INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING, VOL.

2 335-352 (1970)

FINITE ELEMENT ANALYSIS OF TORSIONAL AND

TORSIONAL-FLEXURAL STABILITY PROBLEMS
ROSHDY s. BARSOUM* AND RICHARD H. GALLAGHER?
* Graduate Student, School of Civil Engineering, Cornell University, Ithaca
t Professor of Civil Engineering, Cornell University, Ithaca

SUMMARY
Stiffnessmatrices are formulated for the torsional and lateral stability analysis of structures composed
of flexural members by the matrix displacement method. The formulations are based upon approximate
displacement fields which represent the action of the element in simple flexure. Example problems,
for which exact solutions are known, illustrate the accuracy and convergence characteristics of the
derived formulations.

INTRODUCTION
The capabilities of the electronic digital computer have enabled the performance of linear
structural analyses, excluding elastic instability effects, for large-scale space frameworks. No
fundamental developments in structural theory were necessary since the relationships for the
axial, flexural and torsional behaviour of prismatic members were known and have been employed
for many decades. The relatively recent formulation of generalized procedures, often in matrix
terminology, has served the purpose of facilitating and enhancing the efficiency of such
applications.
Many structures are prone to the effects of elastic instability phenomena, extending from
influences on stable equilibrium states (e.g. beam-column action) where linear theory remains
sufficiently accurate, through to highly non-linear effects as encountered in catastrophic failure
processes. Although classical formulations continue to be applicable to digital computer analyses
of large-scale systems for such circumstances, an opportunity exists for improved efficiency, at
the expense of approximation by virtue of finite element concepts.’ The envisioned efficiencies
may prove to be essential from the view of economically feasible analysis.
The designation ‘finite element concepts’, as employed herein, is intended to characterize the
formulation of relationships between the forces and displacements of a single member via
simplified assumptions as to the behaviour of the element in terms of stress or displacement.
The approach is well established with respect to linear, stable behaviour.lr2 The extension to
deal with problems of elastic instability has advanced rapidly in recent years. Progress to date
with respect to prismatic members has been surveyed in References 3 and 4.
The treatment to date of elastic instability phenomena in structures composed of prismatic
members (e.g. frames) has largely been restricted to flexural behaviour in a single plane. Yet,
consideration of problems associated with space frameworks suggests the need to cope with
torsional instability as well as combined torsional-flexural instability. These problems are of
course of significance in their own right in connection with the instability of single-span or
continuous members or planar frameworks.
Received I August 1969
Revised 12 September 1969
@ 1970 by John Wiley & Sons, Ltd.

335
336 ROSHDY S. BARSOUM A N D RICHARD H . GALLAGHER

The purpose of this paper is therefore to devclop stiffness equations for the torsional and
combined flexural-torsional instability of one-dimensional members of constant cross-section.
Rigorous formulations for the subject type of element, based upon closed-form solution of
the governing differential equations, have been presented by Renton.5
I t should be noted that the problem of torsional-flexural buckling, for various types of open
sections, was extensively studied by Chajes and Winter.6 The study was restricted to fixed and
hinged boundary conditions.

BASIC CONSIDERATIONS AND POTENTIAL ENERGY EXPRESSION

The subject element is portrayed in Figure la. The rectangular co-ordinate system x, y , z is chosen
such that y and z coincide with the principal axes of the cross-section of the element before
buckling, and x coincides with the centroidal axis c1-c2 before buckling.
Let 11, u' denote the displacements of the centroidal axis in the y and z directions, respectively,
+ is the angle of twist, and u, the axial displacement. The quantities u, D, u' and 4 are all assumed
to be of small order, so that their products are negligible when compared with their individual
magnitudes.

Figure 1. Element forces

The usual assumptions for thin-walled beams,' adopted here, are

1. The shear deformations of the middle surface can be assumed to vanish.
2. The cross-section is not deformed in its plane.
 FINITE ELEMENT ANALYSIS 337

The first assumption leads to the definition of the angular displacements 8, $ in terms of the
first derivatives of the transverse displacements :

8=-,du *=-
dw
dx dx
The element shown in Figure 1 is acted upon by an axial force F, with eccentricities eu,ez
from the centroid, the end forces and end moments, Qyl, Q@, Q,,, Qz2, My,, MyB,M,, and MZ2*
acting along axes parallel to the principal axes through the shear centre (Reference 8, l58), the
transverse load qz acting at a distance i, from the shear centre, and the torque M, along an axis
through the shear centre.
The potential energy mP is given by
7 r p = u- v (2)
where U is the strain energy and V is the potential of the applied loads.
The strain energy U is given by (Reference 8, 115 )

‘b +
U = -j [EZ,u”’+ EZ, w“’ EC, $”’+ GKd” +EAu“] dx
where a prime denotes d/dx; E is Young’s modulus; G, modulus of rigidity; I,, I,, moments of
(3)

inertia of the cross-section about its principal axes y and z , respectively; C,,, warping constant of
the section ; K , St.Venant torsion constant; and A, cross-sectional area.
The potential of the applied loads consists of two parts. The first part, V, is the product of
the applied loads and their corresponding displacements :
P = F J u i - 4 + Q,iui+ Q@uz+ Q z i wi + Qz2 ~2
+ Mu1 *I + Mu2 *2 + MZ14 + Mzz 02
+
d42 w dx+ M&l - dz)

The utilization of potential energy concepts in the establishment of relationships for instability
(4)

analysis presumes that prebuckling deformations have taken place and that an examination is
being conducted of adjacent equilibrium states. Thus, only the changes in potential energy due to
flexural and torsional action are sought in finding the second part, p.
is given as a summation of the separate contributions of the different forces.
v= i

= ;(I* z3dA + IA*y’z d A ) -22,

* Numerals I or 2 refer to the position where the force or moment is acting, at end 1 or 2 respectively.
23
338 ROSHDY S. BARSOUM A N D RICHARD H . GALLAGHER

where y,,zo are the co-ordinates of the shear centre (Figure Ib) and I, is the polar moment of
inertia about the shear centre 0 (Reference 9, 245).

P r

+
v”r$[x2 ( I - x ) ~dx]
] - &qzdz
1
dx

Equations (7-1 I ) are derived in Appendix 1.

For a conservative torque,1° the potential is given by

excluding the boundary terms which depend on the type of the applied torque.ll It is also to be
noted that equation (12) and the corresponding strain energy is strictly valid only for the buckling
analysis of a circular shaft.lZ

FUNCTIONAL REPRESENTATION OF DISPLACEMENTS

Transformation of equation (2) into the desired stiffness expressions requires the selection of
displacement functions in description of the behaviour of the element. Each displacement
component is assumed of the general form

A = A Ai (1 3)
where,fi are the ‘shape functions’, describing the displacement field for a unit value of the joint
displacement Ai-with all other joint displacements suppressed. These parameters are functions
of the co-ordinate x. The chosen functions, which are suggested by the exact solutioits of axial
and purely flexural displacement, are
 FINITE ELEMENT ANALYSIS 339

where

.fz = [3 ( y - 2 (y
h = - x [(+I]’

f4 = [p(;)
-x
and
el = - (2);
e2 (g) =-
2

dw dw
$1 = -(& $2 = -(z) 2

xl= -(
d+z);
x 2 = - (z)2
d4
It should be noted that the expression for twist does not satisfy the governing equation excluding
distribution loading and instability terms, i.e. the equation:
ECw&v- GK4” = 0 (17)
The significance of this factor, which has no bearing upon the potential energy aspects of the
formulation, will be discussed in conjunction with the review of numerical solutions.
340 ROSHDY S. BARSOUM A N D RICHARD H. GALLAGHER

or

*
W

lu, w9 *,u, 8, $ 9 XI
TT, =
2 [kI{ ’
e
$
\XI

*
W

+ “ 9 w l *’ 9’ 3‘ $9 XI [n] u - ~ uw,
, tb, 21, e, $1 { F )
2
e
4

[kl =

where k , is the axial stiffness, k,: and k , are the flexural stiffness in the L’ and w directions
respectively, and k+ is the torsioanl stiffness.
The matrix [n]is the element geometric matrix, which could be written as

where F,, Qy, Q,, M,, M,, M, and q, are the element forces and n,, naur,n,, ntV+, niw6 and ni+
are the element geometric submatrices, found from equation (1 8) (i.e.

etc.). It is clear that [n]is a function of the element forces and its geometry.
Both [k] and [n] are given in Appendix 11.
 FINITE ELEMENT ANALYSIS 341

THE PRINCIPLE OF MINIMUM POTENTIAL ENERGY FOR STABILITY PREDICTION

For stable equilibrium we will invoke the principle of minimum potential energy which states
that
6np = 0 (22)
Performing the variation on the expression for 7rP equation (19), results in the element stiffness
equations
{ F } = P I {A}+ b l {A> (23)
From the conditions of equilibrium and continuity the system equations are
{PI = [KI{A)+ " 1 1 4 (24)
In all problems of stability treated in this paper the buckling deformations are assumed
completely independent of the prebuckling deformations, i.e. they are uncoupled. Thus a study
of instability would require suppression of all prebuckling deformations in the analysis. Since
the applied loads are only in the direction of the prebuckling deformations, {P} in equation (24)
will contain zeros elsewhere. It is thus possible to express the element forces and moments as
ratios of the load that introduces instability of the whole structure. The analysis procedure
begins with the choice of an arbitrary value of the applied load, from which the individual
element end forces are found from a prebuckling analysis. The critical load would be then a
factor X times the chosen load F; X here is the instability parameter (eigenvalue). The preceding
argument reduces equation (24) to

where {E} are the buckling deformations only.

For the non-trivial solution the condition is
I [Rl+X"lI =0
or
1 -
3 {A} = - [R]-I [ N ]{h>

It should be noted that equation 26 could also be derived from equation (24) by application of the
condition that the second variation of potential energy be equal to zero for elastic instability.
Variations are taken here with respect to the deformations of adjacent equilibrium position.
As seen from equation (27), the solution of the instability problem requires the determination
of the eigenvalue A. The eigenvalues were determined using a standard eigenvalue subroutine
NR00T.12

COMPARISON ANALYSIS
The validity and adequacy of a new approximate formulation for structural behaviour are
measured by the performance of analyses of problems for which exact or highly accurate solutions
have been derived by alternative means. The formulations derived in this paper enable the
solution of a wide variety of problems. Five separate examples are therefore treated in this section.
In each case the solution accuracy of the present formulation is examined with respect to ideali-
zation refinement. The comparisons are designed in such a manner so as to test for each type of
instability completely independent from the other.
342 ROSHDY S. BARSOUM A N D RICHARD H. GALLAGHER

Torsional buckling
Pure torsional buckling is represented here by the case of an axially loaded uniform-section
member constrained with respect to linear displacements but free to sustain angular displace-
ments 4. The governing differential equation is

with boundary conditions

+(O) = +(L) = 0
for which the exact solution is

Since the critical load is a function of the section properties, a specific choice must be made for
these parameters in an evaluation of the finite element formulation. We choose here a 14WF142
column of length 100 in. For a representation based on a single element (with
u(x)= V ( x ) = W(x)=O, +(O) = 4(L) = 0)
equations (26) give
(0 - P,, c ) -~( b - P,, d)2 = 0
or
(ac -bd) 5 J[(ac-bd)' - (a2- b2)(c2- d 2 ) ]
P,, = (c2-d2)
where
4GKL 4ECw
a=-
30 L
+-
G K L 2ECw
b = --+-
30 L
c = -41, L
30A

d= Ill L
--
30A
This value of P,, (as shown in Figure 2 for the one element solution) differs by 18 per cent from
the correct solution. For two and four elements, the finite element solutions are 0.6 and 0-03
per cent in error, respectively.
Lateral buckling under moment
The problem examined here (Figure 2) is that of a simply supported beam subject to equal
end moments MC,. For simplicity a narrow rectangular beam is considered. In this case the
prebuckling deformations in the x-z plane are suppressed and only the lateral displacements and
twists 4 are left. The governing differential equations in this case are:

with boundary conditions

EI,u"+M,+
GKI$"-M,V" = O

r(0) = u(L)= 4(0)

=0

I
= +(L) = 0
for which the exact solution is
7T
M,, = -J(EI,GK) (32)
L
 FINITE ELEMENT ANALYSIS 343

The finite element representation with a single element gave an error of 10.2 per cent. For
two elements, the finite element solution is 0.3 per cent in error. These results are plotted in
Figure 2.
% Error

11
I

-
Pure torsional
buckling

10

9
Lateral
8 buckling +=0 +=0
A
---a- -- v=o v=o
7
6
5 Lateral
buckling
4 -.+.-
3
2

I 2 3 4 5 6
Number of elements
Figure 2. Convergence of finite element solution for lateral and torsional buckling

Lateral buckling of a cantileoer beam acted upon by a concentrated load at the shear centre
This comparison study tests the element formulation for the case of unequal end moments,
and hence its applicability in more general cases.
The geometric matrix [ N ] is found in the following manner:
1. The bending moment and shearing force distributions are found for a unit load (P, = 1)
at the end of the cantilever beam.
2. The beam is divided into the required number of elements of lengths I, and the moments
and shears Mul, Muz, Qzl and Qa at the ends of each element are calculated.
3. Knowing Myl, My,,Qzl and QS2, the element geometric matrix [n] is calculated, enabling
assembly of the master geometric matrix [ N ] .
Solution of the eigenvalue problem (equation 24) yields the eigenvalues A, which are in this case
the buckling load Pep. The corresponding governing differential equations for this problem are
EI,u"+P*$b(L-x) =0

GK~'-P,(L-x)v'+P,(v(L)-o) = 0 I
344 ROSHDY S. BARSOUM AND RICHARD H. GALLAGHER

with the boundary conditions

v(0) = v'(0) = +(O) =0
and since M, =0 at x = L, therefore
#(L) = 0
Solution of this boundary value problem (Reference 9, 260) is in terms of Bessel functions of
the first kind and will not be cited here. Comparison results are shown in Figure 2.
Lateral buckling of a simply supported beam by a load in the x-z plane
This comparison provides a further study on the applicability of the finite element formulation
in more general cases. Six cases of loading are considered (sources for the governing differential
equations and classical solutions are noted in parentheses).
1. Concentrated load at the shear centre of a narrow rectangular beam (Reference 9, 269).
2. Distributed load at the shear centre of a narrow rectangular section (Reference 9, 269).
3. Concentrated load at the top of an I-beam (Reference 9, 264).
4. Concentrated load at the bottom of an I-beam.
5. Distributed load at the top of an I-beam (Reference 9, 268).
6. Distributed load at the bottom of an I-beam.
The boundary conditions are
u(0) = u(L) = +(O) = +(L)= 0

In each case the element geometric matrix [n] is constructed after finding the end moments and
shears Mgl, My,, Q,, and Q,, of the individual elements due to a unit loading acting on the beam.
The master geometric matrix [ N ] is then assembled for this unit loading.
In loading cases 3 and 4 an additional term, that was not included before, is needed in the
expression for potential energy viz. +P, Cr, $,: where the load P, is acting at a vertical distance Li,
from the shear centre, and is the angle of twist at node i. Taking the variation with respect to
& we get an additional term P,i, in the [ N ] matrix corresponding to the degree of freedom di.
It should be noted that cases 3 and 4 (similarly cases 5 and 6) are found in one operation, since
negative eigenvalues found correspond to a negative ci,, i.e. the load is applied at the bottom
flange. In cases 3, 4, 5 and 6 the I-beam chosen had the constant L2GK/ECw = 4 (Reference 9,
264). It was noticed that with higher values of this constant, a lower percentage error is achieved.
Figure 3 shows the comparison results. The high percentage error in the case of loads not at
the shear centre has previously been discussed at the end of the third section.
Stability of a circular shaft due to a conservative torque
Two types of conservative torques are considered, semitangential torque and quasitangential
torque. A third well-known type of torque, pseudotangential torque, is not considered since it does
not lead to instability.
I . Semitangential torque (Figure 4). In this case the expression for the potential of the torque
is given by equation (12). It was found that the error in the case of a fixed shaft is very high.
The reason for this high error is that the end conditionslO
+
EIw" BM, U' = 0
EIv" - JjM, W' =0 I
could not be satisfied, since we have only first derivatives in our stiffness formulation.
(34)

With 0 = $ = 0 at the ends, while w"#O and u"Z0, equations (34) could not be satisfied,
unless a higher order representation is used. Case 11 has higher accuracy, since it is possible to
 FINITE ELEMENT ANALYSIS 345

O/O Error

=
Ah T.1'iL:
I#J=O
v=o v=o
+=O vf
23 --.+-- Distributed load at the top
(I beam)

I
+Distributed load at the bottom
(I beam

---0--- Distributed load at S.C.

(rect. sec.)

--)- Concentrated load at top

( I beam)
........&..... Concentrated load at the bottom
(I beam)

---+---
Concentrated load at S.C.

1 2 3 4 5 6
Number of elements
Figure 3. Convergence of finite element solution for lateral buckling

satisfy the boundary conditions at one end. This could be easily seen by integrating equation (34)
once.
In cases 111, IV and V of Figure 4, where one end is free or one or both ends are hinged
( O # O , $#O), the second derivative can take some value and does not have to be identically zero.
Therefore in cases 11, 111, IV and V of Figure 4, a higher accuracy is achieved.
2. Quasitangential torque (Figure 5). In this case the potential of the torque is given by
equation (12), plus an additional term (Reference 10, 106)
+ W,(&,w;L)- $0, (35)
This additional term in the potential of the torque produces, for a unit torque, two off diagonal
terms equal to - & corresponding to Oto, and I,+~,
at the top of [ N ] and another two off diagonal
terms corresponding to 80,,y5(L) at the bottom of [ N ] . Cases I and 11, Figure 5, give the same
result as in the case of semitangential torque, because the boundary term, equation (35), is zero
in these cases.
346 ROSHDY S. BARSOUM A N D RICHARD H. GALLAGHER

t
/'a Error

52

T
12

10 Semitangential
torqu,e
8

Number of elements
Figure 4. Convergence of finite element solution for a conservative torque (semitangential)

CONCLUSIONS
Stiffness expressions have been developed for torsional and lateral instability analyses of uniform-
section prismatic elements. Solutions were obtained to such problems by use of these element
relationships and the results were compared with classical solutions.
All results converged to the classical solutions as the analysis grid was refined. Rate of con-
vergence and absolute accuracy were excellent for phenomena governed principally by the flexural
and lateral instability modes. These measures of solution efficacy were less satisfactory for cases
where the torsional mode predominated. This factor stems from the use of a functional repre-
sentation for torsion which does not satisfy the basic governing differential equation.
The formulations derived herein should find applicability in the treatment of non-uniform
and continuous beams for the subject type of instabilities. Torsional and lateral instability
solutions for these types of structures are few in number and difficult in application. The derived
formulations can also be employed as the basis for study of corresponding instability phenomena
for planar and space frameworks, although here it must be cautioned that the treatment of
general stability phenomena for this class of problem will require extension of the presented
relationships.
 FINITE ELEMENT ANALYSIS 347

% Error

Quasitangential
torque

2 3 4 5 6
Number of elements
Figure 5 . Convergence of finite element solution for conservative torque (quasitangential)

ACKNOWLEDGEMENT

The authors wish to acknowledge the support provided by the National Aeronautics and Space
Administration under Grant No. NGR 33-010-070.

APPENDIX I
To derive expressions (7-1 l), consider Figure 6 , in which an element is acted upon by the forces
9, Ql,Q,, MI and M,. The potential of these forces could be easily obtained by considering
Figure 7(a), where the node 1 is fixed. Thus, we consider the work done by 9,Q2 and M2 due to a
lateral deformation of a differential element (i) of length dx. The deformation of the element (9
is decomposed into a bending in the x-y plane, Figure 7(b), and a rotation 4, Figure 7(c). From
Figure 7(b) the horizontal displacements of points 2 and m are given by
d2u
db2 = --(/-x)dx (36)
dx2
and
d2v
dbT,,= -- [dx (37)
dx2
348 ROSHDY S. BARSOUM A N D RICHARD H. GALLAGHER

Figure 6. Element forces

4d€
I I

Pre buckling
position

I I rn I (a 1
I
A
T
1
T
2
I
I

i 1

Bending of (b)
differential element ( i )

I I I
I I
I I

Final deformation
2"

1
1
Plon

Figure 7. Deformation of a differential element

 FINITE ELEMENT ANALYSIS 349

From Figure 7(c); where a twist 4 of the differential element (i) occurs; the vertical displacements
of points 2 and m are given by

and

Due to both the above deformations the end face 2 will rotate about the y-axis by an angle
d$, shown in Figure 7(d).

Hence the work done by Q,, M, and the load q d f at a point (rn) is given by*

and

and for the distributed load from point (i) to point 2, the work done is

d2v
= - $q 7+(/- x)' dx (43)
dx
The total work done due to the deformation of the whole length from 1 to 2 is given by

In order to account for unequal end moments, we have to consider the case where point 2 is
fixed and point 1 is free, as in Figure 8. Using the same arguments, we could write the work
done by the forces q, Q, and Ml as

The only difference between expressions (44) and (45) is the substitution of x for (I-x) and
vice versa, and the change of sign of the second term, which is obvious from Figure 8.

* The subscript S.C. refers to the fact that the loads are applied at the shear centre, and 2 indicates that point 2
is free.
350 ROSHDY S. BARSOUM AND RICHARD H. GALLAGHER

It is clear that VN,c,.l=V'N.c2 for a bar in equilibrium under the same loading, but changing the
idea of which end is free and which end is supported.

dGl I
\
Elevation

Figure 8. Deformation of a differential element

Therefore the work done by all the forces q, Q,, Q,, Ml and M2 due to a general motion of the
element (1-2) is given by:
 FINITE ELEMENT ANALYSIS 35 1

APPENDIX I1
Element stiflness expressions

lkl =

where

["I z -

where
352 ROSHDY S. BARSOUM AND RICHARD H. GALLAGHER

0 = M, /I

R = M,, /2

REFERENCES
I . 0. Zienkiewicz and Y. Cheung, The Finite Element Method, McGraw-Hill, New York,1967.
2. J . Przemieniecki, Theory of Matrix Structural Analysis, McGraw-Hill, New York, 1968.
3. R. Mallett and P. Marcal, ‘Finite element analysis of nonlinear structures’, J . struct. Div. Proc. Am. SOC.
civ. Engrs, 94, ST9 (1968).
4. R. H. Gallagher, ‘A survey of finite element framework stability analysis, Am. SOC.civ. Engrs, National
Meeting, New Orleans (1 969).
5. J. D. Renton, ‘Buckling of frames composed of thin-walled members’, in Thin-walled Structures (Ed. A. H.
Chilver), Wiley, New York, 1967.
6. A. Chajes and G. Winter, ‘Torsional-flexural buckling of thin walled members’, J. struct. Div. Proc. Am.
SOC.civ. Engrs, ST4 (1965).
7. V . Z. Vlasov, Thin-walled Elastic Beams, 2nd edn., 1959. English translation published for U.S.Science
Foundatation by Israel Program for Scientific Translations, 1961.
8. F. Bleich, Buckling Strength of Metal Structures, McGraw-Hill, New York, 1952.
9. S. P. Timoshenko and J. M. Gere, Theory of Elastic Stability, McGraw-Hill, New York, 1961.
10. H. Ziegler, ‘Stabilitatsprobleme bei graden Staben und Wellen’, Z A M P , 2, 265 (1951).
1 I . H. Ziegler, ‘Knickung gerader Stabe unter Torsion’, Z A M P , 3, 96 (1952).
12. M. Beck, ‘Knickung gerader Stabe durch Druck und Konservative Torsion’, Zng. Archiv, 23, 231 (1955).
13. Anon., ‘System/360 Scientific Subroutine Package (360A-CM-03X) Version 11. Programs Manual’, ZEM
Appl. Program Man., H20-0205-2 (1967).