PSD Random Vibration Tutorial For Femap and NX Nastran 2

PSD Random Vibration Analysis in FEMAP
An Introduction to the Hows and Whys of PSD

Adrian Jensen, PE | Senior Staff Mechanical Engineer
Kyle Hamilton | Staff Mechanical Engineer
Random Vibration Analysis 2017
Table of Contents
1. INTRODUCTION ............................................................................................................................................................ 4
2. THE PSD FUNCTION ...................................................................................................................................................... 5
3. THE NX NASTRAN METHOD ......................................................................................................................................... 7
4. PSD UNITS .................................................................................................................................................................... 8
5. EXAMPLE 1: CANTILEVER BEAM ................................................................................................................................... 9
5.1 PROBLEM DEFINITION ........................................................................................................................................................................................9
5.2 ANALYTICAL SOLUTION .....................................................................................................................................................................................10
5.3 DEFINING THE SYSTEM DAMPING .......................................................................................................................................................................11
5.4 CREATING THE PSD FUNCTION ..........................................................................................................................................................................12
5.5 CREATING THE MODAL FREQUENCY TABLE/SETTING UP THE LOAD SET OPTIONS FOR DYNAMIC ANALYSIS ........................................................................13
5.6 LOADING & CONSTRAINING THE MODEL ..............................................................................................................................................................15
5.7 SPECIFYING GROUPS FOR NODAL AND ELEMENTAL OUTPUT ....................................................................................................................................16
5.8 CREATING AN ANALYSIS SET – SIMPLE PSD ..........................................................................................................................................................17
5.9 INTERPRETING THE OUTPUT...............................................................................................................................................................................24
5.10 POSITIVE CROSSINGS ...................................................................................................................................................................................25
5.11 FATIGUE ANALYSIS USING RMS STRESS AND POSITIVE CROSSINGS .......................................................................................................................27
5.12 FATIGUE ANALYSIS – TIME TO FAILURE ...........................................................................................................................................................28
6. EXAMPLE 2: SOLID MESHED BEAM ............................................................................................................................ 29
6.1 ANALYTICAL SOLUTION .....................................................................................................................................................................................30
6.2 PSD FUNCTION INPUT......................................................................................................................................................................................31
6.3 PSD STRESS RESULTS .......................................................................................................................................................................................32
6.4 COMPARING MILES’ APPROXIMATION AND PSD RESULTS ........................................................................................................................................33
Applied CAx / Predictive Engineering White Paper
Page 2 of 45
Please share with your friends and visit us online
7. CONCLUSION.............................................................................................................................................................. 34
8. RECOMMENDED READINGS ....................................................................................................................................... 34
9. APPENDIX................................................................................................................................................................... 35
9.1 FLOW CHART FROM NX NASTRAN THEORETICAL MANUAL ......................................................................................................................................35
9.2 CREATING MODAL FREQUENCY TABLE .................................................................................................................................................................36
9.3 AUTOCORRELATION FUNCTION...........................................................................................................................................................................39
9.4 MULTIPLE EXCITATION SPECTRUMS.....................................................................................................................................................................42
9.5 WHY WE DO A PSD ANALYSIS ............................................................................................................................................................................45

Page 3 of 45
1. INTRODUCTION
Random vibration is vibration which can only be
described in a statistical sense. The magnitude at any
given moment is not known, but is instead described in
a statistical sense via mean values and standard
deviations.
Random vibration problems arise due to earthquakes,

tsunamis, acoustic excitation (e.g., rocket launches),
wind fluctuations, or any loading which is inherently
random. Often random noise due to operating or
transporting conditions can also be considered. These
random vibrations are usually described in terms of a
power spectral density (PSD) function.

Page 4 of 45
2. THE PSD FUNCTION

Random vibration is unique because it can excite all frequencies at once, whereas a sine sweep will excite one frequency
at a time (think slamming all keys on a piano instead of sliding your hand across them). The PSD function is created by
subjecting a structure to white noise vibration and measuring the RMS amplitude of the response of the structure across
a range of frequencies, squaring the response, and dividing it by the frequency range which results in units of G2/Hz.
A typical power spectral density is shown below:
Frequency PSD
(Hz) (G2/Hz)
20 0.01
100 0.05
900 0.05
1,100 0.01
For more of the nitty gritty math details see NASA’s webpage on random vibration here: https://femci.gsfc.nasa.gov/random/

Page 5 of 45
A system subject to random vibration does not have a single resultant stress. Luckily for us, the stress results do typically
follow a Gaussian distribution (think bell-curve):
The Gaussian distribution allows stress results to be reported statistically. FEMAP will generate 1-σ stresses, which
represent the stress that the system will likely see 68% of the time. The 2-σ stress level covers 95% of cases, and 3-σ
covers 99.7%. Most of the time a system is designed to the 3-σ stress level.
Image By Dan Kernler (Own work) [CC BY-SA 4.0 (http://creativecommons.org/licenses/by-sa/4.0)], via Wikimedia Commons

Page 6 of 45
3. THE NX NASTRAN METHOD

Given an input PSD function, an output response can be calculated by using the systems transfer function.
𝑃𝑆𝐷𝑜𝑢𝑡 = |𝑔(𝑤)|2 𝑃𝑆𝐷𝑖𝑛
The g(w) represents the system transfer function. A systems transfer function simply represents its output to input
ratio. NX Nastran performs a frequency response analysis on the system to obtain the system transfer function, and then
does the random vibration analysis as a post processing step based upon this transfer function.
There are several steps to setting up the analysis in FEMAP:

1. Defining the system damping
2. Creating the PSD function
3. Creating a Modal Frequency Table (or Requested Solutions Function)
4. Creating the excitation node and tying it into the model
5. Loading the model
6. Constraining the model
7. Specifying output groups for nodal and elemental output
8. Setting up the analysis in the Analysis Manager

Page 7 of 45
4. PSD UNITS
It can be tricky keeping track of the units in any analysis; this is especially true with PSD analysis. The table below shows
the input and output units for a few of the most common unit systems. It doesn’t matter which system you go with but
be sure you are consistent throughout your analysis. The table below shows the properties for aluminum in each unit
system.
Inputs RMS Outputs

Units Young’s PSD Acceleration
Mass Density Deflection Stress
Modulus Function Load
SI (m,kg,sec) 6.89e10 Pa 2710 kg/m3 1 g2/Hz 9.807 m/s2 1m 1000 Pa
SI (mm,Mg,sec) 6.89e4 MPa 2.71e-9 Mg/mm3 1 g2/Hz 9807 mm/s2 1000 mm 1.0e-3MPa
Imperial (in, snail, sec) 10.0 e6 psi 2.54e-4 snail/in3 1 g2/Hz 386.1 in/s2 39.37 in 0.145 psi

Page 8 of 45
5. EXAMPLE 1: CANTILEVER BEAM

5.1 PROBLEM DEFINITION
A cantilevered aluminum beam 5 inches in
length is used to support a 0.50 lb mass. Our E = 10.0E6 psi
objective is to determine the dynamic
stresses and fatigue life of the beam for
vibration along the vertical axis. 2.0 in
The FEA model is a single beam element. A
picture of the beam element, with its cross
section displayed is shown on the right.
We will compare the FEA results to an
analytical solutionψ. The PSD input (PSDIN)
function used by Steinberg was
𝑃𝑆𝐷𝑖𝑛 = 0.2 𝐺 2 ⁄𝐻𝑧
This excitation was applied to the fixed end

of the beam (where the rectangle is drawn).
0.25 in
Our unit system is lb/in/s and 1 g = 386 in/s2. 0.50 lb Point Mass
Ψ
Steinberg, Dave S. Vibration Analysis for Electronic Equipment. 2nd ed. New York: John Wiley & Sons, 1988. 226-231.

Page 9 of 45
5.2 ANALYTICAL SOLUTION

A cantilever beam with the dimensions previously given and an end load of 0.5 lbf experiences an end deflection of:
𝑊𝐿3
𝑌𝑆𝑡 = = 8.01𝐸 − 4
3𝐸𝐼
Based upon this end deflection, the beam’s resonant frequency and transmissibility can be calculated as:
1 𝑔
𝑓𝑛 = √ = 110.5 𝑄 = 2√𝑓𝑛 = 21
2𝜋 𝑌𝑆𝑡
Miles’ equation can be used to approximate the Gout(RMS) value:

𝜋
𝐺𝑜𝑢𝑡 = √ 𝑃𝑆𝐷𝑖𝑛 ∗ 𝑓𝑛 ∗ 𝑄 = 27.0
2
This output in G. If an equivalent value is desired in English units, simply multiply this by gravity
𝑎𝑐𝑐𝑒𝑙𝑒𝑟𝑎𝑡𝑖𝑜𝑛
27𝐺 = 27 ∗ 𝑔𝑟𝑎𝑣𝑖𝑡𝑦 = 10,422 𝑖𝑛⁄ 2
𝑔𝑟𝑎𝑣𝑖𝑡𝑦 𝑠
The max output PSD can also be obtained using:

𝑃𝑆𝐷𝑜𝑢𝑡 = 𝑄2 ∗ 𝑃𝑆𝐷𝑖𝑛 = 212 ∗ (0.2 ∗ 𝐺 2 ) 𝑤ℎ𝑒𝑟𝑒 𝐺 = 1𝑔 𝑜𝑟 386 𝑖𝑛/𝑠 2
In English units, the max PSDout = 13.14e6 in2/s4. This can also be verified against the FE Model.
Note: When approximating transmissibility (Q), the square root of the natural frequency should be scaled by 0.5 – 2 per Steinberg. In this case there is 100%
mass participation for the first mode, thus 2 is appropriate.

Page 10 of 45
5.3 DEFINING THE SYSTEM DAMPING

Determining how the system is damped can be
complicated. In NX Nastran there are three
ways to do this:
1. If the structural damping
coefficient (G) is known then
function type 6: Structural
Damping vs. Frequency should be
used,
2. If the critical damping ratio is
known, then function type 7:
“Critical Damping vs. Frequency”
should be used,
3. If the Quality/Magnification factor
(Q) is known, then function type 8:
“Q Damping vs. Frequency” should
be used.
An approximation of the transmissibility of the beam is Q = 21. This value yields a critical damping ratio of 2.38%; this is
what we will use.

Page 11 of 45
5.4 CREATING THE PSD FUNCTION

The input to the cantilever beam is a white-
noise vibration with a PSD input of 0.20
G2/Hz from 20 to 2000 Hz.
This is entered directly with no scaling. It
will be scaled for the desired unit system in
the Load Definition dialog (Section 5.6).

Page 12 of 45
5.5 CREATING THE MODAL FREQUENCY TABLE/SETTING UP THE LOAD SET OPTIONS FOR DYNAMIC ANALYSIS
The Modal Frequency Table is a function which defines
which frequencies NX Nastran will obtain a solution
for; that is, each frequency represents a separate
solution that is written out to the results file. The
function can either be created manually, or FEMAP can
create one for you. If you do not know about which
frequencies you’d like the analysis to focus, it is
preferable to have FEMAP set it up, otherwise you will
most likely end up with a large amount of extraneous
output.
To have FEMAP set up the table for you, you must first run an eigenvalue analysis. Once the eigenvalue
analysis has run, FEMAP will know about which frequencies to concentrate.
The normal modes will be used to define the solution frequencies of the Random Analysis. Think of it as
guiding the Random Analysis such that only frequencies of interest (significant frequencies) are processed.
This greatly limits the amount of post-processing that is required for the Random Analysis. More will be
said on this later on….

Page 13 of 45
It is good practice to run the normal modes

analysis first to see how the structure will
behave. In this simple beam model, we have
fixed the end of the beam in all six DOF. The
beam is also massless (material density of 0.0).
This was done to allow us to exactly match the
analytical solution.
After the analysis has finished running, you

should have three modes. In Section 5.8 we will
show you how these Normal Modes are used to
generate the Solution Frequencies for the
Random Analysis.

Page 14 of 45
5.6 LOADING & CONSTRAINING THE MODEL

An acceleration load must be given to the base
node in the direction of the excitation. Since the
PSD is given in G2/Hz, we must scale the load by a 1
g gravitational acceleration in our unit system of
choice. We want our deflection results in inches so
we will enter an acceleration of 386 in/s2.
The Load Constraint constrains the base node in all

six degrees of freedom.
This constraint set should identical to the constraint
set used for the eigenvalue analysis. The node used
to constrain the model is the same node to which
the unit acceleration was applied.
The idealization concept is that the base is fixed in
the TX, TZ, RX, RY, RZ while the structure is excited
in the Y-direction (i.e., there is displacement in the
Y-direction).

Page 15 of 45
5.7 SPECIFYING GROUPS FOR NODAL AND ELEMENTAL OUTPUT

A group can be created to specify certain
nodes and elements to recover data from. For
this analysis we will skip creating a group to
simplify the analysis.
If we wanted a group for the beam element

we could create a single group with our single
element and two nodes. We are not
recovering any data from the Mass Element,
so we can leave it out of the group.

Page 16 of 45
5.8 CREATING AN ANALYSIS SET – SIMPLE PSD

Next up is creating an analysis set. There are a lot of
options to tailor the output to exactly what you need, but
let’s look at a straightforward analysis. This will allow you
to see RMS Stress and positive crossings, which is enough
information for a general PSD stress analysis and fatigue
life estimate.
First, create a new Random Response Analysis Set.
Select Next…

Page 17 of 45
Keep pressing Next… until you arrive at the NASTRAN

Modal Analysis window.
In the modal analysis tab you can decide between a Direct
or Modal Solution Type. For this analysis, we will use a
Modal solution. For more information about the difference
in solution types take a look at the NX Nastran Basic
Dynamic Analysis User Guide, Chapter 6.4 Modal Versus
Direct Frequency Response.
For Range of Interest you can set the maximum frequency
at your upper limit of the PSD spectrum. This will guarantee
your entire PSD spectrum is covered and not spend extra
computing power (and time) processing frequencies above
that.

Page 18 of 45
In the Dynamic Analysis tab, one can specify the damping

function and define the frequency range of the analysis (#
of modes, or Lowest and Highest Frequency).
For Frequency Response, Select the “Modal Freq…” button,

and then choose the modes you would like to create a
Modal Frequency Table from. For this analysis only the first
mode will be selected to match up with the analytical
solution.
It is recommended to use the default values for the Points
per Mode and Frequency Spread. See appendix for details.

Page 19 of 45
In the NASTRAN XY Output for Modal Analysis window, you

can leave all of the options un-checked. This information
can be gathered when you run a standard modal analysis
so there is no need to request it here.
In the NASTRAN Output for Random Analysis window,

select none for nodal and elemental output.
PSD Functions: Generates ‘PSDF’ output set for each frequency in the
Modal Frequency Table
Autocorrelation Functions: Creates output for the autocorrelation
functions if applicable
Root Mean Square: Generates ‘CRMS’ results for each frequency in
the Modal Frequency Table

Page 20 of 45
If you are interested in getting data for your entire

structure, deselect everything in the NASTRAN Output for
Random Analysis window. This will give you 1-σ stress
results for your full model. For this example deselect all.
If you have an extremely large model and you only want
specific nodal outputs, or results from certain elements,
this is where you specify that. You can also use this window
to request specific data such as T2 acceleration for a group
of elements and nodes that you could have created in
Section 5.7. If you select PSDF it will generate a function
with the acceleration vs frequency for a group.

Page 21 of 45
Select your PSD Function and be sure to select Apply. If desired Choose your constraint set and load created for the
you can scale the PSD function in the “Factor” input here. PSD analysis

Page 22 of 45
Choose the output requests desired. For this analysis we

will request Displacements, Equation Force, Acceleration,
and Stress.

Page 23 of 45
5.9 INTERPRETING THE OUTPUT

The PSD output sets are titled RMS Values and Positive In the RMS Values output set you can contour all the usual
Crossings. RMS Values will give all of the traditional output vectors. Beam EndA Pt1 Comb Stress is shown
stress, displacement, and acceleration data. Positive contoured over the beam. This output shown is the RMS
Crossings will output the frequency of positive crossings Stress, and is also known as the 1-σ PSD stress value. This
for each of the requested output vectors. This represents how much stress the beam will experience 68.3% of
frequency is utilized to calculate fatigue damage based the time.
on the duration of excitation.

Page 24 of 45
5.10 POSITIVE CROSSINGS

This is a vibration analysis, so of course we are also
concerned about fatigue. We will use the output from
positive crossings to calculate the fatigue life.
To access data for the positive crossings, Right click on the
Positive Crossings result in the model info tree, and select
“Post Data”
In the Post Data toolbar select the Dynamic Max/Min box

in the upper right
Select the output vector for the positive crossing
frequencies desired. In this model, all stress recovery
points on Beam EndA show the same frequency.

Page 25 of 45
Positive Crossings can also be contoured over the model.

This can help the user understand how the positive crossing
frequency changes throughout the model.

Page 26 of 45
5.11 FATIGUE ANALYSIS USING RMS STRESS AND POSITIVE CROSSINGS

We can see that Beam EndA Pt1 Comb Stress vector gives a positive
crossing frequency of 110.4 Hz. This means that given the white noise PSD
input of 0.2 G2/Hz, the beam will experience a fully reversible stress of
3,162 psi at a frequency of 110.4 Hz.
Statistically speaking, this stress value represents the 1-σ value and will be
experienced 68.3% of the time. A 2-σ stress of 2*3,162 or 6,324 psi will be
experienced 27.1% of the time and a 3-σ value of 9,486 psi will be n1 n2 n3
experienced 4.33% of the time. These values represent 99.73% of the Rn   
stresses the beam will see at point A. It is probable that the beam will see N1 N 2 N3
stresses at and above the 4σ level, but this will only happen 0.27% of the
time, so we will ignore them.
All three σ level stresses fall into the “run-out” range on a fatigue curve for
aluminum. To demonstrate how to treat the problem if this is not the
case, let us assume that there is a small hole in the beam which causes a
stress concentration factor of 3. This would put the 1-σ stress level at
9,486 psi. We can use Miner’s cumulative damage index to get a sense of
how long the beam will last under this condition. Miner’s cumulative
damage is given by the equation on the right.

Page 27 of 45
5.12 FATIGUE ANALYSIS – TIME TO FAILURE

On the right is a table containing values taken from a
fatigue curve for aluminum. For a given stress, the
amount of cycles needed to cause failure is given.
These values can be substituted into Miner’s

equation to calculate how many cycles can occur
until the beam fails. Substituting in the values and
solving for n, yields a beam life of 1.80E6 cycles. If
the beam is vibrating at a frequency (number of
positive crossings) of 110.4 Hz, then it will take the
beam approximately 16,300 seconds or about 4.5
hours to fail.
Point A 1σ 2σ 3σ
As long as the beam is exposed to the while noise
vibration for less than 4.5 hours, it should not fail. Stress 9,486 psi 18,972 psi 28,458 psi
# of Cycles
infinite 11.0E6 cycles 14.0E4 cycles
to Fail
0.6831 n 0.271 n 0.0433 n

1  
 11.0E5 14.0E4

Page 28 of 45
6. EXAMPLE 2: SOLID MESHED BEAM

Let’s take a look at the same beam geometry modeled with The beam properties are shown below:
solid elements. The beam is massless, with a point mass of w = 2 in
0.5lbf (1.30e-3 snails) attached via RBE2 on the end.
T = 0.25 in
Lbeam = 5 in
W = 0.5 lbf
E = 10e6 psi

Page 29 of 45
6.1 ANALYTICAL SOLUTION

Let’s first take a look at the hand calculations to show how the beam is expected to behave.
First up is maximum deflection Ymax
𝑊𝐿3
𝑌𝑚𝑎𝑥 = = 8𝑒 − 4 𝑖𝑛
3𝐸𝐼
Based upon this end deflection, the beam’s first natural frequency and transmissibility can be calculated as:
1 𝑔
𝑓𝑛 = √ ( ) = 110.6 𝐻𝑧 𝑄 = 2√𝑓𝑛 ≈ 21
2𝜋 𝑌𝑚𝑎𝑥
Utilizing Miles’ Equation to estimate Grms we see that Grms is approximately 27 Gs:
𝜋
𝐺𝑜𝑢𝑡𝑅𝑀𝑆 = √ 𝑃𝑆𝐷𝑖𝑛 𝑓𝑛 𝑄 = 27𝐺 ′ 𝑠
2

Page 30 of 45
6.2 PSD FUNCTION INPUT

Then we generate the functions necessary for the PSD Analysis. Note the Modal Frequency Table is centered at the first
natural frequency with 10% spread in both directions.
Damping Function
PSD Spectrum Modal Frequency Table

Page 31 of 45
6.3 PSD STRESS RESULTS

After running the analysis, let’s take a look at the results. The PSD results can be validated by checking the resultant
acceleration against the Miles’ equation prediction. Miles’ equation predicted 27 G’s for the maximum acceleration. The
results show an acceleration of 10,300 in/s2 which matches up with the Miles’ equation prediction.
𝑖𝑛
10,300
𝑠 2 = 27 𝐺𝑠
𝑖𝑛
386.09 2
𝑠

Page 32 of 45
6.4 COMPARING MILES’ APPROXIMATION AND PSD RESULTS

An additional verification is done by comparing the PSD stress results to the static analysis with the acceleration given by
Miles’ equation. The images below show an 8% difference between the two results, with similar stress patterns. In
addition, the hand calculations show ~10% higher stresses than the static analysis.
Hand Calculations:
𝐹𝑑 = 27 ∗ 𝑊 ∗ 𝑆𝑎 = 13.5 𝑙𝑏𝑓
𝑇
𝑀𝑐 (𝐹𝑑 𝐿)( )
𝑆𝑡𝑟𝑒𝑠𝑠 = = 2 = 3,240 𝑝𝑠𝑖
𝐼 𝐼
This comparison between the PSD results, Miles’ equation, and hand calculations offer some insight into the relative
accuracy of the analysis.

Page 33 of 45
7. CONCLUSION
The topic of Random Vibration is complex. What is presented here is a brief introduction to the theory and
implementation of the subject. It is suggested that the user read a bit of the documentation provided on this subject
within the NX Nastran library that is installed with every license of FEMAP & NX Nastran.
For a lot of FEA work, a straightforward recipe to accomplish your analysis task is seldom available and if it does, could
easily lead you down the wrong path. Thus, I’m fond of saying that nothing beats having a good theoretical
understanding of what you are doing and being highly suspicious of any result generated in “color”. Or as I have read
“Computer models are to be used but not necessarily believed.”
8. RECOMMENDED READINGS
 Linear Dynamics for Everyone, Three Part Series on Vibration (www.PredictiveEngineering.com)
 Principles of Vibration Analysis: Normal Modes to PSD to Direct Transient (www.AppliedCAx.com)
 Vibration Analysis for Electronic Equipment by Dave S. Steinberg
 NX Nastran Basic Dynamic Analysis User's Guide

Page 34 of 45
9. APPENDIX
9.1 FLOW CHART FROM NX NASTRAN THEORETICAL MANUAL

Page 35 of 45
9.2 CREATING MODAL FREQUENCY TABLE

When Nastran calculates the RMS stress value for a PSD analysis it first calculates the frequency response at each value
on the Modal Frequency Table, and uses that response to calculate the stress due to PSD excitation.
The chart below shows the frequency response for a simply supported beam with the first mode at 227 Hz, and the next
mode above 300 Hz. The response acceleration ramps up near the natural frequency and gradually drops off as you move
away from it.

Page 36 of 45
PSD analysis is statistical and the 1-sigma stress output is simply the stress the structure will likely see when subject to a
specified acceleration spectrum. The random vibration solver doesn’t calculate the stress at every frequency—it only
solves for the stress at the frequencies specified in the Modal Frequency Table. The simplified process is that it solves for
the stress at each value in the Modal Frequency Table, and then combines those stress results to give the RMS stress.
The red line in the image below shows the response at the natural frequency. The orange lines show the response at
multiple points with a 10% spread from the natural frequency. As you can see, the response drops off as you move away
from the natural frequency so adding more solve points, or a greater spread from the natural frequency does not
improve accuracy of the results, but it does add significant computational cost.

Page 37 of 45
The chart below shows Modal Frequency Tables (X-axis is frequency, Y-axis is arbitrary) with varying number of solution
points and the resultant RMS stress on a simply supported beam. With a single solve point at the natural frequency it
significantly overestimates the RMS stress. The default 5 points with 10% spread gives a more reasonable result, and you
can see even going up to an impractical 55 points with 20% spread gives a result within a few percent of the default table
configuration.
1-σ RMS Stress, 1 Point at fn 1-σ RMS Stress, 5 Points with 10% Spread 1-σ RMS Stress, 55 Points with 20% Spread

Page 38 of 45
9.3 AUTOCORRELATION FUNCTION

A newly supported feature in FEMAP 11.3 is the ability to generate Autocorrelation functions during random vibration
analysis. The Autocorrelation Function Time Lag input is available in the NASTRAN Power Spectral Density Factors dialog
as shown below:
The three inputs available are Lag Intervals, Starting

Lag, and Max Lag. These inputs do not change the
way the random vibration analysis is conducted, it
merely defines the autocorrelation function which
will be generated in addition to the output.
Lag Intervals: How many times to chop up the time

band between the starting lag and maximum lag
Starting Lag: Starting time step
Max Lag: Final time step

Page 39 of 45
If we take our beam example and plot the autocorrelation function for displacement on a couple of nodes we can get a
more intuitive idea of what is going on. No matter the lag time, the autocorrelation at the excitation node is very close to
zero. At lag = 0, the autocorrelation for the node at the end of the beam is high while the node at the middle of the beam
is lower amplitude, and follows the same sinusoidal pattern. From this plot we can infer that a small lag time results in a
high autocorrelation at the beam end, and it tapers off as you increase the lag time. It is worth noting that the period of
the sinusoidal response shown here is 0.009 seconds, which matches the 110 Hz natural frequency of the beam.
Excitation Node
Middle Node
Node at Free End

Page 40 of 45
At this point you may be wondering how to generate the autocorrelation function in your analysis. In the first NASTRAN
Output dialogs, select “Autocorrelation Functions” or “All”. Choose the desired nodal or elemental outputs to plot, and
then enter the desired Autocorrelation Function Time Lag values.

Page 41 of 45
9.4 MULTIPLE EXCITATION SPECTRUMS

In some cases you may need to analyze a system with multiple excitation spectrums. This can be due to a difference in
structure mounting points, wheels driving on different surfaces, or an array of other situations. In this example we will
take a look at a hypothetical red wagon, where the wheels on the –Z side are on a smooth section of road, and the -Z side
is rolling on over a rougher road.
First the boundary conditions must be set up as shown—with your excitation points defined in separate load sets and all
of the constraints in one Constraint Set.

Page 42 of 45
With the boundary conditions set up, prepare the random analysis in the usual manner, but do not specify boundary
conditions. Instead, specify the boundary conditions as subcases in the Analysis Set Manager.

Page 43 of 45
After that is set up, go back to edit the PSD Factors tab. Here you will notice it looks a little different than the previous
method. You can now choose PSD Functions for each subcase, and you can correlate the two sub cases for coupled
analysis if desired. If you do not wish to correlate the sub cases, leave the settings at their default values.

Page 44 of 45
9.5 WHY WE DO A PSD ANALYSIS

Dynamics are tricky. Structures that seem sturdy intuitively may have unexpected responses when excited dynamically.
The images below show a circuit board which was subjected to an intense PSD spectrum where the assembly was
expected to see accelerations near 700 g’s! In this case, the circuit board was not designed for that level of excitation and
the chips removed themselves from the board during testing.

Page 45 of 45

PSD Random Vibration Tutorial For Femap and NX Nastran 2

Uploaded by

Copyright:

Available Formats

PSD Random Vibration Tutorial For Femap and NX Nastran 2

Uploaded by

Document Information

Original Description:

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

PSD Random Vibration Tutorial For Femap and NX Nastran 2

Uploaded by

Copyright:

Available Formats

What is random vibration analysis?

What is random vibration analysis?

How is a PSD function defined and used?

How is a PSD function defined and used?

PSD Random Vibration Analysis in FEMAP

An Introduction to the Hows and Whys of PSD

Applied CAx / Predictive Engineering White Paper

Random vibration problems arise due to earthquakes,

Applied CAx / Predictive Engineering White Paper

2. THE PSD FUNCTION

A typical power spectral density is shown below:

Applied CAx / Predictive Engineering White Paper

Applied CAx / Predictive Engineering White Paper

3. THE NX NASTRAN METHOD

𝑃𝑆𝐷𝑜𝑢𝑡 = |𝑔(𝑤)|2 𝑃𝑆𝐷𝑖𝑛

There are several steps to setting up the analysis in FEMAP:

Applied CAx / Predictive Engineering White Paper

Inputs RMS Outputs

SI (m,kg,sec) 6.89e10 Pa 2710 kg/m3 1 g2/Hz 9.807 m/s2 1m 1000 Pa

Applied CAx / Predictive Engineering White Paper

5. EXAMPLE 1: CANTILEVER BEAM

𝑃𝑆𝐷𝑖𝑛 = 0.2 𝐺 2 ⁄𝐻𝑧

This excitation was applied to the fixed end

Applied CAx / Predictive Engineering White Paper

5.2 ANALYTICAL SOLUTION

Miles’ equation can be used to approximate the Gout(RMS) value:

The max output PSD can also be obtained using:

Applied CAx / Predictive Engineering White Paper

5.3 DEFINING THE SYSTEM DAMPING

Applied CAx / Predictive Engineering White Paper

5.4 CREATING THE PSD FUNCTION

Applied CAx / Predictive Engineering White Paper

Applied CAx / Predictive Engineering White Paper

It is good practice to run the normal modes

After the analysis has finished running, you

Applied CAx / Predictive Engineering White Paper

5.6 LOADING & CONSTRAINING THE MODEL

The Load Constraint constrains the base node in all

Applied CAx / Predictive Engineering White Paper

5.7 SPECIFYING GROUPS FOR NODAL AND ELEMENTAL OUTPUT

If we wanted a group for the beam element

Applied CAx / Predictive Engineering White Paper

5.8 CREATING AN ANALYSIS SET – SIMPLE PSD

First, create a new Random Response Analysis Set.

Applied CAx / Predictive Engineering White Paper

Keep pressing Next… until you arrive at the NASTRAN

Applied CAx / Predictive Engineering White Paper

In the Dynamic Analysis tab, one can specify the damping

For Frequency Response, Select the “Modal Freq…” button,

Applied CAx / Predictive Engineering White Paper

In the NASTRAN XY Output for Modal Analysis window, you

In the NASTRAN Output for Random Analysis window,

Applied CAx / Predictive Engineering White Paper

If you are interested in getting data for your entire

Applied CAx / Predictive Engineering White Paper

Applied CAx / Predictive Engineering White Paper

Choose the output requests desired. For this analysis we

Applied CAx / Predictive Engineering White Paper