Nonlinear finite element analysis of three-dimensional frames

M. SAVARD
Direction rles strlictllres, Mirzisttre rles T ~ N I Z S ~ O
d~
I .Q
iI S~ i t b e c ,Q~ie'Dec,QC G I K 5Z1, Crrrzrrdn
AND

D. BEAULIEUA N D M. FAFARD
Diprrrterr~etztcle girzie civil, Utziversite' Lnvnl, Saitzte-Fay, QC G I K 7P4, Crrrzrrdn

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

Received March 25, 1991

Revised manuscript accepted November 9, 1993
This paper presents a numerical model for the nonlinear analysis of three-dimensional frames using the finite
element method. The model is based on a general variational formulation for thin-walled beams with open or closed
sections; the mathematical development uses an updated Lagrangian description for shallow arches, taking into
account geometrical and material nonlinearities, residual stresses, member initial imperfections, warping, and connection flexibility. The model has been used to evaluate second-order effects and the influence of residual stresses
and joint flexibility on the behavior of plane frames. A se~nirigidspace frame has been analyzed and the gain in rigidity
offered by a horizontal rigid diaphragm has been evaluated.
Key buorr1.s: analysis, connection, finite element, nonlinearity, residual stress, thin-walled beams, warping.
Cet article prCsente un modele numCrique d'analyse non IinCaire par ClCments finis permettant de simuler le
co~nportementdes cadres mktalliques tridimensionnels. Ce modkle est bas6 sur le dCveloppement d'une formulation
variationnelle gCnCrale des poutres i parois minces I? sections ouvertes ou fermkes, en description lagrangienne
actualisee. De plus, le modele permet de tenir compte de la non-IinCaritC gComCtrique, de la plasticitC, des contraintes
rCsiduelles, des dCfauts de rectitude des pieces, du gauchissement des sections et de la rigidit6 partielle des assemblages. Ce modele a CtC utilisC pour Cvaluer I'importance des effets du deuxikme ordre et des contraintes rCsiduelles
ainsi que I'influence de la rigidit6 imparfaite des assemblages pour des cadres plans. De plus, nous avons pu
apprCcier I'apport de rigidit6 qu'offre iln diaphragme rigide horizontal dans u n cadre rigide simple tridimensionnel.
Mots cle's : analyse, assemblage, contraite rCsiduelle, ClCment fini, gauchissement, non-IinCaritC, poutres ii parois
minces.
Can. 1. Civ. Eng.

21. 461-170

(1994)

Introduction
Design considerations for inembers in three-dimensional
frames include a good knowledge of how these members
will behave under the applied loads. Thus, it is essential to
adeq~~ately
evaluate the intensity of each member's internal
forces. For this purpose, a complete analysis of the structure
should take into account the following factors: combination
of axial and shear forces, along with bending and twisting
moments; P-h effects (geometric nonlinearity); elastoplastic
behavior of the material; connection flexibility; residual
stress distribution; and contribution of rigid diaphragms to
the structure's overall rigidity.
To perform the analysis in question, the numerical model
developed by Akoussah et al. ( 1987) was generalized by
introducing new section geometries, i.e., rectangular, circular,
open, and closed sections. Also, a new finite element was
developed to simulate the nonlinear behavior of flexible
connections.
The numerical model was validated using several examples,
each one covering a particular aspect of structural behavior.
Example 1 checks the proposed model's accuracy for sim~ ~ l a t i nflexibly
g
connected frames. Example 2 compares our
numerical results with experimental measuiements taken on
a plane frame. Example 3 studies the influence of a horizontal
diaphragm in a one-story space frame.

Mathematical development
Upon application of a load, the stress to which a solid is
subjected causes a state of equilibrium between external
and internal forces. This equilibrium can be determined
using the weighted residual method and the principle of
virtual work. The result of integrating over the solid's volume
a weighting function (+) multiplied by the differential equations
of equilibrium nus st equal zero.
Using a weighting function composed of virt~ialdisplacements kinematically admissible within predefined boundary
conditions, we can determine the variational formulation of
the elastic problem in terms of the solid's unknown configuration. The choice of an updated Lagrangian description
makes the resolution possible. The Lagrangian description is
termed ~ipdatedwhen the integration domain is defined as the
last equilibrium config~~ration
determined.
We have used C' to represent the known equilibrium configuration (or current configuration) at the beginning of a
load step and w', to represent its corresponding variational
form. The weak integral form of the variational formulation
of the elastic problem of the current configuration may be
expressed as (Fafard et al. 1989)

NOTE:Written discussion of this paper is welcomed and will be

received by the Editor until October 31, 1994 (address inside
front cover).
I'r~olcdin Can;~d;t1 Impilur6 ;m C:tn;td;t

where (&*I)is the Green-Lagrange first variation deformation

tensor, (s') is the second Piola-Kirchhoff stress tensor, ( u * )
is the virtual displacement vector, (fd) is the force per unit

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

C A N . J . CIV. ENG. VOL. 21. 1994

FIG. 1. Member cross section: (a) typical cross section; (b) displacements in the plane of the cross section.
of volume vector, and (f,l) is the external force per unit of
surface vector. Equation [ l ] completely defines the equilibrium conditions for a solid to which surface and volume
loads are applied.
To evaluate the unknown configuration c', or the increment
of displacement { u : ) that occurs when the load is increased
from configuration C 1to configuration C2,a limited Taylor
series of the integral form W: was developed. The first two
terms of this series are
[2]

W:= W ;+

+ ... = 0

W;-I

To solve this equation, the Newton-Raphson method was

used with matrix triangularization at each iteration. This
iterative method lends itself well to an elastoplastic analysis
requiring knowledge of the change in stresses and deformations at each loading step.
Kinematics of open thin-walled beams
Details on the developments in this section can be found
in Akoussah et al. (1987), Vlassov (1961), and Nishino et al.
(1973).
The following relations describe the displacements u, v, and
w for open and closed sections along the X, Y, and Z axes
respectively. Shear and axial forces, bending and twisting
moments, and cross-sectional warping can be developed
simultaneously.
Based on the cross section of a thin-walled member shown
in Fig. 1, the following can be assumed:
- Assumption 1. Cross sections d o not distort in their
own plane.

A slight angle of twist, using the shear center of coordinates z, and yo as the center of rotation, and point o as the
origin of the coordinate system, results in
[41

v(x, s) = v,(x)

(z

z,)@(x)

To evaluate the axial displacement 11 and the corresponding

deformation yn,?,the following must be assumed:
- Assumption 2. Shear strains due to shear stresses in
equilibrium with the changes of normal stresses are negligible.

- Assumption 3. Shear strains in the planes normal to

the middle surface of thin walls are negligible.
Using the equilibrium of the longitudinal forces on a differential portion of the wall, we obtain (Nishino et al. 1973)

[6]
[71

av - z-aw
ax
ax

~ ( x s)
, = u, - y-

y,,-a-

w-

a@

ax

a@

ax

where @ is the angle of twist. The warping function, w, and

the coefficient a can be evaluated using [8] and [9] for open
sections and [ l o ] and [ 111 for closed sections.

[8]

w =

[9]

a=-25

r, ds

T h e vectorial form of the displacement field may b e

expressed as

The kinematic description of any point on the beam cross

section can be obtained by projecting the Cartesian displacements in the curvilinear base (x, s , 5) in order to use the
beam's properties and to introduce the Marguerre approach
to take into account the flattened beam theory (Akoussah
et al. 1987). We have called v, and v 5 the displacements
along the s and 5 axis respectively (Fig. la). The kinematic
displacements of point q can be described using the limited
Taylor series developed around point p of coordinates x, s ,
and 5 = 0 on the middle surface of the cross section shown
in Fig. 1. Based on the Kirchhoff-Love assumption:

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

S A V A R D ET A L

FIG.2. External forces acting on a member.

I.1
-

[16]

( Y+ 1 5 ),~, - ( Z + tn5)vv,,,+ (o- ~ r ; , ) @ , , ~ ]

mv, - lvv, - (r, +

(d,) = (u, v l

W,

@ , oYl oZl @; u2 v2 w2 @2 0).2 0 : ~@;)

With an updated lagrangian formulation, the expression

of the internal virtual work evaluated for the configuration
C' for each elementary subdomain becomes (Akoussah et al.
1987)

c)@

ll~,- 177W,

+ ti,@

where
I;, = m(y

Y,,) - l(z

2,)

'b= l(y - y o ) - m(z

z,)

In these equations, 1 and m are the cosine directors between

the (y, z ) and (s, 5) coordinate systems. Equation 1131 completely describes the real and virtual kinematic fields of an
arbitrary point on the beam cross section in the curvilinear
system (x, s, 5). This equation is used in the development of
the expressions of the internal virtual work, the external
virtual work, and the tangent stiffness matrix necessary with
the iterative procedure chosen for the resolution. These are
given in Appendix 2.

Finite elements
Bent~ieletnent
If a one-dimensional cubic element is selected for bending
and torsion and a one-dimensional C? linear element is chosen
for the axial force, the displacement vector { u } can be
approximated by the following functions (Akoussah et al.
1987):

where {d,} is the nodal displacement vector:

where { M f} represents the resultant force vector integrated

over the entire section, and [B,] represents the degrees of
freedom-to-deformation transformation matrix.
The elementary external virtual work in the current configuration may be expressed as

where

is the vector of the external forces acting on the thin-walled

member. Figure 2 illustrates these forces in their positive
sense of action.
The rate of the internal virtual work may be expressed as

with
[201

[K,l = [K,,I + [Kg]

[221

[Kg] =

jL, ([B,,, I ~ I D ~ ~ I I)[ Bd, ,~~'

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

464

C A N . 1. CIV. ENG. VOL. 21, 1994

The matrix [DO11 takes the constitutive elastoplastic law of

combined stresses (axial and torsional shear forces) into
account. The mixed formulation was used for the variational
principle to avoid membrane locking (c' interpolation for
membrane displacements; Akoussah et al. 1986). The complete
expressions for [B,,,], [B,,,], [Doll, and [DO21 can be found in
Akoussah et al. (1987).
Equations [17] and [19] are expressed using the local coordinate system of an element for which the displacements are
assumed to be small (no large strain). This assumption allows
us to use the Eulerian equation for the virtual work in the
local system (no Lagrangian variable). The domain of integration used is the last known configuration in equilibrium.
If a sufficiently large number of integration points spreaded
over the cross section of the member is used, residual stress
and shear lag will be adequately modeled. Of course, distorsions and related effects cannot be simulated by a beam
element. To correctly account for these effects, a platelshell
element would be required.
To account for large displacements and large rotations, a
transformation matrix calculated with the updated nodal
coordinates of the element was applied to transform the local
residual vector in the global system.
Therefore, the Newton-Raphson procedure yields (Bathe
1981)

where (li) is the relative displacement vector of the connection:

[26]

(li) = (ii

i, iv

8,

8, 8;)

The residual vector {R(ii)}represents the unbalanced forces

in the connection, i.e., the difference between external and
internal forces. Since the external loads applied to a connection
are generally nul, the residual vector may be interpreted as the
vector of the internal forces at each end of the connection.
If K,, K,, and K, are the internal forces in the connection
corresponding respectively to the unit relative displacements
2, C, and iv at both ends of the connection, and K,,,, KO,,and
KO,are the bending moments corresponding to the unit relative
rotations 8,, 8,, and 8, at the ends of the connection, the tangent stiffness inatrix of the connection may be expressed as

Once again, a limited Taylor series development of the

integral form allows us to determine the unknown equilibrium
configuration C' from the previously determined configuration C1, giving us

W:

(~~)([K,(U')]{AU'}
- {~(u')})= 0

The vector ( ~ ( u ' ) )is the equilibrium residual vector of the

entire structure for iteration i.
The standard Newton-Raphson method can be used to
solve [23] for {Ale'}. The correction is then added to the
correct displacement vector evaluated at the previous iteration:

In order to translate this last equation (written as a function

of the relative displacements) in terms of the nodal degrees
of freedom, the following transformation must be performed:
[291

{ f i l , ~=, [ B ~ ~ ] , x( I ,~ 1 41x 1

where
[-I0

0 0 0 0 0-100

0 0 0 01

The calculations are continued as long as the ratio of the

norm of the correction vector to the norm of the displacement
vector is less than a given value (generally 0.001).
Finite elerrierzt approach for flexible corlrzectiorls
To take the flexibility of the connections into account, a
three-dimensional finite element having as many degrees of
freedom as the beam element presented above was developed.
The rigidity matrix for this connection element is constructed
based on the slope of the force-displacement curves and
moment-rotation curves representing the behavior of the
connection. These curves are required as model input data for
each one of the first six nodal degrees of freedom. The seventh
degree of freedom which accounts for cross-sectional warping
is considered unrestrained in the connection model.
The mathematical approach used to simulate the behavior
of the connection is very similar to the beam element
approach. By means of the principle of virtual work and
the weighted residual method, the equilibrium configuration
can be established. The integral form is written as (Savard
et al. 1989)

The rate of the integral form becomes

where

The tangent matrix [K,(u)l defined above can be directly

assembled with the tangent inatrix of the beam elements.
The global tangent matrix is therefore influenced by the
presence of the connections.

Connection offsets
A structure is generally represented by lines passing through
the center of gravity of its members. It is assumed that the

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

S A V A R D ET A L

columns : W250x73

Rotation, 8 ( ~ 1 0 rad)
'~

FIG.3. Semirigid plane frame: (a) geometry and loading;

(13)

finite element model; ( c ) moment-rotation curve for the connections.

structural properties of the members are valid over the entire

length of these lines. Consequently, a beam whose ends are
to a column flange is considered to have uniform properties
over its entire length even though it is recognized that the
end portions, equal to half the depth of the respective columns,
are much more rigid.
To prevent this slight anomaly which could affect the
behavior of the structure when columns are deep and heavy,
we followed the recommendations of Nethercot et al. (1987),
who suggested introducing a rigid panel, whose length equals
half the column depth, at each end of the beam. The flexural
rigidity in these portions of a beam is closer to the flexural
rigidity of a rigid panel than to that of a beam.

Numerical examples
The beam finite element presented here may be used to
analyze two- or three-dimensional structures. The element's
characteristics may be specified singly or simultaneously.
Therefore, it is relatively easy to adjust the model to the
characteristics of other models used in com~arativestudies.
The first example is based on a three-story three-bay plane
frame with flexible connections. A simple plane frame is
analyzed in the second example and our numerical results
are compared with experimental measurements. In the third
example, we analyze a one-story space frame with a rigid
horizontal diaphragm.

Two-dimensional semirigid frame exat~ll~le

To demonstrate the accuracy of the proposed finite element
model for elastic frames with semirigid connections, we
compared our results with those of Brun and Picard (1976).
The geometry and loads acting on the structure are shown
in Fig. 30. The same frame was analyzed by Adams (1968)
in his investigation of P-A effects.
A uniformly distributed load of 40.86 kN/m was applied to
the frame on each floor along with three concentrated horizontal loads, as shown. The finite element model is presented
in Fig. 30 and includes three beam elements for each member.
The connections are represented by springs at the beam ends.
Rigid panels have been included in the analysis where beams
are connected to column flanges.
To represent the behavior of a connection, Brun and Picard
(1976) used a standardized experimental moment-rotation curves
and found the following equation to be representative of the
typical moment-rotation curve for the type of connection used:
[33]

6=

1.04 X lo-' (KM)

1.62 X 10-I' (KM)"

+ 4.62

~o-'"KM)'

In [33],6 is the rotation of the connection, M is the bending

moment, and K is a factor accounting for geometric parameters of the connection. This function is illustrated in Fig. 3c.
In our numerical model, this curve is subdivided into a series

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

466

CAN.J. CIV. ENG. VOL. 21, 1994

Horizontal displacement, A ( mm )
FIG. 4. Sway of the rigid plane frame.

Horizontal displacement, A ( mm )
FIG.5. Sway of the semirigid plane frame.
of linear segments: eight segments were sufficient to correctly
represent the moment-rotation curve shown in the figure.
The results of the frame analysis considering rigid connections are given in Fig. 4. The finite element results are
very similar to those of Brun and Picard. To show the magnitude of the P-A effects, the results of the analysis with
geometrical nonlinearities included have been plotted in the
same figure. For the frame studied, the P-A effects increased
the horizontal displacement, A by 25-30%.
The results of the frame analysis considering semirigid
connections are illustrated in Fig. 5 and show the horizontal
displacement, A, to be about 80% higher when the connection
flexibility is accounted for in the analysis than the results for
the rigid structure (Fig. 4). As expected, the horizontal
displacements are slightly lower when connection offsets
are considered.
When semirigid connections are taken into account, our
results differ up to 5 % from those of Brun and Picard. This

rather small difference can be explained by the fact that the

algorithm used by Brun and Picard to account for the nonlinear
behavior of connections differs from ours as do the convergence
criteria.
Orze-story plane frarne exnirlple
A one-story one-bay frame loaded with four concentrated
vertical loads symmetrically applied to the beam is shown in
Fig. 6. The horizontal load H was the only one varied during
the experimental test, as described by Arnold et al. (1968).
The purpose of this numerical test was to evaluate and compare the horizontal displacement of the point of application
of load H in order to calibrate the finite element model.
Load H was gradually increased while all other loads
remained constant.
The finite element representation in Fig. 6 0 comprises
four beam elements for each column and five for the beam.
In addition, connection offsets and flexible connection elements

S A V A R D ET AL

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

beam : S 250 x 38
columns : W 130 x 28

FIG.6. Experimental plane frame: ( a ) geometry and loading; (b) finite element model; (c) residual stress distribution; (d) stress-strain
relationship for steel.

no residual stresses
Finite element results;
flexible connections;
no residual stresses
A Finite element results;
flexible connections;

20

40

60
80
100
120
140
Horizontal displacement, A ( rnm )

160

180

FIG.7. Load-displacement relationship - experimental plane frame.

were introduced at each end of the beam. Since the momentrotation relationships for the connections were unknown,
we assumed that the curve used in example 1 could be
applied to the beam-to-column connections and assumed
perfectly rigid base connections for the columns.
The residual stress distribution considered over the length
of each member is shown in Fig. 6c. The maximum residual
stress was taken to equal 30% of the material yield stress, as
is usually recommended. The standardized stress-strain
relationship for steel used in the analysis is shown in Fig. 6d.
The load sequence may significantly affect the results. The
load sequence we considered was essentially the same as the
experimental one and consisted in applying the four vertical
loads and then gradually increasing the horizontal load H.
The results obtained for both geometrical and material
nonlinearities are plotted in Fig. 7 along with the experimental

measurements. First, we observed that connection rigidities

influence the behavior of the whole structure. The ultimate
load for the perfectly rigid frame is 8-10% higher than the
corresponding value for the semirigid frame. Residual stresses
seem to have very little effect on the frame's overall response.
With the flexible connection elements, o u r numerical
model suggests a frame response very similar to the one
measured experimentally until the horizontal displacement,
A, reaches 80 mm. Past this point, the model seems to overestimate the displacement. Many factors can explain this
difference. First, the horizontal displacement of 80 mm corresponds to the frame's ultimate load. Past this limit, the
horizontal load must be reduced in order to maintain equilibrium. Plastic deformations may have occurred and the
stress-deformation curve shown in Fig. 6c1, as well as the
moment-rotation curve considered for the connections and

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

C A N . J. CIV. ENG. VOL. 21, 1994

P=476kN

,,,,

,,,,,,

beams: W250x73
columns: W200x46

1
-3048 1

FIG.8. One-story space frame: (a) model overview; ( 0 ) plane view; (c) elevation

loo

10

15

20

25

30

35

Displacement of node 3 along the x axis ( mm )

FIG.9. Load-displacement relationship
the residual stress distribution, may prove inadequate. Also,
rotations may have occurred in the base connections, which
are considered perfectly rigid in the numerical model. Other
factors, such as initial geometric imperfections and out-ofplane displacements could also have affected the frame response.
One-story space fi-a~neesalnple
The main objective of this example was to verify the
model's validity and performance in simulating the behavior
of three-dimensional structures composed of linear members
and rigid diaphragms. To do so, the space frame studied by
Wynhoven and Adams (1970) was reanalyzed.
As shown in Fig. 8a, the finite element model includes
flexible connections and rigid panels wherever a beam is
connected to a column flange. Four beam elements were
used to model each member. The elastic limit, F,, considered
in analysis is equal to 238 MPa.
A vertical load, P, of 476 k N was applied at the upper
end of each column of the frame. These loads were held

one-story space frame.

constant while two horizontal loads were gradually applied,

one at node 3 and the other, four times lower, at node 4.
The results are presented in Fig. 9 where the intensity, H, is
plotted against the displacement of node 3 along the .r axis.
First, we performed a simplified analysis which assumed
all connections to be rigid, no residual stresses and perfect
elastoplastic behavior for the steel. These approximations
were also made by Wynhoven and Adams (1970). In this
case, both models yielded very similar results. The initial
slopes are the same and there is a mere 2% difference
between the ultimate loads evaluated by the models. When
the residual stress distribution shown in Fig. 6c is considered
along with the moment-rotation curve shown in Fig. 3c for
the connection, the ultimate load is the same as in the simplified model, but the initial slope is slightly lower.
The above results were obtained when nodes 1 to 4 were
free to move with respect to one another. In order to evaluate
the frame's behavior when these relative displacements are
constrained, we included a horizontal diaphragm simulating

469

S A V A R D ET AL.

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

a rigid roof in the upper part of the f r a m e . T h e rigid

diaphragm was modeled with two six-node triangular shell
elements (Fafard et al. 1989). The results are shown in
Fig. 9. The load-displacement curves show an initial slope
that is significantly steeper and an ultimate load that is 15%
higher than in the previous case. This expected behavior
tends to validate the model and illustrates the importance
of considering horizontal rigid diaphragms in the analysis of
space frames.

Conclusions
The results obtained with the new beam and connection
elements presented in this paper show that the proposed
finite element models are adequate for simulating of the
behavior of two- and three-dimensional frames under the
action of an increasing load, taking into account geometrical
and material nonlinearities, connection flexibility, residual
stresses, and cross-section warping.
The lack of data on published experimental tests inay justify
the differences observed between numerical and experimental
results. To determine a structure's ultimate state of loading,
a good approximation of the real moment-rotation curve of
the connection and column bases is needed, along with initial
member imperfections and out-of-plane displacements. The
proposed numerical model can also take into account parameters that are difficult to measure. As soon as precise measurements are available, it will be possible to include them
adequately in a numerical analysis.
Finally, the fact that the different characteristics of this
numerical model can be specified singly or simultaneously
is very useful for comparative studies. The model allows
evaluation of the validity and performance of simplified
analyses designed for special applications. It can also be a
valuable tool to improve actual design specifications or help
develop new ones.
Acknowledgments
T h e authors wish to thank the Natural Sciences and
Engineering Research Council of Canada, the Fonds pour
la formation de chercheurs et l'aide i la recherche of the
Quebec Ministry of Education, and the Canadian Steel
Industries Construction Council for their financial contribution
to this project.
Adams, P.F. 1968. The design of steel-beam columns. Canadian
Steel Industries Construction Council, Willowdale, Ont.
Akoussah, E., Beaulieu, D., and Dhatt, G. 1986. Curved beam
element via penaltylmixed formulation for nonlinear in-plane
analysis. Com~nunicationsi n Applied Numerical Methods, 2:
6 17-623.
Akoussah, E., Beaulieu, D., and Dhatt, G. 1987. Analyse no11
IinCaire des structures B parois minces par ClCments finis et
son application aux biti~nentsindustriels. Report GCT-87-07,
Department of Civil Engineering, UniversitC Laval, SainteFoy, Que.
Arnold, P., Adams, P.F., and Lu, L.W. 1968. Strength and behavior
of an inelastic hybrid frame. ASCE J o ~ ~ r n of
a l the Struct~~ral
Division, 94(ST1): 243-266.
Bathe, K.J. 1981. Finite element procedures i n engineering
analysis. Prentice-Hall Inc., Eqglewood Cliffs, N.J.
Brun, P., and Picard, A. 1976. Etude d'un assemblage imparfaite~nentrigide et des effets de son utilisation dans un multiCtagC. Report GCI-76-03, Department of Civil Engineering,
UniversitC Laval, Sainte-Foy, Que.

Fafard, M., Dhatt. G., and Batoz, J.L. 1989. A new discrete
Kirchhoff platelshell element with updated procedures.
Computer and Structures, 4: 591-602.
Malvern, L.E. 1969. Introduction to the mechanics of a continuous
medium. Prentice-Hall Inc., Englewood Cliffs, N.J.
Massicotte, B., and Beaulieu, D. 1984. Effct des systkmes
entremises-diaphrag~nessur la stabilitC des poteaux. Report
GCT-84-05, Department of Civil Engineering, UniversitC Laval,
Sainte-Foy, Que.
Nethercot, D.A., Kirby, P.A., and Rifai, A.M. 1987. Colu~nnsin
partially restrained construction: analytical studies. Canadian
Journal of Civil Engineering, 14(4): 485-497.
Nishino, F., Kasemset, G., and Lee, S.L. 1973. Variational formulation of stability problems for thin-walled members. IngenieurArchives, 43: 58-68.
Savard, M., Beaulieu, D., and Fafard, M. 1989. Mod6lisation
par ClCnlents finis des bfitiments B ossature mCtallique. Report
GCT-89-05, Department of Civil Engineering, UniversitC Laval,
Sainte-Foy, Que.
Vlassov, V.Z. 196 1. Thin walled elastic beams. Translated from
Russian. Israel Program for Scientific Translation, Jer~~salem,
Israel.
Wynhove, J.H., and Adams, P.F. 1970. Elastic-plastic analysis
of three-dimensional structures. Struct~~ral
Engineering Report
No. 29, Department of Civil Engineering, University of Alberta,
Edmonton, Alta.

171,

Appendix 1. List of symbols

degrees of freedom-to-deformation transformation matrix
relative displaceinents-to-nodal degrees of
freedom transformation matrix
nodal displacement vector
differential element of surface
differential element of volume
virtual deformation rate tensor
surface forces
volume forces
distributed forces along s, y , and z axes
center of gravity
tangent stiffness matrix
distributed moments along s, y , and z axes
interpolation fi~nctions
shear center
equilibrium residual vector
second Piola-Kirchhoff stress vector
displacements along s,y, and z axes
virtual displacement vector
increment of displacement vector
relative displacement vector
integral form
internal virtual work
first derivative for x
second derivative for x
coordinates of the shear center
weighting functions
increase or variation
Green-Lagrange first variation deforination
vector
axial strain along s and 5 axes
angle of twist around longit~idinalaxis
shear strain in the s-5 plane
Cauchy stress tensor
warping function

470

CAN. I. CIV. ENG. VOL. 21, 1994

Can. J. Civ. Eng. Downloaded from www.nrcresearchpress.com by University of Laval on 01/23/13

For personal use only.

Appendix 2
Virtual work theory
The internal virtual work may be expressed as (Malvern
1969)

where [D:" is the virtual deformation rate matrix and [u]

the Cauchy stress tensor. In order to introduce assumptions 1,
2, and 3 into the internal virtual work expressions, [ A l l
must be transformed in the curvilinear system (x, s, 5) shown
in Fig. 1. This transformation accounts for the curvature of
the flattened beam. The Marguerre theory was used with the
(?,)', (z,,y,,) 4 1. If we also assume that
assumption that by,)',
the section is composed of a series of assembled flat elements,
the internal work may be expressed as follows after integration on the cross section (Akoussah et al. 1987):

section, and f, and f; are applied at a distance dy and dz

from the shear center, as shown in Fig. 2.
Tangent stiffness matrix
The tangent stiffness matrix needed to be defined to solve
the nonlinear problem. This matrix may be expressed as
(Akoussah et al. 1987)

where

+ rv:Aw,,

[A7n] q t , = v:Av,,

[A7b] (q;) = (-(v:A@

+ @*A",,)

(@+
.I' A@

where

(w;A@

+ @'2Aw,,r))

+ @*[email protected]'

)(

,,v

A@ ,.r

))

and
[A801 AN =

jA AS,

dA

[A861 (AM) = (AM? AM; AT,, AM,)

[A401 (M) = (MI,M, T,,, M,)

AM; =

(y + li)AS,,

,.I

AM, =
M, =

jA (

, dA

The resulting force N and moments (M) are defined by

[A4a] and [A4b]. The combined influence of torsion and
flexion on the virtual axial strain can be observed in [A301
and on the virtual flexural strain in [A3b].
External v i r t ~ ~ work
al
Using the Cartesian system of coordinates as the kinematic
variables, the external virtual work, in the absence of volume
forces, may be expressed as
[A51

+ (f;dy - fydz)@*
KXt
= j L ( f r > f ,v*<, + f..
+ m,@* + m,w':, + nz-v;,,)
11

(o - ~ 5 ) A s . ~dA
.~

where u is evaluated using [9] for an open section and [ I l l

for a closed section.
Also,
[A91

(V) = (V, V,)

v z = - JA
[A101

l r r s ~ ;

v1 = j A~ n r , , d ~

= ((My - z,N) (M, - Y,N)

q,

MP)

M, = JA[((y + 15) - yo)' + ((z + mi) - z012]u.I'.r dA

W:

dL

Loadfr is applied at the center of gravity of the beam cross

TI

I, / ; L ~ X S