a兲 shows some frequency response curves of the un-
„b… Close up of the curves near the points P and Q damped primary system ( ␨ 1 ⫽0) with an optimally tuned DVA

Table 3 Approximate solutions to the H ⴥ optimization in the design of viscous damped absorber attached to undamped primary
system „derived from the fixed-points method…
Transfer Abscissas ␭ P and ␭ Q Height of the

No. function Optimum tuning ␯ opt Optimum damping ratios ␨ P and ␨ Q of the fixed points fixed points
1 x1
x0
冏冏冑
⫽
N a共 ␭ 兲
D共 ␭ 兲
1
1⫹ ␮ 冑冉冑冊 ␮
8 共 1⫹ ␮ 兲
3⫿
␮
2⫹ ␮
冑 1
1⫹ ␮
冉冑冊
1⫿
2⫹ ␮
␮
冑 ␮␮
2⫹
2
冏冏冏冏
ẋ 1
␻ 1x 0
⫽␭
x1
x0
1
1⫹ ␮
冑 2⫹ ␮
2
1
4 共 2⫹ ␮ 兲
冑冑
␮ 关 24⫹24␮ ⫹5 ␮ 2 ⫿ 共 4⫹ ␮ 兲 ␮ 共 8⫹5 ␮ 兲兴
1⫹ ␮
冑 4⫹3 ␮ ⫿ 冑␮ 共 8⫹5 ␮ 兲
2 共 1⫹ ␮ 兲共 2⫹ ␮ 兲
冑␮ 2⫹ ␮
共 1⫹ ␮ 兲
3 冏冏冏冏
ẍ 1
␻ 21 x 0
⫽␭ 2
x1
x0 冑 1
1⫹ ␮
冑冋冑册
1
2
3␮
2⫹ ␮
1⫿
␮
2 共 1⫹ ␮ 兲
冑 1
2⫹ ␮
冉冑冊
2⫿
1⫹ ␮
2␮
冑␮ 2
共 1⫹ ␮ 兲
4 冏冏冏冏
␻ 21 y 1
ẍ 0
1 y1
⫽ 2
␭ x0
1
1⫹ ␮
冑 2⫺ ␮
2
冑冉冑冊 ␮
4 共 1⫹ ␮ 兲共 2⫺ ␮ 兲
3⫾
␮
2
冑 1⫹ ␮
1
冉冑冊
1⫿
␮
2
共 1⫹ ␮ 兲冑␮
2
5
冏冏冏冏
␻1y 1 1 y 1
ẋ 0
⫽
␭ x0
1
1⫹ ␮
冑 3␮
8 共 1⫹ ␮ 兲
冑 4⫹ ␮ ⫿ 冑␮ 共 8⫹ ␮ 兲
4 共 1⫹ ␮ 兲
冑 ␮␮
2 共 1⫹ 兲
6
冏冏冑
y1
⫽
N r共 ␭ 兲 1
1⫹ ␮
冑 2⫹ ␮
2
冑 ␮
冉冑冊
3⫿
␮
冑 2⫹ ␮
冉冑冊
1⫿
␮ 冑 ␮␮2⫹
x0 D共 ␭ 兲 8 共 1⫹ ␮ 兲 2⫹ ␮ 2 共 1⫹ ␮ 兲 2⫹ ␮
whose mass is one-tenth of the main mass 共i.e., ␮⫽0.1兲. Three resonant amplitudes are adjusted to equal heights in the exact
response curves corresponding to zero, optimum and infinite ab- solution and they are smaller than the higher resonant amplitude
sorber damping ␨ 2 are drawn in this figure. If there is no primary in the approximate solution.
damping, then all response curves pass through two fixed points,
named P and Q, as shown in Fig. 3共a兲. It has been observed that 3.2 Case 2: Hysteretic Damped Absorber. Similarly,
there is a trade-off relationship between the heights of the fixed Tables 5 and 6 show the approximate and exact solutions of the
points. The fixed-points method is essentially based on this obser- H ⬁ optimization design of the hysteretic damped absorber. This
vation. On the basis of this method, one can find the solution such time, the expressions based on our method are as simple as those
that first by a proper choice for ␯ the two fixed points P and Q are of the fixed-points theory, except for No. 2 and 5. Figure 5 com-
adjusted to equal heights, and then by a proper choice for ␨ 2 the pares the approximate and exact solutions for the condition that
response curve passes simultaneously with a horizontal tangent ␮⫽0.1. This figure reconfirms us in the view that the two resonant
through the two fixed points. However, the special values of ␨ 2 , amplitudes of the system are equal in heights in the exact solution.
say ␨ P and ␨ Q , for which the curve passes horizontally through In comparison with Figs. 4 and 5, it can be seen that the optimized
the points P and Q, respectively, are generally not equal to each maximum amplitude responses of the system with hysteretic
other 关see Fig. 3共b兲兴. The average of them is usually adopted as an damped absorber is greater than those of the system with viscous
optimum damping ␨ 2opt , but the two resonant amplitudes of the damped absorber. This means that the hysteretic damped absorber
curve are not exactly equal in this case, as shown in Fig. 3共b兲. will give a poor performance than the viscous damped absorber
Hence we recognize that, in general, we cannot reach the exact when they are equal in size.
solution to the H ⬁ optimization by the fixed-points method.
3 Approximate and Exact Solutions to the H ⴥ Optimi-

4 Exact Solution to the Damped Primary System
zation
Unlike the fixed-points method, our method is applicable to the
3.1 Case 1: Viscous Damped Absorber. Table 3 shows the design of DVA attached to the damped primary system. However,
solutions to the H ⬁ optimization design of the viscous damped for the viscous damped systems shown in Fig. 1, we could not find
absorber obtained from the fixed-points method. All solutions are the algebraic solution for all transfer functions defined in Table 1.
expressed by one parameter, mass ratio ␮. As mentioned previ- On the other hand, for the hysteretic damped systems shown in
ously, the optimum tuning parameter ␯ opt described here is such Fig. 2, we obtained the closed-form exact solution for the transfer
that the two fixed points P and Q have equal heights; the abscissas functions 1 f , 1 and 6. The solution is given in Table 7. We can
and heights of them are shown in this table. They are approximate demonstrate that if the primary system has no damping or ␩ 1
solutions except for No. 5, since the two damping ratios ␨ P and ␨ Q ⫽0, then the expressions shown in Table 7 reduce to those in
are different from each other. Of the double sign of the expres- Table 6. A numerical representation for typical primary damping
sions the upper sign represents ␨ P and the lower ␨ Q . No. 5 is the ␩ 1 is given in Table 8. Figure 6 shows an example of the response
only exact solution obtained from the fixed-points method. curves of the system with optimally tuned and damped DVA. It
We propose the exact solution shown in Table 4. The procedure may be observed that the two peak values of the curves shown in
for obtaining the solution is refer to 关6兴. The expressions are more solid lines are exactly the same; whereas those of the dashed lines
complicated than those in Table 3, but most of them are easy to 共the solution to the undamped primary system兲 are different. It is
calculate on the hand-held calculators. The calculation examples interesting to note that the optimum tuning and damping param-
for a mass ratio ␮⫽0.1 are shown in Fig. 4, where the exact eters, ␯ opt and ␨ 2opt of the force excitation system are the same as
solution is drawn with a solid line and the approximate one with a those of the motion excitation system 共compare the expressions of
dashed line. It may be observed from this figure that the two No. 1 f and No. 1兲.

Journal of Vibration and Acoustics
Table 4 Exact solutions to the H ⴥ optimization in the design of viscous damped absorber attached to undamped primary system „derived from our method…
Transfer
No. function Optimum tuning ␯ opt Optimum damping ␨ 2opt Maximum amplitude
1
冏冏
x1
x0
2
1⫹␮
冑 2 关 16⫹23␮ ⫹9 ␮ 2 ⫹2 共 2⫹ ␮ 兲冑4⫹3 ␮ 兴
3 共 64⫹80␮ ⫹27␮ 2 兲
1
4
冑 8⫹9 ␮ ⫺4 冑4⫹3 ␮
1⫹ ␮
1
3␮
冑共 8⫹9 ␮ 兲 2 共 16⫹9 ␮ 兲 ⫺128共 4⫹3 ␮ 兲 3/2
3 共 32⫹27␮ 兲
2
冏冏
ẋ 1
␻ 1x 0
1
1⫹ ␮
冑⫺ 共 1⫹ ␮ 兲 ⫹ 冑2 共 1⫹ ␮ 兲共 2⫹ ␮ 兲冑 1
2
⫺
1
冑2 共 1⫹ ␮ 兲共 2⫹ ␮ 兲
冑冋1 2⫹ ␮
␮ 2 共 1⫹ ␮ 兲
⫹ 冑 2⫹ ␮
2 共 1⫹ ␮ 兲
册
3 冏冏ẍ 1
␻ 21 x 0
2
8⫹5 ␮
冑 2
3
共 16⫹7 ␮ ⫹ 冑64⫺16␮ ⫺26␮ 2 兲
1
8⫹5 ␮
冑 64⫹88␮ ⫹33␮ 2
2
⫺ 共 4⫹3 ␮ 兲冑c 0
c 0 ⫽64⫺16␮ ⫺26␮ 2
8
3␮
冑 16⫹25␮ ⫺2 冑64⫺16␮ ⫺26␮ 2
3 共 32⫹27␮ 兲
1
1⫹ ␮
冑 1
共 2r 1 ⫺ ␮ ⫹ 冑16r 21 ⫺4r 1 ␮ ⫹ ␮ 2 兲
1
冑 1⫺r 1
1⫹ ␮
共 2r 1 ⫹ ␮ ⫹ 冑16r 21 ⫺4r 1 ␮ ⫹ ␮ 2 兲
1⫹ ␮
6 2r 1
冑1⫺r 21
p 0 ⫽⫺262144⫹196608␮ ⫹233472␮ 2 ⫺598016␮ 3 ⫹448320␮ 4 ⫺120624␮ 5
冏冏冋冑冑冑册冦
⫹6371␮ 6 ⫹48冑6 ␮ 3/2共 64⫺56␮ ⫹13␮ 2 兲 3/2冑共 2⫺ ␮ 兲共 32⫹40␮ ⫺ ␮ 2 兲
4
␻ 21 y 1
ẍ 0 再冎
r1
r2
⫽
1
16
␮⫾
p 1 ⫹q 1
3
⫾
2q 1 ⫺p 1
3
⫿4 ␮ q 0
3
p 1 ⫹q 1
p 1 ⫽p 1/3 ⫺1/3
0 ⫹p 0 共 8⫺7 ␮ 兲共 512⫹192␮ ⫺168␮ 2 ⫺31␮ 3 兲
q 0 ⫽96⫺72␮ ⫹13␮ 2
q 1 ⫽128⫺32␮ ⫺19␮ 2
再冎
r3
r4
⫽
1
16
␮⫿冋冑 p 1 ⫹q 1
3
⫾ 冑 2q 1 ⫺p 1
3
⫾4 ␮ q 0 冑 3
p 1 ⫹q 1
册
5
冏冏
␻1y 1
ẋ 0
1
1⫹ ␮
冑 3␮
8 共 1⫹ ␮ 兲
冑 2 共 1⫹ ␮ 兲
␮
冑冑
JULY 2003, Vol. 125 Õ 401
冑
共 8⫹9 ␮ 兲 2 共 16⫹9 ␮ 兲 ⫺128共 4⫹3 ␮ 兲 3/2
8⫹9 ␮ ⫺4 冑4⫹3 ␮
冏冏
1
1 1 1
共 16⫹9 ␮ ⫹4 冑4⫹3 ␮ 兲
6 y1
3␮ 3 共 32⫹27␮ 兲
x0 2 共 1⫹ ␮ 兲 6 4 1⫹ ␮
Fig. 5 Comparison between exact and approximate solutions
Fig. 4 Comparison between exact and approximate solutions to the H ⴥ optimization of the system with a hysteretic damped
to the H ⴥ optimization of the system with a viscous damped DVA „a… Transfer function No. 1 „b… Transfer function No. 4
DVA „a… Transfer function No. 1 „b… Transfer function No. 4
Table 5 Approximate solutions to the H ⴥ optimization in the design of hysteretic damped absorber attached to undamped
primary system „derived from the fixed-points method…
Transfer Optimum tuning Abscissas ␭ P and ␭ Q Height of the

No. function ␯ opt Optimum loss factors ␩ P and ␩ Q of the fixed points fixed points
1
冏冏冑
x1
x0
⫽
N a共 ␭ 兲
D共 ␭ 兲
1
1⫹ ␮
冑 ␮␮2⫹
关 3⫹2 ␮ ⫿2 冑␮ 共 2⫹ ␮ 兲兴冑冉冑冊冑
1
1⫹ ␮
1⫿
2⫹ ␮
␮ 2⫹ ␮
␮
2
冏冏冏冏
ẋ 1
␻ 1x 0
⫽␭
x1
x0
1
1⫹ ␮
冑 2⫹ ␮
2
冑␮ 关 24⫹50␮ ⫹30␮ 2 ⫹5 ␮ 3 ⫿2 共 5⫹5 ␮ ⫹ ␮ 2 兲冑␮ 共 8⫹5 ␮ 兲兴
共 2⫹ ␮ 兲 2
冑冑冑
4⫹3 ␮ ⫿ ␮ 共 8⫹5 ␮ 兲
2 共 1⫹ ␮ 兲共 2⫹ ␮ 兲
2⫹ ␮
␮ 共 1⫹ ␮ 兲
3 冏冏冏冏
ẍ 1
␻ 21 x 0
⫽␭ 2
x1
x0 冑 1
1⫹ ␮
1
2⫹ ␮
冑3 ␮ 关 2⫹3 ␮ ⫿2 冑2 ␮ 共 1⫹ ␮ 兲兴冑冉冑冊冑
1
2⫹ ␮
2⫿
2␮
1⫹ ␮
2
␮ 共 1⫹ ␮ 兲
4 冏冏冏冏
␻ 21 y 1
ẍ 0
⫽
1 y1
␭2 x0
1
1⫹ ␮
冑 2⫺ ␮
2
冑␮ 共 6⫺ ␮ ⫿2 冑2 ␮ 兲
2⫺ ␮
冑冉冑冊
1
1⫹ ␮
冑1⫿
␮
2
共 1⫹ ␮ 兲
2
␮
5
冏冏冏冏
␻1y 1 1 y 1
ẋ 0
⫽
␭ x0
1
1⫹ ␮ 冑 3 ␮ 关 4⫹ ␮ ⫿ 冑␮ 共 8⫹ ␮ 兲兴
8
冑冑冑
4⫹ ␮ ⫿ ␮ 共 8⫹ ␮ 兲
4 共 1⫹ ␮ 兲
2 共 1⫹ ␮ 兲
␮
6
冏冏冑
y1
x0
⫽
N r共 ␭ 兲
D共 ␭ 兲
1
1⫹ ␮
冑 2⫹ ␮
2
冑 ␮ 关 3⫹2 ␮ ⫿2 冑␮ 共 2⫹ ␮ 兲兴
2⫹ ␮
冑冉冑冊冑
2⫹ ␮
2 共 1⫹ ␮ 兲
1⫿
2⫹ ␮
␮
2⫹ ␮
␮

Table 6 Exact solutions to the H ⴥ optimization in the design of hysteretic damped absorber attached to undamped primary system „derived from our method…
No. T.F. Optimum tuning ␯ opt Optimum damping ␩ 2opt Maximum amplitude Nomenclature
1 共 3⫹ ␮ 兲 2 共 1⫹ ␮ 兲 p 0 ⫽ 关 ⫺2 ␮ 2 (1⫹ ␮ ) 2 兴 1/3
1 x1
2
x0
冏冏 1⫹␮
冑␮ 冑 ␮ p 1 ⫽2(1⫹ ␮ ⫹ ␮ 2 )⫹3p 0
共 q⫺1⫺ ␮ 兲冑1⫹ ␩ 22opt⫺2 1 3
Journal of Vibration and Acoustics

ẋ1 ⫺b⫹ 冑b 2 ⫺4ac
2 p 2 ⫽2 共 1⫹ ␮ ⫹ ␮ 2 兲 ⫺ p
2 0
␻1x0
冏冏冑共 q⫺1⫹ ␮ 兲共 1⫹ ␩ 22opt兲 ⫺2 共 1⫹ ␮ 兲冑1⫹ ␩ 22opt
冑 2a
冑 q⫺3⫺ ␮
ẍ 1 2 冑2 共 2⫺ ␮ 兲冑3 ␮ 共 8⫺ ␮ 兲 4 p1 2 共 1⫺ ␮ 兲共 2⫹ ␮ 兲共 1⫹2 ␮ 兲
3 q⫽ ⫹ p 2⫹
␻ 21 x 0 4⫹ ␮ 2 共 2⫺ ␮ 兲 2
冏冏冑␮ 共 8⫺ ␮ 兲
冑冑冑2p 1
␻ 21 y 1 1 2⫺ ␮ 冑3 ␮ 共 8⫺ ␮ 兲 4 共 1⫹ ␮ 兲
4 a⫽a 0 q 3 ⫹a 1 q 2 ⫹a 2 q⫹a 3
ẍ 0 2
冏冏 1⫹ ␮
冑 2 共 2⫺ ␮ 兲冑␮ 共 8⫺ ␮ 兲
1 共 q⫺1⫹ ␮ 兲冑1⫹ ␩ 22opt⫺2 共 1⫹ ␮ 兲 ⫺b⫹ 冑b 2 ⫺4ac 1⫹ ␮ b⫽⫺2(b 0 q 3 ⫹b 1 q 2 ⫹b 2 q⫹b 3 )
5 ␻1 y 1
ẋ0
冏冏 1⫹␮
冑共 q⫺1⫺ ␮ 兲共 1⫹ ␩ 22opt兲 ⫺2 冑1⫹ ␩ 22opt
冑 2a
冑 q⫺3⫺ ␮ c⫽⫺ ␮ (c 0 q 3 ⫹c 1 q 2 ⫹c 2 q⫹c 3 )
y1 2 共 3⫹ ␮ 兲 2 共 1⫹ ␮ 兲
6
2
x0
冏冏冑共 1⫹ ␮ 兲共 2⫹ ␮ 兲
冑␮ 冑 ␮
a 0 ⫽5120⫺94464␮ ⫹50688␮ 2 ⫺281728␮ 3 ⫺129492␮ 4 ⫺106667␮ 5 ⫺43693␮ 6 ⫺10176␮ 7 ⫺1024␮ 8
a 1 ⫽⫺6144⫹156160␮ ⫺95616␮ 2 ⫹913632␮ 3 ⫹449296␮ 4 ⫹667848␮ 5 ⫹381560␮ 6 ⫹146199␮ 7 ⫹32064␮ 8 ⫹3072␮ 9
a 2 ⫽⫺23552⫹301824␮ ⫹397568␮ 2 ⫺507584␮ 3 ⫹776140␮ 4 ⫺227683␮ 5 ⫺727803␮ 6 ⫺399091␮ 7 ⫺156903␮ 8 ⫺33600␮ 9 ⫺3072␮ 10
a 3 ⫽⫺12288⫹123904␮ ⫹191872␮ 2 ⫹494176␮ 3 ⫹548168␮ 4 ⫹778822␮ 5 ⫹643981␮ 6 ⫹495675␮ 7 ⫹193878␮ 8 ⫹61309␮ 9 ⫹11712␮ 10⫹1024␮ 11
b 0 ⫽⫺1024⫹37376␮ ⫺36224␮ 2 ⫹56976␮ 3 ⫺7868␮ 4 ⫺22581␮ 5 ⫺18779␮ 6 ⫺7040␮ 7 ⫺1024␮ 8
Coefficients b 1 ⫽1024⫺62208␮ ⫹66432␮ 2 ⫺250672␮ 3 ⫹5564␮ 4 ⫹41644␮ 5 ⫹93648␮ 6 ⫹66753␮ 7 ⫹22656␮ 8 ⫹3072␮ 9
b 2 ⫽5120⫺124928␮ ⫺104064␮ 2 ⫹434608␮ 3 ⫺32052␮ 4 ⫹18799␮ 5 ⫹28627␮ 6 ⫺89049␮ 7 ⫺72753␮ 8 ⫺24192␮ 9 ⫺3072␮ 10
b 3 ⫽3072⫺49920␮ ⫺57472␮ 2 ⫺102288␮ 3 ⫹1140␮ 4 ⫹29134␮ 5 ⫹93055␮ 6 ⫹72685␮ 7 ⫹66494␮ 8 ⫹31691␮ 9 ⫹8576␮ 10⫹1024␮ 11
c 0 ⫽⫺17152⫺12288␮ ⫺172624␮ 2 ⫺160188␮ 3 ⫺161061␮ 4 ⫺84723␮ 5 ⫺24768␮ 6 ⫺3072␮ 7
c 1 ⫽21248⫹25664␮ ⫹446624␮ 2 ⫹517192␮ 3 ⫹800496␮ 4 ⫹602424␮ 5 ⫹290889␮ 6 ⫹78912␮ 7 ⫹9216␮ 8
c 2 ⫽77568⫹132736␮ ⫹227504␮ 2 ⫹651828␮ 3 ⫺200997␮ 4 ⫺713205␮ 5 ⫺614421␮ 6 ⫺314361␮ 7 ⫺83520␮ 8 ⫺9216␮ 9
c 3 ⫽3(13056⫹29888␮ ⫹96768␮ 2 ⫹184128␮ 3 ⫹286918␮ 4 ⫹289257␮ 5 ⫹224231␮ 6 ⫹114406␮ 7 ⫹42977␮ 8 ⫹9792␮ 9 ⫹1024␮ 10)
JULY 2003, Vol. 125 Õ 403

Table 8 Numerical exact solutions for damped primary system
for 1 and 6
„Transfer function is 円 x 1 Õ x 0 円 and mass ratio ␮ is equal to 0.1…
for 1 f
Primary Optimum Optimum loss Maximum
damping ␩ 1 tuning ␯ opt factor ␩ 2opt amplitude
q 1 ⫺2 ␮ 共 3⫹6 ␮ ⫺ ␮ 2 兲 ⫹2 共 1⫺ ␮ 兲共 3⫺ ␮ 2 兲 ␩ 21
0.00 0.909091 0.393700 4.69042
q 1 ⫺2 ␮ 共 3⫹6 ␮ ⫺ ␮ 2 兲 ⫹2 共 1⫺ ␮ 兲共 3⫺ ␮ 2 兲 ␩ 21
0.01 0.907941 0.395643 4.51256
0.02 0.906801 0.397566 4.34740
0.05 0.903437 0.403232 3.91648
Maximum amplitude h max
0.10 0.898000 0.412355 3.36195

Table 7 Exact solutions to the optimization problem in the design of hysteretic damped absorber attached to damped primary system
0.15 0.892754 0.421124 2.94908
6 共 1⫹ ␮ 兲共 1⫹ ␩ 21 兲
6 共 1⫹ ␮ 兲
0.20 0.887677 0.429584 2.63249

0.50 0.859966 0.475493 1.68819
⫹2 ␮ (3⫹ ␮ )(9⫹15␮ ⫹18␮ 2 ⫺6 ␮ 3 ⫺3 ␮ 4 ⫺ ␮ 5 )⫹4 ␮ 2 (1⫺ ␮ )(6⫺9 ␮ ⫹3 ␮ 2 ⫹9 ␮ 3 ⫹7 ␮ 4 ) ␩ 21 ⫺8 ␮ 4 (1⫺ ␮ ) 2 (3⫹ ␮ ) ␩ 41 其
p 1 ⫽⫺ ␮ (3⫹ ␮ ) ⫹3(1⫺ ␮ )(3⫹ ␮ ) 3 (9⫺9 ␮ ⫹21␮ 2 ⫺5 ␮ 3 ) ␩ 21 ⫺12␮ 2 (1⫺ ␮ ) 2 (45⫺3 ␮ 2 ⫺2 ␮ 3 ) ␩ 41 ⫺8 ␮ 4 (1⫺ ␮ ) 3 ␩ 61

e 0 ⫽4 ␮ 2 q 21 ⫹8 ␮ q 1 关 3⫺6 ␮ ⫺6 ␮ 2 ⫺6 ␮ 3 ⫺ ␮ 4 ⫹ ␮ (1⫺ ␮ )(3⫹ ␮ 2 ) ␩ 21 兴 ⫹4(3⫺6 ␮ ⫺6 ␮ 2 ⫺6 ␮ 3 ⫺ ␮ 4 ) 2
e 1 ⫽(1⫺ ␮ ) 2 关 q 1 ⫹2 ␮ (3⫹ ␮ ) 2 ⫹4 ␮ 2 (1⫺ ␮ ) ␩ 21 兴关 q 1 ⫺2 ␮ (3⫹6 ␮ ⫺ ␮ 2 )⫹4 ␮ 2 (1⫺ ␮ ) ␩ 21 兴

⫺16␮ 2 (1⫺ ␮ )(9⫺6 ␮ ⫹12␮ 2 ⫹18␮ 3 ⫹3 ␮ 4 ⫺4 ␮ 5 ) ␩ 21 ⫹32␮ 4 (1⫺ ␮ ) 2 (3⫺ ␮ 2 ) ␩ 41
e 2 ⫽⫺2(1⫺ ␮ ) 兵 ␮ q 21 ⫺q 1 关 3⫹18␮ ⫹6 ␮ 2 ⫹6 ␮ 3 ⫺ ␮ 4 ⫹2 ␮ 2 (1⫺ ␮ )(3⫺ ␮ ) ␩ 21 兴

冑
q 0 ⫽ ␮ 2 关 p 1 ⫺3(1⫺ ␮ ) 2 ␩ 1 冑3 关 ⫺2 ␮ ⫹(1⫺ ␮ ) ␩ 21 兴关 (3⫹ ␮ ) 3 ⫹8 ␮ 2 ␩ 21 兴 3 兴

e 3 ⫽12␮ (1⫹ ␮ )(1⫺ ␮ ) 2 ␩ 1 关 q 1 ⫺2(3⫹3 ␮ 2 ⫹2 ␮ 3 )⫹4 ␮ 2 (1⫺ ␮ ) ␩ 21 兴
p 0 ⫽(3⫹ ␮ ) 4 ⫺4(1⫺ ␮ )(3⫹ ␮ )(9⫺3 ␮ ⫹2 ␮ 2 ) ␩ 21 ⫹4 ␮ 2 (1⫺ ␮ ) 2 ␩ 41
冑
共 1⫺ ␮ 兲
共 1⫺ ␮ 兲
Optimum damping ␩ 2opt
b⫽e 3 共 3e 1 ⫺e 2 兲
⫺b⫹ 冑b 2 ⫺4ac
c⫽e 1 e 2 ⫺3e 23
a⫽e 0 e 1 ⫺e 23
2a
q1⫽ 1/3 ⫹q 1/3

0
␮2p0
q0
再 for 6
for 1 f and 1
关 2 共 3⫹3 ␮ 2 ⫹2 ␮ 3 兲 ⫺4 ␮ 2 共 1⫺ ␮ 兲 ␩ 21 ⫺q 1 兴共 1⫹ ␩ 22opt兲
Fig. 6 Absolute displacement response of the damped pri-

mary system with optimally tuned and damped DVA
2 共 3⫹3 ␮ 2 ⫹2 ␮ 3 兲 ⫺4 ␮ 2 共 1⫺ ␮ 兲 ␩ 21 ⫺q 1
5 Concluding Remarks
6 共 1⫹ ␮ 兲共 1⫹ ␩ 22opt兲
6 共 1⫹ ␮ 兲
Optimum tuning ␯ opt
The H ⬁ optimization of a dynamic vibration absorber attached

to a linear primary system is a classical optimization problem.
Although the system is very simple, it is difficult to find the exact
solution. The conventional method, well-known as the fixed-
points method, is only an approximate one to solve the problem.
In 2002, we proposed a new method to find the exact solution to
the H ⬁ optimization.
First in this paper, we applied this method to solve the H ⬁
optimization problem to different transfer functions and DVAs
冑
冑
共viscous damped DVA and hysteretic damped DVA兲 under the

共 1⫹ ␮ 兲共 1⫺ ␮ 兲
condition that the primary system has no damping, and found

共 1⫺ ␮ 兲共 1⫹ ␩ 21 兲
closed-form exact solutions for all transfer functions. Some of the

1
solutions are simple and others complicated, but we can handle

1⫹ ␮
them by the hand-held calculators. And now, we can evaluate the

accuracy of the conventional solutions by comparing them with
冑
our solutions. From the comparison of these solutions we see that

冑冑
the difference between the exact and approximate solutions is neg-

ligibly small in the practical range of mass ratio 共␮ is less than
Transfer function
about 0.2兲. Nevertheless, we believe that the conventional expres-

sions shown in Tables 3 and 5 should be replaced by the new
冏冏 f 0 /k1
expressions in Tables 4 and 6, respectively, because the latter are

冏冏
冏冏
y1
x1
x0
x0
x1
exact solutions.
Expressions
Next, we tried to solve the same optimization problem for a

generalized system where damping is present in the primary sys-
tem, and found the closed-form exact solutions for the hysteretic
No.
damped system.
1f

The last unsolved problem is the H ⬁ optimization of the vis- 2 Problem Formulation and Initial Analysis
cous damped dynamic vibration absorber attached to damped pri-
Consider an equation of motion describing parametric excita-
mary systems. However, the algebraic approach is probably im-
tion of the form
possible to this system. We proposes simultaneous equations for
calculating numerically the exact solution 关6兴. M u ,tt ⫹Gu ,t ⫹Ku⫹␧p 共 t 兲共 M ⬘ u ,tt ⫹G ⬘ u ,t ⫹K ⬘ u 兲 ⫽0 (1)
where M, G, K, M ⬘ , G ⬘ , and K ⬘ are linear, real, spatial differen-
References tial 共or matrix兲 operators, u(x,t) is the system response, x is the
关1兴 Frahm, H., 1911, ‘‘Device for Damping Vibrations of Bodies,’’ U.S. Patent, spatial coordinate, t is time, ␧Ⰶ1 is a small parameter, p(t) is a
No. 989, 958, pp. 3576 –3580. periodic function in time, and a comma indicates partial differen-
关2兴 Ormondroyd, J., and Den Hartog, J. P., 1928, ‘‘The Theory of the Dynamic
Vibration Absorber,’’ ASME J. Appl. Mech., 50共7兲, pp. 9–22.
tiation. The operators M, G, and K are associated with a complex
关3兴 Hahnkamm, E., 1932, ‘‘Die Dämpfung von Fundamentschwingungen bei ve- inner product 具,典 and enjoy the properties
ränderlicher Erregergrequenz,’’ Ing. Arch., 4, pp. 192–201, 共in German兲.
关4兴 Brock, J. E., 1946, ‘‘A note on the Damped Vibration Absorber,’’ ASME J. M ⬎0 K⬎0 M * ⫽M G * ⫽⫺G K * ⫽K (2)
Appl. Mech., 13„4…, p. A-284.
关5兴 Den Hartog, J. P., 1956, Mechanical Vibrations, 4th ed., McGraw-Hill, New where the * indicates the adjoint. In other words, the unperturbed
York. 共␧⫽0兲 problem is a positive definite gyroscopic system. The
关6兴 Nishihara, O., and Asami, T., 2002, ‘‘Closed-Form Solutions to the Exact eigensolutions associated with the unperturbed system are
Optimizations of Dynamic Vibration Absorbers 共Minimizations of the Maxi-
mum Amplitude Magnification Factors兲,’’ ASME J. Vibr. Acoust., 124共4兲, pp.
兵 i ␻ n , ␾ n 其 where, due to 共2兲, all the eigenvalues are imaginary,
576 –582. i.e., ␻ n is real. These solutions are numbered such that n⫽⫾1,
关7兴 Korenev, B. G., and Reznikov, L. M., 1993, Dynamic Vibration Absorbers, p. ¯ 其 where the overbar
⫾2,⫾3 . . . and 兵 i ␻ ⫺n , ␾ ⫺n 其 ⫽ 兵 ⫺i ␻ n , ␾ n
11, John Wiley & Sons, New York.
denotes complex conjugation.
The adjoint eigenfunction is the eigenfunction itself; hence
eigensolutions can be normalized such that
␻ n ␻ m 具 M ␾ n , ␾ m 典 ⫹ 具 K ␾ n , ␾ m 典 ⫽ ␦ nm (3)
Determining Stability Boundaries
where ␦ nm is the Kronecker delta. The eigenexpansion
Using Gyroscopic Eigenfunctions ⫾⬁
Anthony A. Renshaw
u⫽ 兺
n⫽⫾1,⫾2, . . .
d n共 t 兲 ␾ n (4)
Associate Professor, Dept. of Mechanical Engng. can always be constrained such that 关3兴
Columbia University, New York, NY 10027 ⫾⬁ ⫾⬁
u ,t ⫽ 兺
n⫽⫾1, . . .
d n,t ␾ n ⫽ 兺
n⫽⫾1, . . .
i ␻ nd n␾ n (5)
By taking advantage of modal decoupling and reduction of order, This last constraint 共5兲 enables one to decouple the unperturbed
we derive a simplified procedure for applying the method of modal equations 关3兴. Substituting Eq. 共4兲 into 共1兲, taking the inner
multiple scales to determine the stability boundaries of paramet- product with ⫺i ␻ m ␾ m and utilizing Eq. 共5兲 gives a system of
rically excited, gyroscopic systems. The analytic advantages of the modal equations in which the left hand side, unperturbed equa-
procedure are illustrated with three examples. tions are a set of first order, decoupled, ordinary differential equa-
关DOI: 10.1115/1.1569944兴 tions:
1 Introduction
d m,t ⫺i ␻ m d m ⫽␧i ␻ m p 共 t 兲冉兺 n
⬘ ⫹d n,t G nm
d n,tt M nm ⬘ ⫹d n K nm
⬘ 冊
The complex eigenfunctions of positive definite, linear, gyro- (6)
scopic systems possess a number of properties that make them ⬘ ⫽ 具 M ⬘ ␾ n , ␾ m 典 , etc.
where M nm
advantageous and efficient in analytic methods. First and fore- The parametric instability boundaries are determined from Eq.
most, these eigenfunctions can be used to decouple system equa- 共6兲 by applying the method of multiple scales. Define the time
tions, thus providing exact modal solutions even for continuous scales T 0 ⫽t and T 1 ⫽␧t and let d n (t)⫽D n0 (T 0 ,T 1 )
systems 关1–3兴. These eigenfunctions can also be used in pertur- ⫹␧D n1 (T 0 ,T 1 )⫹ . . . . Substitution into Eq. 共6兲 gives the leading
bation studies 关4兴. Unfortunately, these eigenfunctions do not ap- order equation
pear to offer meaningful advantages for numerical methods 关5–7兴.
In this Tech Brief, we demonstrate that complex gyroscopic D m0,T 0 ⫺i ␻ m D m0 ⫽0 (7)
eigenfunctions are particularly efficient for determining the stabil-
ity boundaries of parametrically excited systems using the method whose solution is
of multiple scales. This efficiency is derived from two properties: D m0 ⫽A m0 共 T 1 兲 e i ␻ m T 0 (8)
decoupling and reduction of order. Our contribution here is one of
presentation and the examples; we make no claim of originality The unknown coefficient A m0 (T 1 ) is determined by eliminating
for the method which has been used in previous parametric sta- secular terms of the form e i ␻ m T 0 from the right hand side of the
bility studies 关8 –10兴. Our aim here is to present the analysis in as order ␧ equation, given by
general and simple a manner as possible so that the efficiency of
the procedure is obvious and its application is accessible to those
unfamiliar with it. To that end, the procedure is used to quickly
D m1,T 0 ⫺i ␻ m D m1 ⫽⫺A m0,T 1 e i ␻ m T 0 ⫹i ␻ m p 共 T 0 兲兺An
n0 e
i␻nT0
␤ nm
analyze three parametrically excited systems. The efficiency of the (9)
procedure is evident when our brief analyses are compared with where
the alternative approaches given in the literature.
⬘ ⫹i ␻ n G nm
␤ nm ⫽⫺ ␻ 2n M nm ⬘ ⫹K nm (10)
in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Aug. 2002; After this step, the procedure varies from problem to problem; we
Revised Jan. 2003. Associate Editor: L. A. Bergman. illustrate different approaches with the following examples.
Example 1. Classical Parametric Resonance D jm1,T 0 ⫺i ␻ jm D jm1
For the classical case p(t)⫽cos(⍀t). The order ␧ equations
become ⫽⫺A jm0,T 1 e i ␻ jm T 0
D m1,T 0 ⫺i ␻ m D m1 ⫽⫺A m0,T 1 e i ␻ m T 0 ⫹

1
i ␻ 共 e i⍀T 0
2 m ⫹i ␻ jm 冉兺k,n
A kn0 e i ␻ kn T 0 R kn 共 r 0 兲 R jm 共 r 0 兲 e i 共 j⫺k 兲 ⍀T 0 冊
⫹e ⫺i⍀T 0 兲兺An
n0 e
i␻nT0
␤ nm (11) (21)
The detuning parameter for a two mode secular combination of
When the first term on the right hand side is the only secular term, terms on the right hand side is
only bounded solutions result. We therefore look for combinations
of two secular terms. Define the detuning parameter ␴ by ␻ jm ⫹␧ ␴ ⫽ ␻ kn ⫹ 共 j⫺k 兲 ⍀ (22)
␻ m ⫹␧ ␴ ⫽ ␻ n ⫹⍀ (12) giving the equations

for some n and m, where, by assumption, ␴ is real. Eliminating the ⫺A jm0,T 1 ⫹i ␻ jm R 2jm 共 r 0 兲 A jm0
secular terms requires both
1 ⫹i ␻ jm R jm 共 r 0 兲 R kn 共 r 0 兲 e i ␴ T 1 A kn0 ⫽0 (23)
⫺A m0,T 1 ⫹ i ␻ m A n0 e i ␴ T 1 ␤ nm ⫽0 (13)
2 ⫺A kn0,T 1 ⫹i ␻ kn R 2kn 共 r 0 兲 A kn0
1
⫺A n0,T 1 ⫹ i ␻ A e i ␴ T 1 ␤ mn ⫽0 (14) ⫹i ␻ kn R kn 共 r 0 兲 R jm 共 r 0 兲 e ⫺i ␴ T 1 A jm0 ⫽0 (24)
2 n m0
Assuming solutions of the form
whose solution has the form
A m0 ⫽a m e ␣ T 1 A n0 ⫽a n e 共 ␣ ⫺i ␴ 兲 T 1 (15) A jm0 ⫽a jm e ␣ T 1 A kn0 ⫽a kn e 共 ␣ ⫺i ␴ 兲 T 1 (25)
Solving these gives and solving, we obtain unstable solutions whenever
i 共 ␴ ⫹ ␻ jm R 2jm 共 r 0 兲 ⫹ ␻ kn R 2kn 共 r 0 兲兲 2 ⫺4 ␴␻ jm R 2jm 共 r 0 兲 ⬍0 (26)
␣ ⫽ 关 ␴ ⫾ 共 ␴ 2 ⫹ ␻ n ␻ m ␤ nm ␤ mn 兲 1/2兴 (16)
2
The most important instability produced by the spring is diver-
The first order stability boundary occurs when ␣ changes from gence instability of a mode with a fixed number of nodal diam-
purely imaginary to complex. In other words, eters. For this case, we let ␻ jm ⫽ ␻ 0 ⬎0, ␻ kn ⫽⫺ ␻ 0 , j⫽⫺k, and
R jm (r 0 )⫽R kn (r 0 ). The conditions for instability are then
⍀⫽ ␻ m ⫺ ␻ n ⫾␧ 冑⫺ ␻ n ␻ m ␤ nm ␤ mn (17)
when the factor under the square root is positive. If it is not ␴ 共 ␴ ⫺4 ␻ 0 R 2jm 共 r 0 兲兲 ⬍0 (27)
positive, the unstable region vanishes. Primary instability occurs In other words, for a combination resonance of two modes with
for ␻ m ⫽⫺ ␻ n . Combination resonance of the summation type the same number of nodal diameters, the stability boundary is
occur when ␻ n ␻ m ⬍0; this is not possible if ␤ nm ␤ mn ⬍0 and real, demarcated by the two lines
i.e., when M ⬘ and K ⬘ are skew symmetric and G ⬘ is symmetric.
Combination resonance of the difference type occur when ␻ n ␻ m ␻ 0 ⫽ j⍀ and ␻ 0 ⫹2 ␻ 0 R 2jm 共 r 0 兲 ⫽ j⍀ (28)
⬎0; this is not possible when ␤ nm ␤ mn ⬎0 and real, i.e., when M ⬘
and K ⬘ are symmetric and G ⬘ is skew symmetric. The system is unstable at rotation speeds immediately above the
Example 2. A Stationary Spring Loading a Rotating Circu- speed at which the natural frequency divided by the number of
lar Plate nodal diameters equals the rotation speed. This, of course, is a
For an axisymmetric rotating circular plate in the rotating frame well known result 关11,12兴. Compare this analysis with 关13兴.
of reference 关11兴, the eigensolutions of the unperturbed system are We also note that when R jm (r 0 )⫽0, the instability region van-
doubly indexed with eigenvalues i ␻ km and eigenfunctions of the ishes. This result is derived in a quite different manner in 关4,14兴.
form Example 3. Washboarding of Circular Saws
Washboarding of industrial circular saws is a self-excited vibra-
␾ km ⫽e ik ␪ R km 共 r 兲 (18) tion of the saw that occurs at certain cutting speeds and leaves the
where (r, ␪ ) are polar coordinates and k and m are integers and cut surface of the workpiece with a sinusoidally varying profile
m⭓0. This describes a mode with 兩 k 兩 nodal diameters and m rather than the desired flat surface. Tian and Hutton 关15兴 present a
nodal circles. We do not need to know the actual forms of M, G, model for washboarding based on the classical regenerative force
K, R km , and ␻ km in order to analyze stability; these are only model used to analyze machine tool chatter 关16兴. Here we con-
needed if explicit values of the frequencies and boundaries are sider a simplified version of their model.
required. A circular saw has N evenly spaced teeth located on its periph-
Here we consider a perturbation comprising a stationary, con- ery at r⫽r 0 and at fixed angles ␪ q ⫽2 ␲ q/N for q⫽1, . . . ,N,
centrated spring transversely loading the rotating plate. In the ro- where these angles are measured in the rotating frame of reference
tating frame of reference, we have p⫽1 and 共i.e., fixed to the saw兲. The saw rotates at angular speed ⍀ so that
the time interval between teeth passing the same angular location
1 on the non-rotating workpiece is ⌬T⫽2 ␲ /N⍀. In an actual saw,
K ⬘⫽ ␦ 共 r⫺r 0 兲 ␦ 共 ␪ ⫹⍀t 兲 (19) only an angular sector of the saw cuts the workpiece at any given
r
time. For our simplified analysis, we make the assumption that the
where ⍀ is the rotation speed of the plate and r 0 is the radial entire periphery of the saw cuts the workpiece at all times. While
location of the spring. M ⬘ ⫽G ⬘ ⫽0. The inner product is this assumption is physically unrealistic, it gives results that are
similar to those derived from more realistic but complicated mod-
⬘ jm ⫽ 具 K ⬘ e ik ␪ R kn ,e i j ␪ R jm 典 ⫽R kn 共 r 0 兲 R jm 共 r 0 兲 e i 共 j⫺k 兲 ⍀t
K kn
eling.
(20) In the rotating frame of reference, the regenerative cutting force
Equation 共9兲 is therefore is given by p⫽1 and

N
1 References
K ⬘ u⫽ 兺
q⫽1 r
␦ 共 r⫺r 0 兲 ␦ 共 ␪ ⫺ ␪ q 兲关 u 共 r 0 , ␪ q ,t 兲关1兴 Meirovitch, L., 1975, ‘‘A Modal Analysis for the Response of Linear Gyro-
scopic Systems,’’ ASME J. Appl. Mech., 42共2兲, pp. 446 – 450.
关2兴 Wickert, J. A., and Mote, C. D., Jr., 1990, ‘‘Classical Vibration Analysis of
⫺u 共 r 0 , ␪ q⫹1 ,t⫺⌬T 兲兴 (29) Axially Moving Continua,’’ ASME J. Appl. Mech., 57, pp. 738 –744.
关3兴 Renshaw, A. A., 1997, ‘‘Modal Decoupling of Systems Described by Three
where u(r, ␪ ,t) is the transverse displacement of the saw and ␦共兲 Linear Operators,’’ ASME J. Appl. Mech., 64, pp. 238 –240.
is the delta function. In other words, the transverse load produced 关4兴 Hryniv, R. O., Lancaster, P., and Renshaw, A. A., 1999, ‘‘A Stability Criterion
by a given tooth cutting the workpiece is proportional to the dif- for Parameter Dependent Gyroscopic Systems,’’ ASME J. Appl. Mech., 66, pp.
ference between its current position and the location of the previ- 660, 664.
ous tooth when it cut through the same angular location of the 关5兴 Jha, R. K., and Parker, R. G., 2000, ‘‘Spatial Discretization of Axially Moving
Media Vibration Problems,’’ ASME J. Vibr. Acoust., 122, pp. 290–294.
workpiece. 关6兴 Lee, K.-Y., and Renshaw, A. A., 1999, ‘‘Solution of the Moving Mass Problem
For this model, Eq. 共9兲 becomes Using Complex Eigenfunction Expansions,’’ ASME J. Appl. Mech., 67, pp.
D jm,T 0 ⫺i ␻ jm D jm ⫽⫺A jm0,T 1 e i ␻ jm T 0
823– 827.
关7兴 Lee, K.-Y., and Renshaw, A. A., 2002, ‘‘A Numerical Comparison of Alterna-
N
tive Galerkin Methods for Eigenvalue Estimation,’’ J. Sound Vib., 253共2兲, pp.
兺兺R
359–372.
i 共 j⫺k 兲 ␪ q
⫹i ␻ jm kn 共 r 0 兲 R jm 共 r 0 兲 e 关8兴 Parker, R. G., and Lin, Y., 2001, ‘‘Parametric Instability of Axially Moving
q⫽1 k,n Media Subjected to Multifrequency Tension and Speed Fluctuations,’’ ASME
J. Appl. Mech., 68, pp. 49–57.
⫻A kn0 e i ␻ jm T 0 关 1⫺e 2 ␲ i 共 k⍀⫺ ␻ kn 兲 /N⍀ 兴 (30) 关9兴 Mockensturm, E. M., Perkins, N. C., and Ulsoy, A. G., 1996, ‘‘Stability and
Limit Cycles of Parametrically Excited, Axially Moving Strings,’’ ASME J.
We eliminate secular response of a single mode by setting Vibr. Acoust., 118, pp. 346 –351.
关10兴 Lee, K.-Y., and Renshaw, A. A., 2002, ‘‘Stability Analysis or Parametrically
A jm0,T 1 ⫽i ␻ jm NR 2jm 共 r 0 兲 A jm0 关 1⫺e 2 ␲ i 共 j⍀⫺ ␻ jm 兲 /N⍀ 兴 (31) Excited Systems Using Spectral Collocation,’’ J. Sound Vib., to appear.
关11兴 Iwan, W. D., and Moeller, T. L., 1976, ‘‘The Stability of a Spinning Elastic
whose solution is Disk with a Transverse Load System,’’ ASME J. Appl. Mech., 43, pp. 485–
490.
A jm0 ⫽exp关 i ␻ jm T 1 R 2jm 共 r 0 兲 N 共 1⫺exp共 2 ␲ i 共 j⍀⫺ ␻ jm 兲 /N⍀ 兲兴关12兴 Renshaw, A. A., and Mote, Jr., C. D., 1992, ‘‘Absence of One Nodal Diameter
(32) Critical Speed Modes in an Axisymmetric Rotating Disk,’’ ASME J. Appl.
Mech., 59, pp. 687– 688.
Therefore, the response is stable only when
关13兴 Mote, Jr., C. D., 1970, ‘‘Stability of Circular Plates Subjected to Moving
sgn共 ␻ jm 兲 sin共 2 ␲ 共 j⍀⫺ ␻ jm 兲 /N⍀ 兲 ⬍0 (33) Loads,’’ J. Franklin Inst., 290共4兲, pp. 329–344.
关14兴 Renshaw, A. A., and Mote, Jr., C. D., 1996, ‘‘Local Stability of Gyroscopic
The response changes from stable to unstable as the disk rotation Systems Near Vanishing Eigenvalues,’’ ASME J. Appl. Mech., 63, pp. 116 –
speed increases through critical speed (⍀ cr ⫽ ␻ jm / j). Since cir- 120.
关15兴 Tian, J. F., and Hutton, S. G., 2001, ‘‘Cutting-Induced Vibration in Circular
cular saws rarely operate above their critical speed, the more im- Saws,’’ J. Sound Vib., 242共5兲, pp. 907–922.
portant factor is whether or not the response is stable between 0 关16兴 Stone, E., and Askari, A., 2002, ‘‘Nonlinear Models of Chatter in Drilling
⬍⍀⬍⍀ cr . As ⍀→0 ⫹ , we obtain an infinite number of alternat- Processes,’’ Dynamical Systems, 17共1兲, pp. 65– 85.
ing stable and unstable zones of rotation speed for each mode, the
width of each zone shrinking as ⍀→0. This form of response is Rotor Isolation for Vibration and
similar to the one determined by Tian and Hutton 关15兴 using nu-
merical methods. Based on experimental observations, they focus Noise Reduction
their study on the regions at which the tooth passing frequency
approximately equals a natural frequency of the system in the
stationary frame of reference, i.e., Timothy A. Brungart
␻ jm ⫾ j⍀⫽N⍀⫹ ␴ (34)
where ␴ is a detuning parameter. Substitution of Eq. 共34兲 into 共33兲 Eric T. Riggs
shows that ␴⫽0 is always the location of a transition from stable e-mail: [email protected]
to unstable behavior, as shown in their results.
The Pennsylvania State University, Applied Research
3 Concluding Comments Laboratory, PO Box 30, State College, PA 16804-0030
The fundamental simplification derived by using gyroscopic
eigenfunctions is the reduction of order in 共6兲. For this first order
ordinary differential equation, we obtain only one potential secu- A technique for reducing the vibration and noise from marine
lar solution. In the traditional analysis, the decoupled equation is propulsors is examined theoretically and demonstrated experi-
second order and two potential secular solutions must be investi- mentally. A single layer of elastomer is incorporated into a rotor
gated. In practice, one of these solutions is always stable and the to form a rotating vibration isolation mount. The mount isolates
critical stability criteria are determined by the other solution. The the rotor hub and shaft from the unsteady forces generated on the
procedure presented here eliminates the need to investigate the blades due to their interaction with circumferentially non-uniform,
stable secular solution. time-averaged inflow and turbulence. This results in reduced vi-
Positive definite, self-adjoint systems are a special case of posi- bration and noise. Nonrotating isolated rotor transmissibility is
tive definite gyroscopic systems, so the analysis described here well predicted with a simple lumped parameter rotor model for
applies to these systems as well. In fact, because they are per- frequencies below the first bending mode of the blades. Tests per-
formed in the rotating frame of reference, Examples 2 and 3 are formed under operational (i.e., rotating) conditions have shown
self-adjoint. that rotor isolation provides a reduction of up to 15 dB in trans-
The restriction to positive definite systems (M ⬎0, K⬎0) en- missibility compared to an identical rotor without isolation.
sures that all the natural frequencies ( ␻ j ) are real. This makes it 关DOI: 10.1115/1.1569948兴
trivial to determine which terms on the right hand side of 共9兲 are
secular. For systems in which the natural frequencies are not all Contributed by the Technical Committee on Vibration and Sound for publication
real 共e.g., damped systems兲, the analysis would be identical except in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received May 2002;
that more care would be required to determine secular solutions. Revised February 2003. Associate Editor: J. Wickert.
Introduction
Propulsors used in marine applications are a source of tonal and
broadband vibration and noise. Unsteady forces at blade rate and
harmonic frequencies are generated by the interaction of rotor
blades with mean inflow spatial nonuniformities, due to upstream
obstructions, such as control surfaces, stators and support struts.
Broadband unsteady forces are generated by the interaction of
rotor blades with turbulence, formed in the boundary layers and
wakes of upstream surfaces, that is ingested into the rotor. The
tonal and broadband hydrodynamically-induced rotor unsteady
forces are transmitted to the watercraft through the rotor hub and
drive shaft. This results in watercraft vibration and noise. Blake
关1兴 provides an in-depth analysis of these vibration and noise gen- Fig. 1 Schematic of isolated rotor
erating mechanisms.
Vibration and noise control is achieved by reducing the un-
steady forces generated at the source and/or by inhibiting the inder is driven through the base plate with the attached shaker 关3兴.
transmission of the unsteady forces to the watercraft. An example The transmissibility can also be shown, using the transmissibility
of the latter type of control is the use of resilient mounts that theorem, to equal the absolute value of the complex ratio of the
isolate the watercraft from vibrating components such as power force transmitted to the cylinder to the force driving the mass 关3兴.
generating equipment and pumps 关2兴. When power generating This is illustrated in Fig. 3 where the foundation represents the
equipment and pumps are mounted independently, flexible shaft cylinder, M is the mounted mass, V 1 and V 2 are the foundation
couplings are often used to accommodate shaft misalignment and and mounted mass velocities, respectively, and F 1 and F 2 are the
provide isolation of component torsional vibrations 关2兴. Similar forces acting on the foundation and mounted mass, respectively.
isolation techniques can be incorporated directly into the rotor for With a shaker driving the cylinder of the ECTF, the signals
vibration and noise control. from two phase and amplitude matched accelerometers, mounted
on opposite sides of the elastomer, were measured with a dynamic
signal analyzer. The accelerometer transfer functions were mea-
Approach sured over a frequency range of 0 to 2 kHz in 2.5 Hz frequency
A vibration isolation mount incorporated into the rotor hub can bands with 256 ensemble averages. Values of G ␻ and ␦ G ␻ for
attenuate the unsteady forces generated on the rotor blades before mounts fabricated from Hapco Inc. Hapflex 581, 570, 560 and 540
they are transmitted to the shaft and watercraft. Such a mount is elastomers were determined by fitting Eq. 共1兲 to the measured
shown schematically in Fig. 1. A single layer of elastomer sepa- transmissibility. Hapflex 540 was found to be the softest of the
rates the outer blade ring from the inner hub. In the event that the four elastomers considered and, as a result, to provide the greatest
rotor’s axial or torsional load exceeds the mount design specifica- isolation capability. Axial load/deflection curves were measured
tions or the elastomer fails, mating axial and circumferential stops on an Instron Corporation Series IX Automated Materials Testing
are machined into the rotor. The stops are not engaged under
normal operating loads since this would create a shorting path
around the mount.
Quantification of Elastomer Properties. In order to build a
suitable vibration isolation mount, the dynamic mechanical prop-
erties of candidate elastomers must be known, but may not be
adequately provided by product vendors. Elastomer properties are
measured with the Elastomer Characterization Test Fixture
共ECTF兲 shown in Fig. 2. A mass, M, is suspended within a cylin-
der by casting and curing the candidate elastomer into a vibration
isolation mount similar to that utilized in the isolated rotor. A base
plate attaches the cylinder to a shaker. The elastomer dynamic
shear modulus, G ␻ , and loss factor, ␦ G ␻ , are determined by per-
forming a least-squared error fit of the measured mount transmis-
sibility, T, to the lumped parameter expression for transmissibility
given by Snowdon 关3兴
T⫽ 冏
kG ␻
␻
关 ␦ G␻⫺ j 兴
Z
, 冏 (1)
Fig. 2 Schematic of elastomer characterization test fixture
„ECTF…
where Z is the impedance of the mechanical system
Z⫽
kG ␻
␻ G␻
␦ ⫹ j ␻M⫺冋kG ␻
␻
, 册 (2)
and k is a mount stiffness-related parameter

2␲rMᐉ
k⫽ . (3)
r c ⫺r M
Here r is the radius of the suspended mass 共M兲 or cylinder 共C兲,
and ᐉ is the bond length of the elastomer.
The transmissibility is the absolute value of the complex veloc-
ity or acceleration ratio measured across the mount when the cyl- Fig. 3 Vibration isolation mount transmissibility

the ‘‘clearance derivative’’ which is defined as the reduction in
efficiency 共points兲 for a one percent increase in normalized
clearance 关4兴.
Results
Isolated Rotor Static Transmissibility. The axial transmissi-
bility of the mount embedded in the isolated rotor was measured
in both air and water in a manner similar to that employed with
the ECTF. The experimental set-up is shown in Fig. 4. A shaker
was mounted to the inner hub of the rotor and supported with
elastic shock chords from overhead. The outer blade ring is un-
constrained in this set-up, as required by the transmissibility theo-
rem. Two phase and amplitude matched accelerometers were
mounted opposite one-another across the elastomer and oriented
in the direction of the shaker-applied excitation. The measured
Fig. 4 Experimental set-up for static measurement of isolated transmissibility, and transmissibility predicted from Eq. 共1兲, with
rotor transmissibility
G ␻ and ␦ G ␻ measured with the ECTF, are shown in Fig. 5. The
measured transmissibility for a nonisolated 共i.e., solid兲 rotor is 1.0
which is also shown in Fig. 5. The frequency is nondimensional-
ized by the rotor blade passing frequency at the subject operating
System with the Hapflex 540 in the ECTF. The data indicated that condition (1⫻BR).
Hapflex 540 also provided suitable stiffness, therefore, Hapflex The in-air measured transmissibility is within 3 dB of the
540 was used for the construction of the isolated rotor. simple lumped parameter model 共Eq. 共1兲兲 prediction up to a fre-
In order to assess the static stiffness of the mount fabricated quency of approximately 4⫻BR. The lumped parameter model
from Hapflex 540, isolated rotor axial load/deflection curves were tends to under-predict the transmissibility somewhat at higher fre-
also measured on an Instron Corporation Series IX Automated quencies. The transmissibility measured in water is within 3 dB of
Materials Testing System. From the curves, the mount stiffness the Eq. 共1兲 prediction up to a frequency of approximately 1.7
was determined to be approximately 1056 N/mm, 965 N/mm, and ⫻BR. Equation 共1兲 also under-predicts the in-water transmissibil-
863 N/mm at loading rates of 12.7 cm/min, 1.27 cm/min, and ity at higher frequencies but to a greater extent than the in-air
0.127 cm/min, respectively. The mating stops machined into the results. Water loading is responsible for the reduction in mount
rotor are designed to engage when the thrust moves the blade row resonance frequency and corresponding modifications to the trans-
forward 共approximately 0.32 cm兲 at speeds just above the maxi- missibility for the in-water measurements compared to the in-air
mum speed of operation. Adequate clearance 共hub-to-blade tip兲 is measurements. It is reasonable to expect that the measured trans-
available to permit such movement. Rotor isolation is not ex- missibility will deviate from the lumped parameter-based predic-
pected to affect the powering performance of the propulsor since tion at frequencies where the rotor begins to exhibit distributed
the blade tip clearance does not change significantly throughout system behavior. Structural modal analyses of the rotor, performed
the rotor’s axial range of motion. An increase in blade tip clear- in-air and in-water, indicate that the lowest order mode, the first
ance has an adverse effect on propulsor efficiency 关4兴. The extent bending mode, occurs at frequencies of approximately 6.5⫻BR in
of the performance degradation has been quantified in terms of air and 2.8⫻BR in water. Thus, we attribute the deviations be-
Fig. 5 Comparison of statically-measured and predicted of isolated rotor

transmissibility

Fig. 6 UUV model unsteady forces test set-up
Fig. 7 Typical shaft unsteady thrust spectrum measured with and without
rotor isolation
Fig. 8 Comparison of operationally-measured and predicted isolated rotor

transmissibility

tween the measured and predicted transmissibility, at frequencies ⫻BR, the Eq. 共1兲 predictions are well below the statically and
above approximately 4⫻BR in air and 1.7⫻BR in water, to the operationally measured transmissibility since the lumped param-
occurrence of the rotor blade bending modes. eter assumption inherent to Eq. 共1兲 is violated. Resonances asso-
ciated with the dynamometer and shaft assembly add to the diffi-
Isolated Rotor Transmissibility Under Operational Condi-
culty of interpreting comparisons of the measured static
tions. The hydrodynamically-induced rotor unsteady forces
transmissibility to the transmissibility measured under operational
transmitted to the shaft were measured on an Unmanned Undersea
conditions. Even so, the Fig. 8 transmissibility comparisons show
Vehicle 共UUV兲 model. Lauchle et al. 关5兴 give a detailed descrip-
that the isolated rotor attenuates the unsteady forces transmitted to
tion of the 1.22 m Water Tunnel facility in which the measure-
the shaft up to approximately 15 dB in level compared to an
ments were performed. A schematic of the test set-up is shown in
identical non-isolated rotor.
Fig. 6.
The rotor was housed in a duct with a downstream stator and
was driven with a shaft attached to two 75 Hp electric motors in a Conclusions
downstream-mounted dynamometer. An unsteady force sensor A vibration isolation mount has been incorporated into the hub
was mounted in the shaft and used to measure the rotor-generated of a rotor used to propel a marine vehicle. The mount transmissi-
unsteady axial force 共thrust兲. Unsteady thrust spectra were ac- bility is predicted reasonably well under static conditions with a
quired with a dynamic signal analyzer from 0 to 1 kHz in fre- simple lumped parameter transmissibility model at frequencies
quency with 0.3125 Hz resolution. The spectra were averaged below the first bending mode of the rotor blades. Under opera-
over 256 ensembles. Typical results are shown in Fig. 7. The tional conditions, the mount resonance is eliminated but otherwise
isolation reduces the shaft unsteady thrust measured at the 1 the measured transmissibility is comparable to the static measure-
⫻BR and harmonic frequencies up to 10 dB in level, and the ments. The isolated rotor reduces the unsteady forces transmitted
measured broadband shaft unsteady thrust up to 15 dB in level. to the shaft up to 15 dB in level compared to an identical non-
The transmissibility measured under operational conditions is isolated rotor.
shown in Fig. 8 and compared to the transmissibility measured
statically in water and to the Eq. 共1兲 prediction. The transmissibil- Acknowledgment
ity increase measured statically in water at the mount’s resonance
frequency (0.6⫻BR) is reduced approximately 15 dB under op- This work was sponsored by the Office of Naval Research,
erational conditions. In-water static transmissibility measurements Code 333, and monitored by Dr. Kam Ng. Their support is grate-
performed with and without the duct were identical. This result is fully acknowledged. The contributions of Dr. Jason T. Gomez and
consistent with calculations that indicate the duct does not affect Mr. Stephen J. Plunkett, of the Naval Undersea Warfare Center
the transmissibility since its impedance is negligible compared to Division Newport, in various aspects of this work, are acknowl-
the impedance of the mechanical system. The 15 dB transmissi- edged as well.
bility reduction at the mount’s resonance frequency, measured un-
der operational conditions, appears to be associated with the dy- References
namics of the rotating dynamometer and shaft assembly since the 关1兴 Blake, W. K., 1986, Mechanics of Flow-Induced Sound and Vibration, Vol-
resonance amplification is approximately 15 dB when the isolated umes I and II, Academic Press, Inc., Orlando.
rotor is mechanically excited on the nonrotating dynamometer 关2兴 Skaistis, S., 1988, Noise Control of Hydraulic Machinery, Marcel Dekker,
shaft. We speculate that the rotation of the shaft and dynamometer New York.
关3兴 Snowdon, J. C., 1968, Vibration and Shock in Damped Mechanical Systems,
assembly alters the phase relationship between the excitation force John Wiley and Sons, New York.
and the axial motion of the rotor at resonance resulting in reduced 关4兴 Wisler, D. C., 1986, ‘‘Advanced Compressor and Fan Systems,’’ Fluid Dynam-
transmissibility. ics of Turbomachinery, Gas Turbine Division, American Society of Mechani-
At frequencies below approximately 0.4⫻BR and between cal Engineers, ASME Turbomachinery Institute, Ames, Iowa.
关5兴 Lauchle, G. C., Billet, M. L., and Deutsch, S., 1989, ‘‘High Reynolds Number
0.9⫻BR and 1.9⫻BR, both the Eq. 共1兲 prediction and measured Liquid Flow Measurements,’’ Lecture Notes in Engineering, 46, Frontiers in
static transmissibility are within ⫾3 dB of the transmissibility Experimental Fluid Mechanics, M. Gad-el-Hak, ed., Springer-Verlag, Berlin,
measured under operational conditions. At frequencies above 1.9 pp. 95–158.

Asme - Journal of Vibration and Acoustics - July 2003

Uploaded by

Copyright:

Available Formats

Asme - Journal of Vibration and Acoustics - July 2003

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Asme - Journal of Vibration and Acoustics - July 2003

Uploaded by

Copyright:

Available Formats

Transactions Journal of Vibration

of the ASME and Acoustics ®

Journal of Vibration and Acoustics Volume 125, Number 3 JULY 2003

The ASME Journal of Vibration and Acoustics is abstracted and

Since the enthalpy must remain limited, the constant C 1 has to

250 Õ Vol. 125, JULY 2003 Transactions of the ASME

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 251

1 Introduction Assume the lug shape can be described by

2.1 Theory on Tire Noise. Tire pitch noise is approximated

冕 h 共 t 兲 ␦ 共 t⫺t i 兲 dt⫽h 共 t i 兲 . (2)

Fig. 2 Magic shape

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 253

P rection described by w(x), noise amplitude is proportional to

254 Õ Vol. 125, JULY 2003 Transactions of the ASME

Fig. 5 Verification of magic angle „measurement…

n is the number of harmonics, A n is the parameter for A-weighted

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 255

256 Õ Vol. 125, JULY 2003 Transactions of the ASME

Introduction frequencies on measured noise transmission. These prior studies

Acoustic Transmission Loss

where m, n, o, p, q are modal indices and rជ is a position vector.

258 Õ Vol. 125, JULY 2003 Transactions of the ASME

phone measurement points were achieved with a boom that was

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 259

Fig. 4 Details of FEA model components

260 Õ Vol. 125, JULY 2003 Transactions of the ASME

Table 2 Numerical and experimental structural modal properties.

Predicted Measured Measured

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 261

262 Õ Vol. 125, JULY 2003 Transactions of the ASME

Fig. 8 Representative FRF between speaker and cavity microphone

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 263

Predicted Measured Measured

Fig. 9 Noise transmission measurement setup

264 Õ Vol. 125, JULY 2003 Transactions of the ASME

Fig. 11 Measured noise reduction spectrum „NRS… showing various

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 265

266 Õ Vol. 125, JULY 2003 Transactions of the ASME

1 Introduction introduces the modal analysis approach to calculating the fre-

Denote ⌿̄ sT f s ⫽p, the coefficient matrix ␰ can be computed in the

3 Sensitivity Analysis of Coupled Systems

Suppose that the forced frequency responses are expanded by

268 Õ Vol. 125, JULY 2003 Transactions of the ASME

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 269

Structural Acoustic Coupled

5.2 Case 2. In the second case, the unit harmonic excitation

270 Õ Vol. 125, JULY 2003 Transactions of the ASME

Mode Structural frequency Coupled frequency

Table 4 Trend of SPL reduction in case 2

Table 5 Natural frequencies of the final design in case 2

Mode Structural frequency Coupled frequency

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 271

272 Õ Vol. 125, JULY 2003 Transactions of the ASME

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 273

Introduction Test Gear and Testing Machine

Contributed by the Technical Committee on Vibration and Sound for publication

Where, the ␺ (t) indicates the conjugate of a mother wavelet func-

Journal of Vibration and Acoustics JULY 2003, Vol. 125 Õ 275

Fig. 4 Gabor function f 共 j 兲共 t 兲 ⫽ f j⫺1 共 t 兲 ⫹g j⫺1 共 t 兲 ⫽g j⫺1 共 t 兲 ⫹g j⫺2 共 t 兲 ⫹g j⫺3 共 t 兲 ⫹•••

k 共 x 兲 ⫽K n 兺关 g⫹x cos ␩ ⫺ 共 A/Bn 兲 sin ␩ 兴