Computers and Structures 81 (2003) 39–51

www.elsevier.com/locate/compstruc

Stiffness method of thin-walled beams with closed cross-section

A. Proki
c
Department of Civil Engineering, University of Subotica, Kozara
cka 2a, Koce Kapetana 44, Beograd 11000, Yugoslavia
Received 9 April 2001; accepted 15 September 2002

Abstract
The aim of the present study is to present a general stiffness method capable of analyzing three-dimensional thin-
walled straight beams with closed cross-sections. The method based on the assumptions introduced by Benscoter is
suited for automatic computation on computers. Starting from the principle of virtual displacement, an exact stiffness
matrix and vector of fixed-end reactions for the analysis of thin-walled beam with an arbitrary closed cross-section are
derived. The method is illustrated by example.
Ó 2002 Elsevier Science Ltd. All rights reserved.

Keywords: Static structural analysis; Thin-walled beams; Stiffness method; Beams with closed cross-section; Matrix method

1. Introduction publications on the subject have been published [5–7] no

general stiffness matrix for analyzing space frames made
Thin-walled beams with a closed, generally multi- up of thin-walled beams with closed sections is available
cellular cross-section, made of high strength materials in the literature. The present paper fulfills this require-
are used extensively in the aerospace industry, civil en- ment.
gineering, ship construction and etc. The basic purpose of this paper is matrix formulation
An exact analysis of such beams is usually very of the problem of structural thin-walled beams with
complex and is too complicated for routine calculations. arbitrary closed cross-section based on the assumptions
Starting from theory due to Umanski [1] and Karman introduced by Benscoter, in a way which has been found
and Christensen [2] many approximate theories for extremely convenient for the application of computers.
closed profiles have been developed. We here consider We try to follow, as close as possible, already existent
the theory presented by Benscoter [3]. The Vlasov as- techniques used in cases of assemblages of solid beams.
sumption for open profiles [4], setting the shear strain in So it is easy to implement given stiffness matrix into any
the middle surface equal to zero, is not made because it computer program for static analysis of structures. It my
will cause that no warping could develop. The warping be said that if the theory of thin-walled beams with
function is described by a function given beforehand, closed cross-section is labeled as exact the technique to
and is the same as in the case of the Saint VenantÕs be presented will yield also exact results since it does not
function of free-warping for beams with closed profiles. introduce any additional simplifications.
The function that defines the warping intensity repre- The adopted notation in this paper is very close to
sents a new unknown that may be derived as a function that of Kollbrunner and Hajdin [5].
of the angle of rotation of the profile.
With the advent of computers and their widespread
applications it is natural that BenscoterÕs theory be 2. Basic theory
translated into a stiffness matrix. Though numerous
A straight thin-walled beam with a closed, generally
multicellular cross-section, is considered. In order to
E-mail address: [email protected] (A. Proki
c). determine the geometry of the beam two coordinate

0045-7949/02/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 5 - 7 9 4 9 ( 0 2 ) 0 0 3 4 5 - 0
40 A. Proki
c / Computers and Structures 81 (2003) 39–51

systems are used. The first of these is an orthogonal nðs; zÞ ¼ vP sin a þ uP cos a þ uhnP
Cartesian coordinate systems for which the z-axis coin- ð1Þ
gðs; zÞ ¼ vP cos a  uP sin a þ uhP
cides with the longitudinal centroidal axis, while x and y
coincide with principal axes of the section. The second where a denotes the angle between the x and n axes, hnP
coordinate system is a curvilinear coordinate system represents the perpendicular distance from normal at
ðn; s; zÞ where n is the normal coordinate measured along point S to the point P given by
the normal to contour (middle line of a cross-section),
hnP ¼ ðx  xP Þ sin a  ðy  yP Þ cos a ð2Þ
and s is profile coordinate measured along the contour
line from arbitrarily taken starting point O. The wall and hP represents the perpendicular distance from tan-
thickness is denoted by tðsÞ. gent at point S to the point P given by
hP ¼ ðx  xP Þ cos a þ ðy  yP Þ sin a ð3Þ
2.1. Suppositions
hnP and hP are positive when normal n and tangent t
The theory of thin-walled beams with closed sections respectively are rotating counterclockwise about the
rests on the following simplifying assumptions: pole P , when observed from positive z direction.
The longitudinal displacement w of an arbitrary
1. The cross-section is perfectly rigid in its own plane. point on contour may be found by using the hypothesis
2. The part of the shear strains in the middle surface of concerning the absence of shearing strain in the middle
the wall, due to the bending moment, is negligible. surface due to bending moment
3. The distribution of warping distortion is the same as
in the case of Saint Venant torsion. w ¼ wo  u0P x  v0P y  #xP ð4Þ

The first three terms on the right hand side of Eq. (4)
In the technical theory of thin-walled beams with
describe longitudinal displacements of the cross-section
closed sections we assumed that normal stresses are
as a plane surface. The last term describes the warping of
constant over the entire thickness of the wall.
the cross-section, and is given as the product of two
functions. The first function #ðzÞ defines the warping
2.2. Kinematics and strains intensity and can not be directly expressed as a function
of the other parameters used to describe the plane de-
According to the first assumption the cross-sectional formation of the cross-section. This function represents
behavior can be described by only three displacement a new unknown. The second function xP ðsÞ, being the
components, two translations uP and vP and an angle of same for all cross-sections of a member, describes
twist u of arbitrarily taken pole P (Fig. 1). From geo- warping of the cross-section qualitatively. This function
metric considerations, normal and tangential displace- depends only of geometrical properties of cross-section
ments of an arbitrary point S with coordinates x, y on and is defined with the solution of Saint VenantÕs torque
the contour, where the angle of twist is sufficiently small, Z s
are xP ðsÞ ¼ ðhP  s~Þ ds ð5Þ
o

y
where s~ represents Saint Venant shear flow for Gu0 ¼ 1.
Component deformations, different from zero, are
xP x- x P given by
ow
ez ¼ ¼ w0o  u00P x  v00P y  #0 xP
h nP

oz ð6Þ
og ow
czs ¼ cT ¼ þ ¼ u0 hP  #ðhP  s~Þ
oz os
s
P
h

η ξ
t
n 2.3. Stresses and stress resultants
y- yP

vP S
From HookeÕs law for normal stress rz we get:
ϕ P uP O rz ¼ Eez ¼ Eðw0o  u00P x  v00P y  #0 xP Þ ð7Þ
yP

α
The shear stresses szs , uniformly distributed over the
C x thickness, may be expressed as a sum of the shear
stresses sT produced by torsion and the shear stresses sV
Fig. 1. Section geometry. produced by bending
 A. Proki
c / Computers and Structures 81 (2003) 39–51 41

szs ¼ sT þ sV ð8Þ externally applied moment per unit length about x, y and
z axes and external distributed bimoments, respectively.
The shear stresses sT can be derived directly from cor- The geometrical properties of cross-section are defined
responding strains by the following quantities
sT ¼ GcT ¼ G½u0 hP  #ðhP  s~Þ ð9Þ Z Z
Ixx ¼ x2 dF
The shear stresses sV cannot be obtained by simply ap- F
Z Z
plying HookeÕs Law, but only from the axial equilibrium
Iyy ¼ y 2 dF
condition. F
Reducing the normal stresses on the center of gravity Z Z
and shear stresses on the pole P we get for stress resul- Ixx ¼ x2 dF
tants the following expressions Z ZF ð14Þ
Ihh ¼ h2D dF
normalRforce
R Z ZF Z Z
N¼ F z
r dF
bending Rmoment with respect to the x axes K¼ s~2 dF ¼ s~hD dF
R F F
Mx ¼ F
rz y dF Ihh
o
bending moment with respect to the x axes Ixx ¼ Ixx
RR Ihh  K
My ¼  F rz x dF
ð10Þ
shear forceR Rin the x direction Parameter # is given by
Vx ¼  F szs sin a dF  
o o
shear force EIxx 1 EIxx
R R in the y direction #¼ u000 þ u0 þ m0D  mx
Vy ¼ F zs
s cos a dF GðIhh  KÞ GðIhh  KÞ GIhh
torsion Rmoment
R RR ð15Þ
TP ¼ s h dF or TP ¼
F zs P
s h dF
F T P
Forces in the cross-section may be defined in terms of
As it is well known, in the theory of thin-walled beams, componental displacements as shown below
we introduce a new generalized force
Z Z N ¼ EFw0o
M xP ¼ rz xp dF ð11Þ My ¼ EIxx u00D
F
Mx ¼ EIyy v00D
This force, due to warping, is called bimoment. o
EIxx ð16Þ
o
Mx ¼ EIxx u00  mD
GIhh
2.4. Differential equations o
EIxx
o
TD ¼ EIxx u000 þ GKu0 þ mx  m0
The sectorial coordinate which satisfies the condi- GIhh D
tions
Z Z It is convenient to consider TD as composed of two parts
SxP ¼ xP dF ¼ 0 TD ¼ Ts þ Tx ð17Þ
Z ZF
IxxP ¼ xxP dF ¼ 0 ð12Þ where
Z ZF
IyxP ¼ yxP dF ¼ 0 Ts ¼ GKu0 Saint Venant torque
F EI o
o
Tx ¼ EIxx u000 þ mx  xx m0D warping torque
is denoted as normalized sectorial coordinate x. The GIhh
pole P coincides now with the shear center D of the ð18Þ
cross-section. In that case, the differential equations of
equilibrium are uncoupled and can be written as
EFw00o ¼ pz 3. The matrix method approach
0
EIxx uiv
D ¼ px  my

EIyy viv 0 ð13Þ The stiffness method of analyzing a structure may be

D ¼ py þ mx
o subdivided into two parts.
E Ixx
o
EIxx uiv  GKu00 ¼ mD þ m0x  m00 Part I: A beam as the individual element of the
G Ihh D structure is considered. The vector of generalized nodal
In Eqs. (13) px , py , pz , mx , my , mz and mx represent ex- forces R of an element, using the principle of superpo-
ternal loads per unit length in the x, y and z directions, sition, may be expressed as the sum of
42 A. Proki
c / Computers and Structures 81 (2003) 39–51

(a) the vector of generalized nodal forces Rq of the ele-

ment when subjected only to its nodal displace-
ments,
(b) the vector of generalized nodal forces Ro , of the ele-
ment when subjected to the given external loads with
its ends fixed.

Part II: The structure made up of thin-walled beams

connected together to a number of nodes is considered.
Using
Fig. 2. Generalized displacements and forces of thin-walled
(a) the compatibility equations, i.e. equating the appro- beam.
priate nodal displacements of those elements which
have common node,  T
qsy ¼ ui uyi uk uyk
(b) the equilibrium equations, i.e. equating at each node
qsx ¼ ½ vi uxi vk uxk T
the external load with the nodal forces of elements ð24Þ
which meet in that nodes, qa ¼ ½ wi wk T
(c) boundary conditions, qt ¼ ½ uzi #i uzk #k T
are vectors of nodal parameters associated with bending
we get the stiffness equations for computing the un-
about y and x axes, axial loading and torsion respec-
known displacements.
tively.
The column matrices of the associated nodal forces
3.1. The stiffness matrix may be defined as

Eqs. (1) and (4) can be converted to matrix form R ¼ ½ Rsy Rsx Ra R t T ð25Þ
2 3 where
2 3 2 3 uD
n cos a sin a 0 hnD 0 6 7
6 vD 7 Rsy ¼ ½ Vxi Myi Vxk Myk T
4 g 5 ¼ 4  sin a cos a 0 hD 0 56 w 7 ð19Þ
6 o7 Rsx ¼ ½ Vyi Mxi Vyk Mxk T
w x dzd y dzd 1 0 x 4 u 5 ð26Þ
# Ra ¼ ½ Ni Nk T
Rt ¼ ½ Ti Mxi Tk Mxk T
Since, according to (15)
Parameters uD , vD , wo and u can be defined in terms of
q1 the shape functions and the associated nodal parameters
# ¼ 2 u000 þ u0 ð20Þ 2 3 2 32 3
k uD N sy qsy
6 vD 7 6 N 76 qsx 7
where 6 7¼6 sx 76 7 ð27Þ
4 wo 5 4 Na 54 qa 5
sffiffiffiffiffiffiffiffiffiffi u Nt qt
GK Ihh
k¼ ; q¼ ð21Þ where the shape functions N sy , N sx , N a and N t will be
o
EIxx Ihh  K
determined later on.
it follows that Substituting (27) into (22) we get:
2 3
2 3 2 cos a 32 u 3 n
n sin a 0 hnD D 6 7
76 vD 7 4g5¼
4g5¼64  sin a cos a 0 hD  56 7
d3
4 wo 5 w
w x dzd y dzd 1 x dzd þ q1
k2 dz3 2 3
u cos aN sy sin aN sx 0 hnD N t
ð22Þ 6  sin aN cos aN sx 0 hD N t 7
6 sy 7
4 5
Let us denote the vector of generalized nodal dis- xN 0sy yN 0sx N a x N 0t þ q1 k2
N 000
t
2 3
placements (Fig. 2) in the following way: qsy
 T 6q 7
6 sx 7
q ¼ qsy qsx qa qt ð23Þ 6 7 ¼ Aq ð28Þ
4 qa 5
where qt
 A. Proki
c / Computers and Structures 81 (2003) 39–51 43
Z Z
As there are only two independent and nonzero
dRs ¼ RT dq þ pT du dF
 ð36Þ
strain components, the strain vector is given by F

e ¼ ½ ez csz T ð29Þ where dF is the element of middle surface of the beam F .

The virtual work of the actual stresses trough the virtual
where strains is
Z Z Z
ow dRu ¼  rT de dV ; ð37Þ
ez ¼
oz ð30Þ
V
og ow
csz ¼ þ If the vector of virtual displacements du is adopted in the
oz os
same form as a vector of real displacements, than,
Substituting (28) into (30), we get equating the external work with the internal work we
obtain
e ¼ Bq ð31Þ
R ¼ kq þ Ro ð38Þ
where where
" # Z Z Z
o
oz
B¼ o o
A¼ k¼ B T DB dV
oz os V
2  3 Z Z ð39Þ
xN 00sy yN 00sx N 0a x N 00t þ q1 N iv Ro ¼  AT p dF
6 k2 t 7
4 5 F
0 0 q1 000
0 0 0 hD N t  ðhD  s~Þ N t þ k2 N t
are the stiffness matrix and vector of fixed-end reactions
ð32Þ respectively.
Since
If we denote with D the matrix of elastic constants
  q  1 iv
E N 00t þ N t ¼ qN 00t ð40Þ
D¼ ð33Þ k2
G
then, by inserting (32) and (33) in (39-1) we farther ob-
for stress vector tain

Z Z Z
k¼ B T DB dV
V
2 3
Ex2 N 00T 00
sy N sy
6 7
Z Z Z 6 Ey 2 N 00T 00
sx N sx 7
6 7
6 EN 0T 0
a Na h 7
¼ 6 7 dV
V 6 Eq2 x2 N t N t þ G s~2 N t N t  k 2 ðhD s~  s~2 ÞðN t N t þ N t N t Þ 7
00T 00 0T 0 q1 0T 000 000T 0
6 7
4 i 5
ðq1Þ2 2 2 000T 000
þ k 4 ðhD  2hD s~ þ s~ ÞN t N t

ð41Þ

r ¼ ½ rz ssz T ð34Þ Taking account of Eqs. (14) and (21) for stiffness matrix
k we finally get
we get
2 3
r ¼ De ¼ DBq ð35Þ kby
6 kbx 7
k¼6
4
7
5 ð42Þ
ka
The principle of virtual displacements will be used to kt
give a stiffness matrix and a vector of fixed-end reac-
tions.
In a virtual displacement, the distributed forces p and
generalized nodal forces R, perform virtual work where
44 A. Proki
c / Computers and Structures 81 (2003) 39–51
Z
uðzÞ ¼ a1 þ a2 kz þ a3 sh kz þ a4 ch kz ð46Þ
kby ¼ EIxx N 00T 00
by N by dz
ZL where ai (i ¼ 1; . . . ; 4) are some arbitrary constants to be
kbx ¼ EIyy N 00T 00
bx N bx dz determined from boundary conditions
Z L 
u ¼ uzi
ka ¼ EF N 0T 0
a N a dz
z¼0
L # ¼ #i
Z    ð47Þ
o 00T 00 2 0T 0 q  1 000T 000 u ¼ uzk
kt ¼ EIxx qN t N t þ k N t N t þ 2 N t N t dz z¼L
L k # ¼ #k
ð43Þ
Taking into account relation between the functions #ðzÞ
and uðzÞ given by Eq. (20), and after elimination of
are stiffness matrices of flexure in two principal planes, the constants ai from Eq. (46), we find the angle of rota-
axial loading and torsion. tion uðzÞ as a function of nodal displacements qt ¼
The quasidiagonal form of the stiffness matrix (42) ½ uzi #i uzk #k T
reflects the fact that the system of differential equations
(13) is uncoupled, i.e. the axial deformation, bending uðzÞ ¼ N t qt ð48Þ
and torsion in a single element may be analyzed inde-
where
pendently. The axial and bending deformation can be
treated by classical engineerÕs theory in the usual manner N t ¼ ½N1 ðzÞN2 ðzÞN3 ðzÞN4 ðzÞ;
and the correspondent stiffness matrices are given here in 1
final form N1 ðzÞ ¼ ½1 þ qk sh k  ch k  q sh kkz þ sh k sh kz
D
2 3 þ ð1  ch kÞ ch kz
12 6L 12 6L
1
EIxx 6
6 6L 4L2
6L 2L2 7
7 N2 ðzÞ ¼
qkD
½qk ch k  sh k þ ðq  q ch kÞkz
kby ¼ 3 6 7
L 4 12 6L 12 6L 5 þ ð1 þ qk sh k  ch kÞ sh kz
6L 2L2 6L 4L2 þ ðsh k  qk ch kÞ ch kz
2 3
12 6L 12 6L
ð44Þ 1
6
EIyy 6 6L 4L2 6L 2L2 7 N3 ðzÞ ¼ ½1  ch k þ q sh kkz  sh k sh kz
7 D
kbx ¼ 3 6 7
L 4 12 6L 12 6L 5  ð1  ch kÞ ch kz
6L 2L2 6L 4L2 1
  N4 ðzÞ ¼ ½sh k  qk þ ðq  q ch kÞkz  ð1  ch kÞ sh kz
EF 1 1 qkD
ka ¼
L 1 1 þ ðqk  sh kÞ ch kz
D ¼ 2ð1  ch kÞ þ qk sh k k ¼ kL
To obtain the torsional stiffness matrix we have first to ð49Þ
determine the vector of shape functions N t as solution of
the differential equation (13-4) governing the problem of The torsion stiffness matrix may now be written from
torsion (in absence of external loads) (43-4) and (49) as

2 3
qk 3 sh k k 2 ðch k  1Þ qk 3 sh k k 2 ðch k  1Þ
o 6 k ðch k  1Þ
2 k
ðqk ch k  sh kÞ k 2 ðch k  1Þ  q ðqk  sh kÞ 7
k
EIxx 6 q 7
kt ¼ ð50Þ
D 4 qk 3 sh k 2
k ðch k  1Þ qk 3 sh k k 2 ðch k  1Þ 5
k 2 ðch k  1Þ k 2
 q ðqk  sh kÞ k ðch k  1Þ k
q
ðqk ch k  sh kÞ

d 4u d 2u It is convenient to write the elements of vector of

 k2 2 ¼ 0 ð45Þ
dz 4 dz nodal displacements and vector of associated nodal
forces in the following order:
The general solution of Eq. (45) can be written as

 T
q ¼ ui vi wi uxi uyi uzi #i uk vk wk uxk uyk uzk #k
 q q q q q q q T ð51Þ
Rq ¼ Vxi Vyi Ni Mxi Myi Ti Mxi Vxkq Vykq Nkq Mxiq Myiq Tkq q
Mxk
 A. Proki
c / Computers and Structures 81 (2003) 39–51 45

In this case, taking into account (44) and (50), for a gen-
eral space thin-walled beam with closed cross-section the
final form of the stiffness matrix is given by relation (52)

2 3
12EIxx 6EIxx 12EIxx 6EIxx
6 L3  7
6 L2 L3 L2 7
6 12EIyy 6EIyy 12EIyy 6EIyy 7
6    7
6 L3 L2 L3 L2 7
6 7
6 EF EF 7
6  7
6 L L 7
6 7
6 4EIyy 6EIyy 2EIyy 7
6 7
6 L L2 L 7
6 7
6 4EIxx 6EIxx 2EIxx 7
6 7
6  7
6 L L2 L 7
6 o
EIxx o
EIxx o
EIxx o
EIxx 2 7
6 7
6 qk 3 s k 2 ðc  1Þ  qk 3 s k ðc  1Þ 7
6 D D D D 7
6 o o o 7
6 EIxx k EIxx EI k 7
6 ðqkc  sÞ  k 2 ðc  1Þ  xx ðqk  sÞ 7
6 D q D D q 7
k¼6
6
7
7
6 12EIxx 6EIxx 7
6  7
6 L3 L2 7
6 12EIyy 6EIyy 7
6 symmetric 7
6 7
6 L3 L2 7
6 EF 7
6 7
6 7
6 L 7
6 7
6 4EIyy 7
6 7
6 L 7
6 7
6 4EIxx 7
6 7
6 L 7
6 7
6 o
EIxx EIxxo 7
6 qk 3 s  k 2 ðc  1Þ 7
6 D D 7
6 7
4 o
EIxx k 5
ðqkc  sÞ
D q

ðs ¼ sh k; c ¼ ch kÞ ð52Þ

Z Z
3.2. Vector of fixed-end reactions
px ¼ ðcos a
pn  sin a
pg Þ ds ¼ px ds

s s
The components of the vector of fixed-end reactions Z Z
are obtained by substituting the expression for A into py ¼ ðsin a
pn þ cos a
pg Þ ds ¼ py ds

s s
Eq. (39-2) Z
82 3 pz ¼ pz ds

>
> cos aN Tsy  sin aN Tsy xN 0T
sy s
>6
> 7 Z
Z Z > <6 sin aN Tsx cos aN Tsx yN 0T 7
6 sx 7 mx ¼ pz ds
y
Ro ¼  6 7
F > >6 0 0 N Ta 7 s
>
>4 5 Z ð55Þ
>
: hnD N T hD N Tt x N 0T q1 000T
t þ k2 N t
t
my ¼  pz ds
x
9 s
2 3> >
> Z
pn > >
=
6 7 mD ¼ ðhnD 
pn þ hD 
pg Þ ds
4 pg 5 dF ð53Þ s
>
> Z
pz >>
>
; ¼ ½ðx  xD Þ
py  ðy  yD Þpx  ds
s
Z
After integration along the middle line we get mx ¼ pz ds
x
s
2 3
px N Tsy þ my N 0T
sy represent external loads in the x, y and z directions; ex-
Z 6 7
6 py N Tsx þ mx N 0T 7 ternally applied moments about x and y axes and ex-
R ¼ 6 7 dz
o sx
6 p z N Ta ð54Þ ternally distributed torsional moments and bimoments,
L 4 7
5
mD N Tt  mx N 0T þ q1
N 000T per unit length of the beam, respectively.
t k2 t
The vector of nodal forces of the element with its
ends fixed subjected to the given distributed torsion
in which moments and bimoments is given by
46 A. Proki
c / Computers and Structures 81 (2003) 39–51

Z   
T 0T q  1 000T mωk
Rot ¼ mD N t  mx N t þ 2 N t dz ð56Þ
L k
m ωi
where N t is given by (49).
In the case of torques mD and bimoments mx con-
centrated at a finite number of discrete points zr Ti
O i k O

O
Tk
(r ¼ 1; 2; . . .) Eq. (56) might be written as M ωi Mωk
O

X 
q  1 000T

o T 0T
Rt ¼  mDr N t ðzr Þ  mxr N t ðzr Þ þ 2 N t ðzr Þ Fig. 4. Beam subjected to distributed bimoments.
r
k
ð57Þ
in which
We finally calculate the components of the vector of _

fixed-end reactions for some elementary cases of load D ¼ 2ð1  ch kÞ þ k sh k ð59Þ

distribution. (b) Distributed bimoments (Fig. 4)
_
(a) Distributed torsional moments (Fig. 3) qD
! Tio ¼  ðmxi þ mxk Þ
_ 2D
o L k2 D k3  2 
Ti ¼  2  þ q sh k mDi L k
k 4 D 12D
o
Mxi ¼ 2 ð1  ch kÞ þ qk2 ch k  D  k sh k mxi
! Dk 2
_  
L k2 D k3 L k2
 2 þ  q sh k mDk þ 2 ð1  ch kÞ  qk2 þ D þ k sh k mxk
k 4 D 12D Dk 2
 _
o L2 k qk2 qk2 qD
Mxi ¼ 2 sh k   ch k þ D Tko ¼ ðmxi þ mxk Þ
qk D 2 6 3 2D
  
þ ðq  1Þð1  ch kÞ mDi L k2
o
Mxk ¼ 2 ð1  ch kÞ  qk2 þ D þ k sh k mxi
 Dk 2
L2 k qk2 qk2  
þ 2 sh k   ch k L k2
qk D 2 3 6 þ 2 ð1  ch kÞ þ qk2 ch k  D  k sh k mxk
 Dk 2
 ðq  1Þð1  ch kÞ mDk ð60Þ
_ ! ð58Þ
(c) Concentrated torque (Fig. 5)
o L k2 D k3
Tk ¼  2 þ  q sh k mDi mD
k 4 D 12D Tio ¼  ½1 þ qk sh k  ch k  q sh kkzo
_ ! D
L k2 D k3 þ sh k sh kzo þ ð1  ch kÞ ch kzo 
 2  þ q sh k mDk
k 4 D 12D o mD
 Mxi ¼  ½qk ch k  sh k þ ðq  q ch kÞkzo
L2
k qk2 qk2 qkD
o
Mxk ¼ 2 sh k   ch k þ ð1 þ qk sh k  ch kÞ sh kzo
qk D 2 3 6

þ ðsh k  qk ch kÞ ch kzo  ð61Þ
 ðq  1Þð1  ch kÞ mDi mD
o
 Tk ¼  ½1  ch k þ q sh kkzo  sh k sh kzo
L2 k qk2 qk2 D
 2 sh k   ch k  ð1  ch kÞ ch kzo 
qk D 2 6 3
 o mD
Mxk ¼  ½sh k  qk þ ðq  q ch kÞkzo
þ D þ ðq  1Þð1  ch kÞ mDk qkD
 ð1  ch kÞ sh kzo þ ðqk  sh kÞ ch kzo 

m Dk
mD
m Di
O i k O
Ti Tk
O i k Tk
O O
M ωi M ωk
O
Ti O O
M ωi M ωk

Fig. 3. Beam subjected to distributed torsional moments. Fig. 5. Beam subjected to concentrated torque.
 A. Proki
c / Computers and Structures 81 (2003) 39–51 47

mω cannot be ascertained by consideration of warping w-

displacement distributions at the end cross-section of
O i k O
two elements meeting at an angle, because these distri-
Ti Tk butions are incompatible for any nonzero angle between
O O
M ωi Mωk the elements.
There are many papers that deal with this problem.
Fig. 6. Beam subjected to concentrated bimoment. In Ref. [5] Eq. (63) has been implied without any
justification. Bazant and Nimeiri [9] found that diagonal
(d) Concentrated bimoment (Fig. 6) coefficient for # or Mx in Eq. (63) must be 1.
mx qk Yang and McGuire [10] introduced an elastic support
Tio ¼ ½sh k þ sh k ch kzo þ ð1  ch kÞ sh kzo  with a ‘‘warping spring’’, similar in concept to the semi-
D
m x
rigid connection element often used to describe the
o
Mxi ¼ ½1  ch k þ ð1 þ qk sh k  ch kÞ ch kzo flexural resistance of a connection, which can be used to
D
þ ðsh k  qk ch kÞ sh kzo  represent the partial warping restraint for each end of
ð62Þ the beam element. In the work by Krenk and Damkilde
m x qk
Tko ¼ ½sh k  sh k ch kzo  ð1  ch kÞ sh kzo  [11] suitable models for the coupled warping at joints are
D developed. However, as the only type of joint considered
o m x
Mxk ¼ ½1  ch k  ð1  ch kÞ ch kzo in their analysis is one between thin-walled I-beams,
D
their transformations for bimoment and warping are not
þ ðqk  sh kÞ sh kzo 
applicable to other type of joints.
For other load distributions (such as parabolic, In any case, farther theoretical and experimental in-
piece-wise continuous etc.) one directly performs the vestigations about this matter are necessary.
integration according to the relation (56).
5. Determination of stresses
4. Transformation matrix
In order to obtain the normal stress rz and shear
This paper introduces the concept of continuous stresses ss and sw the cross-section forces N , Vx , Vy , Mx ,
warping. It is supposed that the connections between the My , Tx , Ts and Mx must be computed before [8].
elements are such that they enable direct coupling of As is well known a thin-walled beam is always stat-
warp at the joints. In that case, the transformation ically indeterminate regardless of torsional moments Ts
matrix T, which transforms the local components of the and Tx and bimoments Mx . To determine the distribu-
generalized nodal forces and displacements as well as stiff- tion of torsional moments and bimoments in the beam
ness matrix to global and vice versa, is shown as below one has to determine the function uðzÞ solving the dif-
ferential equation
d4 uðzÞ d2 uðzÞ f ðzÞ
4
 k2 ¼ o ð65Þ
dz dz2 EIxx
ð63Þ in which
f ðzÞ ¼ mD ðzÞ þ m0x ðzÞ  vm00D ðzÞ
q1 ð66Þ
v¼
qk 2
Submatrix t is the 3 3 rotation matrix given by The solution of this differential equation is
2 3
t11 t12 t13
uðzÞ ¼ a1 þ a2 kz þ a3 sh kz þ a4 ch kz þ up ðzÞ ð67Þ
t ¼ 4 t21 t22 t23 5 ð64Þ
t31 t32 t33
where
where tnm (n; m ¼ 1; 2; 3) is direction cosine of the local x,
y and z axis relative to the global X , Y and Z axis.
All elements of matrix T are determined from kine-
matics, except those which multiply # or Mx . The fact
that warping parameter # in Eq. (63) transform with
unity and is independent from other generalized dis-
placements means that all the members rigidly fixed into
a joint experience an equal amount of warping. This Fig. 7. Beam subjected to distributed torsional moments.
48 A. Proki
c / Computers and Structures 81 (2003) 39–51
Z z 
1 1
up ðzÞ ¼  zn sh kðz  nÞ f ðnÞ dn ð68Þ
GK o k

is the particular solution depends on external loads.

The constants a1 , a2 , a3 and a4 are evaluated by re-
quiring that the solution (67) satisfies the boundary
Fig. 8. Beam subjected to concentrated torque. conditions given on the left end of the beam which are
known from previous calculation
8
>
> uðzÞ ¼ uzi
>
>
>
> #ðzÞ ¼ qvu000 ðzÞ þ u0 ðzÞ þ GIqhh
>
>
>
>
< ½vm0D ðzÞ  mx ðzÞ ¼ #i
z¼0 ð69Þ
> Mx ðzÞ ¼ EIxx
> o
u00 ðzÞ  vmD ðzÞ ¼ Mxi
>
>
>
>
> TD ðzÞ ¼ EIxx
>
o
u000 ðzÞ þ GKu0 ðzÞ
>
>
:
þmx ðzÞ  vm0D ðzÞ ¼ Ti
Fig. 9. Beam subjected to distributed bimoments.
After elimination of the constants ai from (67) it follows
1 1
uðzÞ ¼ uzi þ sh kz#i þ ð1  ch kzÞMxi
qk GK
 
1 1 1
 z sh kz Ti þ ð1  ch kzÞvmD ð0Þ
GK qk GK
 
1 1
 z  sh kz ½mx ð0Þ  vm0D ð0Þ
GK k
Z z 
1 1
 z  n  sh kðz  nÞ f ðnÞ dn ð70Þ
Fig. 10. Beam subjected to concentrated bimoment. GK o k

Fig. 11. Example: (a) plane grid subjected to torque and (b) generalized displacements.

Table 1
Geometric Properties
Elements Joint i Joint k L (m) k (m) o
Ixx =Ixx (m2 ) q cos a sin a
1 1 2 5.0 0.4 2.0 4.0 1 0
2 1 3 5.0 0.4 2.0 4.0 0 1
 A. Proki
c / Computers and Structures 81 (2003) 39–51 49

Fig. 12. Diagrams of internal forces.

 
For some elementary cases of load distribution we GK 1
finally calculate the stress resultants Ts , Tx and Mx , Ts ðzÞ ¼ ch kz#i  k sh kzMxi  1  ch kz Ti
q   q
Z z
i.e. 1
 1  ch kðz  nÞ mD ðnÞ dn
o q
GK 1
Ts ¼ GKu0 ; o
Tx ¼ EIxx u000 þ mx  vm0D ; Tx ðzÞ ¼  ch kz#i þ k sh kzMxi  ch kzTi
q
Z z q ð72Þ
1
 ch kðz  nÞmD ðnÞ dn
o
Mx ¼ EIxx u00  vmD ð71Þ o q
GK 1
Mx ðzÞ ¼  sh kz#i þ ch kzMxi  sh kzTi
qk
Z z qk
(a) Distributed torsional moments f ðnÞ ¼ mD ðnÞ  1
 sh kðz  nÞmD ðnÞ dn
vm00D ðnÞ (Fig. 7) o qk
50 A. Proki
c / Computers and Structures 81 (2003) 39–51

(b) Concentrated torque (Fig. 8) (d) Concentrated bimoment (Fig. 10)

 
GK 1
Ts ðzÞ ¼ ch kz#i  k sh kzMxi  1  ch kz Ti  
q q GK 1
Ts ðzÞ ¼ ch kz#i  k sh kzMxi  1  ch kz Ti
#  q q
# 1
 ## 1  ch kðz  zo Þ mD
q þ jk sh kðz  zo Þmx

GK 1 GK 1
Tx ðzÞ ¼  ch kz#i þ k sh kzMxi  ch kzTi Tx ðzÞ ¼  ch kz#i þ k sh kzMxi  ch kzTi
q q q q
#
#1  jk sh kðz  zo Þmx
 ## ch kðz  zo ÞmD
q
GK 1
Mx ðzÞ ¼  sh kz#i þ ch kzMxi  sh kzTi
GK 1 qk qk
Mx ðzÞ ¼  sh kz#i þ ch kzMxi  sh kzTi
qk qk
 j ch kðz  zo Þmx
#
# 1 ð75Þ
 ## sh kðz  zo ÞmD
qk
ð73Þ

The terms behind vertical lines should be added for

z > zo . 6. Numerical example
(c) Distributed bimoments f ðnÞ ¼ m0x ðnÞ (Fig. 9)
The example considered is a simple plane grid as-
  sembled from two identical thin-walled beams with
GK 1 closed cross-sections shown in Fig. 11a. The grid is
Ts ðzÞ ¼ ch kz#i  k sh kzMxi  1  ch kz Ti
q q loaded by a concentrated torque mD acting at the point
Z z A. The choice of basic nodal parameter is shown in Fig.
þ k sh kðz  nÞmx ðnÞ dn 11b.
o The geometric properties are given in Table 1. Table
1 is used to calculate the following parameters
GK 1
Tx ðzÞ ¼  ch kz#i þ k sh kzMxi  ch kzTi
q q
Z z k ¼ kL ¼ 0:4 5 ¼ 2:0
 k sh kðz  nÞmx ðnÞ dn
o s ¼ sh k ¼ 3:62686
ð76Þ
GK 1 c ¼ ch k ¼ 3:76220
Mx ðzÞ ¼  sh kz#i þ ch kzMxi  sh kzTi
qk qk
Z z D ¼ 2ð1  ch kÞ þ qk sh k ¼ 23:49049
 ch kðz  nÞmx ðnÞ dn
o
ð74Þ For plane grid, from (52) and (63), we get

ð77Þ
 A. Proki
c / Computers and Structures 81 (2003) 39–51 51
2 3
1 References
6 cos a  sin a 7
6 7
6 sin a cos a 7 [1] Umanskij AA. Krucenije i izgib tankostennih avio-kon-
6 7
6 1 7 strukcij. Moskva, Russia; 1939 [in Russian].
T¼6
6
7
7 [2] Karman T, Christensen NB. Methods of analysis for
6 1 7
6 cos a  sin a 7 torsion with variable twist. J Aero, Sci 1944;2(II):110–
6 7
4 sin a cos a 5 24.
1 [3] Benscoter SU. A theory of torsion bending for multi-cell
beams. J Appl Mech 1954;21(1):25–34.
ð78Þ [4] Vlasov VZ. Tankostennie uprugie sterzni. Stroiizdat,
Moakva, Russia; 1940 [in Russian].
Performing the standard matrix method for analyzing [5] Kollbrunner CF, Hajdin N. Dunnwandige stabe. Band 1.
structures we get the generalized displacements and Berlin: Springer; 1972 [in German].
forces at the ends of the beams, and than, from ex- [6] Gjelsvik A. The theory of thin-walled bars. New York:
pressions (73) forces Ts , Tx and Mx at arbitrary points John Wiley & Sons; 1981.
along the axes of the beams, Fig. 12. [7] Murray NW. Introduction to the theory of thin-
walled structures. New York: Oxford University Press;
1984.
[8] Prokic A. Computer program for determination of geo-
7. Conclusion metrical properties of thin-walled beams with open-closed
section. Comput Struct 2000;74(6).
[9] Bazant Z, Nimeiri M. Large-deflection spatial buckling of
The stiffness method of analyzing space structure
thin-walled beams and frames. J Eng Mech 1973;99(EM6):
composed of thin-walled beams with closed cross-sec-
1259–81.
tion is presented. The proposed method can be used by [10] Yang YB, McGuire W. A procedure for analysing space
practicing engineers for obtaining accurate analysis re- frames with partial warping restraint. Int J Numer Meth
sults of such constructions. The stiffness matrix may be Eng 1984;20:1377–98.
easy implemented into any computer program for static [11] Krenk S, Damkilde L. Warping of joints in I-beam
analysis of constructions. assemblages. J Eng Mech 1991;117(11):2457–74.