Thin-Walled Structures 82 (2014) 1–12

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

A model for warping transmission through joints of steel frames

S. Shayan, K.J.R. Rasmussen
The University of Sydney, Australia

art ic l e i nf o a b s t r a c t

Article history: A simple approach is developed in this paper which considers the effect of partial warping continuity
Received 26 July 2013 through the joints of thin-walled steel frames when using beam finite element analysis. Using a
Received in revised form condensed stiffness matrix for the joint generated by the substructuring technique, warping springs are
23 March 2014
introduced to represent the condition of partial warping restraint at intersections between members.
Accepted 24 March 2014
The performance of the proposed model is demonstrated through a number of numerical examples.
Excellent agreement is achieved between the results of beam finite element analysis using the suggested
Keywords: joint model and accurate shell finite element analysis.
Thin-walled structures & 2014 Published by Elsevier Ltd.
Connections
Warping
Warping spring
Substructure
Finite element analysis

1. Introduction In subsequent studies, the focus shifted from isolated members to

frames by realising that at joints, warping displacements in one
Thin-walled steel frames with slender open-section members member may redistribute and produce warping and twisting in
may experience significant torsion and therefore large warping other connected members. This implies that when one member
displacements under applied loads. The warping deformation is warps, the flanges of adjoining members must rotate, causing a
generally defined as the longitudinal displacement caused by distortion of the cross-section. Thus, the resistance of adjoining
torsion. For a doubly symmetric I-section, the warping displace- members to distortion provides a level of restraint on the warping
ment consists of linear longitudinal displacements in the flanges in torsion of the loaded member. Several experiments were conducted
opposite directions. to determine the warping restraint at member ends [16,17]. All
The boundary conditions for warping at the ends of an isolated experiments reported the difficulties of restraining warping and
member can be divided into three main groups: completely free, demonstrated that even very stiff end connections do not provide
fully restrained or partially restrained. The most important con- full torsional warping restraint. Consequently, the concepts of
tributions in this context have been made by Timoshenko [1,2], continuous warping and partially restrained warping were
Wagner [3] and Vlasov [4] who studied the warping (or non- introduced.
uniform) torsion of I-beams and derived a general theory for thin- Austin et al. [18] studied the subject of elastic end warping
walled members. In the numerical implementation of the theory, restraint but no information was given to evaluate the degree of
many researches [5–7] introduced the first derivative of the twist restraint. Trahair [19] introduced the ratio between the elastic flange
rotation as the seventh degree of freedom to represent warping and the fixed-ended flange moments as the degree of warping
deformation. Toward this objective, conventional 12
12 stiffness restraint. Tong et al. [20] suggested a model for warping transmission
matrices were replaced by the new ones with warping considered based on modifying the traditional thin-walled beam element matrix
as an additional degree of freedom. In these studies, the end at the joints. Ettouney and Kirby [21] proposed a warping restraint
warping condition was assumed to be either completely free factor, which is the ratio between the bimoments of the partially
[8–10] or fully restrained [11–13] at both ends of the member. and fully restrained cases similar to the warping “spring” concept
The flexural–torsional behaviour of plane frames has been introduced by Yang and McGuire [22]. For both studies, static
studied by numerous investigators, but in most cases either warp- condensation was used to eliminate undesired degrees of freedom.
ing at joints was neglected by assuming six degrees of freedom for Although the basic idea of the two methods was same, the Yang and
beam elements [14], or considered to be fully prevented [15]. McGuire's procedure seems to be more representative of partial
warping restraint between two members as it operates with the
warping deformation which is easier to measure than the bimoment.
E-mail addresses: [email protected] (S. Shayan), The model featured a hypothetical warping rigidity applied as an
[email protected] (K.J.R. Rasmussen). internal spring at the joint.

http://dx.doi.org/10.1016/j.tws.2014.03.017
0263-8231/& 2014 Published by Elsevier Ltd.
2 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

The warping and distortion at angle joints composed of two thickness. Also, the method does not extend to 3D joints with
steel I-section members of the same cross-section were investi- members adjoining in three orthogonal planes, as in common 3D
gated by Vacharajittiphan and Trahair [23]. Four types of joints frames.
(unstiffened, diagonal-stiffened, box-stiffened and diagonal/box- The present study is concerned with developing a new method
stiffened joints) were considered in that study for which the for modelling joints including warping effects. The method ben-
numerical values of end warping restraint were calculated [24]. efits from the accuracy of 3D shell finite element modelling and
The study concluded that warping and distortion are interdepen- from the computational efficiency of using 1D beam elements. The
dent and depend on the joint configuration details. Subsequently, new joint model is developed using a combination of the sub-
these particular types of joints composing two or more channels structuring technique and linear springs. In the model, the joint
or I-sections were studied by Sharman [25], Morrel et al. [26] and accepts warping deformations from adjoining loaded members
Massarira [27] to determine the effect of warping transmission and redistributes the deformation to all connected beams and
through joints (Fig. 1). An important contribution to the research columns. In fact, the suggested joint model acts as a flexible
on the transfer of warping through joints was presented by interface between members and provides partial warping restraint
Basaglia et al. [28–31] using a numerical model which considered by means of springs. The model is general and can be applied to
transmission of warping torsion and local displacement compat- any kind of joint in 2D and 3D thin-walled structural frames.
ibility at frame joints of various configurations. The results of the
model were compared with shell element FE analysis using ANSYS
and excellent agreement was achieved. 2. Substructuring and static condensation
A brief review of the literature on the transmission of warping
through joints of thin-walled steel frames shows that all suggested Substructuring is a technique commonly used to overcome the
models need substantial numerical or computational effort. Due to difficulty of working with large dimensional problems [33]. In
the complexity of current models, the partial transmission of principle, a structure can be subdivided into smaller parts and
warping through joints is ignored in most design cases. Even if a each part analysed separately. The basic idea of substructuring
designer wanted to consider transmission of warping through analysis is that only certain degrees of freedom are retained while
joints, available commercial finite element software packages are others are eliminated by static condensation. This methodology is
limited to either completely prevent warping or allow warping to available in many finite element software packages and offers many
occur freely at joints when using beam finite element analysis advantages: (i) a substantial reduction in analysis time is achieved by
(B-FEA). At this point in time, there appears to be no FE software modelling only the joints using 3D shell finite elements rather than
available that allows the seventh degree of freedom (warping) to the entire frame, (ii) the substructure stiffness matrix needs only to
be partially transmitted. The only option to model warping be computed once for each type of joint with similar geometry, and
accurately is using shell elements (S-FEA), which is not a desired (iii) by writing a script to generate the substructure, the stiffness
method for complex structures due to its high computational cost. matrix can be calculated automatically and there is no need to create
A few models can be found in the literature for the partial the joint manually when changing the geometry of the joint.
transmission of warping at joints when using beam finite elements According to the conventional finite element method, the
[30,32]. global stiffness matrix K is obtained by assembling the stiffness
Basaglia et al. [30] developed a simple kinematic model to matrices of all elements. The global stiffness equation can be
simulate the warping transmission or restraint at the joints of expressed as
thin-walled frames in the context of beam finite element analysis.
Ku ¼ F ð1Þ
The model relies on the facility of most structural analysis soft-
ware (e.g. ABAQUS and ANSYS) to impose “linear constraint where K is an n
n matrix, u and F are n
1 node-displacement
equations” which establish constraint conditions between the and load vectors respectively and n is the total number of degrees
torsion warping degrees of freedom of the member end nodes. of freedom. In the substructure analysis, Eq. (1) is modified to
Despite its simplicity, the model is only applicable to four specific
Kn ur ¼ Fn ð2Þ
types of joints (see Fig. 1) and cannot model all possible cases with
n n
partially restrained warping. For example, the fully prevented where K and F are the stiffness matrix and force vector
warping assumption implied for diagonal/box stiffened joints respectively of smaller dimension than n obtained after static
may be rather conservative if the stiffeners have relatively small condensation. To obtain Eq. (2), in first step, the displacement

Fig. 1. Configurations of joints between channel members [28]: (a) unstiffened with flange continuity, (b) diagonal stiffened, (c) box stiffened, (d) diagonal/box stiffened
joints.
 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12 3

vector (u) is divided into two parts: ur containing degrees of 3. Joint model
freedom which are to be retained and (ue ) containing degrees of
freedom which are to be eliminated, i.e. 3.1. The basics–2D model
" #
ur A simple 2D angle joint (Fig. 2(a)) will be used to explain the
u¼ ð3Þ
ue model. Subsequently, the model will be set out for the more
general case of a 3D joint with multiple adjoining members.
In partitioned matrix form, the equation between forces and The basic idea of the model is to use shell finite elements to
displacements can be written as discretise the joint, and applying modes of deformation compatible
with the warping deformation of adjoining beams to obtain the
" #" # " # stiffness of the joint under this mode of deformation. Substructuring
Krr Kre ur Fr is here used to enable a warping deformation to be applied to the
¼ ð4Þ
Ker Kee ue Fe joint as a single degree of freedom, which can be linked to the
warping degree of freedom of the corresponding adjoining member.
A spring model is used to implement the warping stiffness exerted by
the joint in the beam-element finite element model.
in which Fr and Fe are force vectors of retained and eliminated In the first step, the warping deformation must be mapped
DOFs respectively and the K-terms are stiffness sub-matrices. onto the 3D joint model by a series of constraint equations which
Solving the second set of Eq. (4) with respect to the eliminated tie the longitudinal displacement of one corner node to the
DOFs ðue Þ provides, longitudinal displacements of other nodes on the same face so
as to represent a warping displacement in the direction of the
1
ue ¼ Kee ðFe  Ker ur Þ ð5Þ attached beam (Fig. 2(a)). A simple physical interpretation of the
model is a set of rigid bar linkages attached to the edge of each
The condensed equilibrium equation is obtained by substituting component plate which through pivots allow the warping defor-
Eq. (5) into the first set of Eq. (4): mation to be represented by a single degree of freedom, shown as
u1B in Fig. 2(b). The corresponding force required to produce the
1 1 displacement u1B is denoted by F 1B . In this terminology, the sub-
ðKrr Kre Kee Ker Þur ¼ ðFr  Kre Kee Fe Þ ð6Þ
script “B” refers to “beam” and the superscript “1” refers to the
global x1-direction. Mathematically, and in the finite element
1
which reproduces Eq. (2) with Kn ¼ Krr Kre Kee Ker and implementation of the model, the warping mode of deformation
1
Fn ¼ Fr  Kre Kee Fe . After applying static condensation, only the is implemented using constraint equations, which can be expressed
retained degrees of freedom are present in the stiffness in the general form,
equation (Eq. (6)), and each joint can be represented by an
equivalent stiffness matrix ( Kn ). a1 U 1i þa2 U 1j ¼ 0 ð7Þ

Fig. 2. (a) Longitudinal displacements to generate a single warping degree of freedom, (b) rigid bar model, (c) retained degrees of freedom for an angle box joint, (d) joint
deformations due to applied warping.
4 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

where U 1i and U 1j are the longitudinal displacements at adjacent

nodes i and j in the same plate element (e.g. nodes E and F in
Fig. 2(a)). For sections composed of three plate elements (e.g.
channel- and I-sections), three constraint equations are required
for each connected face of the joint. For the particular case of a
doubly symmetric I-section, the longitudinal warping displace-
ments of the corner nodes of each component plate element are
equal and opposite of each other, i.e. U 1i ¼  U 1j and hence
2
a1 ¼ a2 ¼ 1: Thus, the specific constraint equations for the joint
shown in Fig. 2(a) are U 1H ¼ U 1G , U 1G ¼  U 1E , and U 1E ¼  U 1F . In 3 1
general, the constants ða1 ; a2 ) are functions of the sectorial coordi-
nate and may be determined as ða1 ; a2 Þ ¼ ða1 ;  ðωi =ωj Þa1 Þ, where ωi
and ωj are the sectorial coordinates of nodes i and j respectively.
For each connected face of the joint, one degree of freedom is
chosen to represent the warping deformation. For the angle joint,
the degrees of freedom are u1B and u2C (Fig. 2 (c)), both chosen as
the longitudinal displacement of a corner node. It follows that the
number of degrees of freedom retained in the substructuring
process equals the number of adjoining beams and columns.
Having chosen the degrees of freedom and corresponding con-
straint equations, warping deformations can be readily applied to Fig. 3. Warping compatibility of adjoining members.
the shell finite element substructure, as exemplified in Fig. 2(d).
For the particular angle joint with two degrees of freedom into Eq. (11), whereby
(u1B , u2C ), the condensed stiffness matrix has four components: one B ¼ K φ0 φ 0 ð12Þ
for the warping displacement (u1B ) of the vertical face connected to
the beam (K BB Þ, one for the warping displacement (u2C ) of the where
horizontal face attached to the column (K CC Þ and two off-diagonal " #
K BB K BC
components (K BC and K CB ) representing interaction between the Kφ 0 ¼ ωKu ω ¼ ð13Þ
K CB K CC
warping deformations, e.g. K BC is the force (F 1B ) generated in
the direction of u1B as a result of a unit warping displacement in The terms (K BB ; K BC ¼ K CB ; K CC ) of the warping stiffness matrix
the direction of u2C . The stiffness matrix for the substructure after (Kφ0 ) relating bimoments to warping degrees of freedom are
condensation takes the form: readily obtained once the matrix Ku has been obtained from the
" # substructuring process and the sectorial coordinate matrix (ω) has
K BB K BC
Ku ¼ where K BC ¼ K CB ð8Þ been obtained from a cross-section analysis.
K CB K CC In the particular case where the adjoining beam and column
are doubly symmetric I-sections of the same width b and depth d,
Since the retained degrees of freedom are longitudinal displace-
the sectorial coordinates are
ments, their multiplication by the stiffness matrix produces forces,
" 1# " 1# 1
ωB ¼ ωC ¼ ω ¼ bh ð14Þ
FB uB 4
F ¼ Ku u where F ¼ 2 and u ¼ 2 ð9Þ
FC uC whereby
 
1 0
The warping deformations of the joint must be compatible with ω¼ω ¼ ωI ð15Þ
the warping deformations of the adjoining beams. As shown in 0 1
Fig. 3, the warping degrees of freedom of the end nodes of the so that
beam and column are denoted by φB' and φC' respectively. The
corresponding warping displacements at the points where the K φ 0 ¼ ω2 K u ð16Þ
displacements u1B and u2C defined are φB' ωB and φC' ωC , respectively, The stiffness Eq. (12) relates the bimoments (B) at the ends of the
where ωB and ωC are the normalised sectorial coordinates of the adjoining beams to the warping degrees of freedom (φ0 ). In
same points. Thus, principle, the stiffness terms could be incorporated directly into
" # " # the global stiffness matrix for the frame. However, few finite
0
ωB 0 ' φB'
u ¼ ωφ where ω ¼ and φ ¼ ð10Þ element programs, including commercial programs, have a facility
0 ωC φC' for such implementation. Hence, an alternative approach is used,
in which the effect of joint warping stiffness is implemented in the
The stress resultants corresponding to the warping degrees of
beam finite element model by means of springs. Specifically, the
freedom (φB' , ω0C ) are bimoments (BB ,BC ), as shown in Fig. 3. Since a
model consists of using a combination of a spring and a linear
concentrated force in the longitudinal direction of a beam pro-
constraint equation for each member connected to the joint.
duces a bimoment equal to the force times the sectorial coordinate
For the particular angle box joint shown in Fig. 2, two springs
at the point of application of the force, the bimoments in the beam
and two constraint equations are needed. To obtain the required
and column (BB ,BC ) are related to the forces acting on the face of
linear constraint equations, the relation between bimoment and
the joint (F 1B ; F 2C ) through,
warping is written out in full,
" # " # " #" #
BB BB K BB K BC φB'
B ¼ ωF where B ¼ ð11Þ ¼ ð17Þ
BC BC K CB K CC φC'

To obtain the relationship between the bimoments (B) and the The first diagonal warping stiffness term is then nominated as a
warping degrees of freedom (φ0 ), Eqs. (9) and (10) are substituted fixed value (K BB ) and the other terms are written as a fraction of
 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12 5

that value, i.e.

" #
K BB b1 K BB
Kφ ¼
0 b1 ¼ b2 ð18Þ
b2 K BB b3 K BB

where b1 ¼ K BC =K BB ; b2 ¼ K CB =K BB ¼ b1 and b3 ¼ K CC =K BB . Thus,

the stiffness equations may be rewritten as, 2
(
BB ¼ K BB ðφB' þ b1 φC' Þ 3 1
ð19Þ
BC ¼ K BB ðb2 φB' þ b3 φC' Þ

or
(
BB ¼ K BB ΦB
ð20Þ
BC ¼ K BB ΦC

where
Fig 5. 3D corner joint.
(
ΦB ¼ φB' þ b1 φC'
ð21Þ
ΦC ¼ b2 φB' þ b3 φC'

Eq. (20) describes a linear relationship between the bimoments The relationship between the longitudinal joint displacement
and the new variables (ΦB ,ΦC ). It may therefore be implemented in degrees of freedom and corresponding forces can be written as
the finite element model using linear springs and constraint 2 3 2 32 1 3
equations to define the spring deformations ΦB and ΦC as per F 1B1 K B1 B1 K B1 B2 K B1 C 1 K B1 C 2 uB1
6 7 6 76 3 7
Eq. (21). 6 F3 7 6 KB B K K K B 2 C 2 76 u B 7
6 B2 7 6 2 1 B2 B2 B2 C 1
76 2 7
6 7
Keeping in mind that the frame should be modelled in such a way 6 F2 7 ¼ 6 6 K K K K 76 2 7
C 1 C 2 76 u C 7
ð22Þ
6 C1 7 4 C B
1 1 C B
1 2 C C
1 1
54 1 5
that separate nodes are used for the adjoining members at the joint 4 5
K C 2 B1 K C 2 B2 K C 2 C 1 K C 2 C 2 u2C 2
and that these nodes are located at the perimeter of the joint, then F 2C
2

new “dummy” nodes are created near the ends of the adjoining
To obtain the relationship between the bimoments and the warping
elements and springs are attached to these nodes (Fig. 4). It is
degree of freedom, Eqs. (12) and (13) apply in which the sectorial
obvious that the stiffness of all springs is same and equal to K BB . The
degrees of freedom corresponding to the springs are ΦB and ΦC . coordinate matrix now is defined as
2 3
While the seventh degrees of freedom (warping) of the adjoining ωB 1 0 0 0
beams are partially transferred using the presented spring model, the 6 0 7
6 0 ωB 2 0 7
other six degrees of freedom are transmitted directly between ω¼6 6 0
7
7 ð23Þ
4 0 ω C 1
0 5
members connected to the joint. In this paper, the joint is considered
0 0 0 ωC 2
to be rigid with respect to translation and rotation, and so rigid links
are used to ensure compatibility of these six degrees of freedom where (ωB1 ; ωB2 ; ωC 1 ; ωC 2 ) are the sectorial coordinates of the
(Fig. 4). points in the cross-sections where the displacements (u1B1 , u3B2 ,
u2C 1 , u2C 2 ) are defined. Finally, the stiffness equations in terms of
3.2. 3D model bimoments can be written as
8
>
> BB1 ¼ K B1 B1 ΦB1
The proposed joint model was introduced in the previous >
>
>
< BB ¼ K B B ΦB
section with reference to a simple 2D angle joint with two 2 1 1 2
ð24Þ
adjoining members. However, it applies equally to 3D joints with >
> BC 1 ¼ K B1 B1 ΦC 2
>
>
multiple adjoining members, as will be demonstrated in this >
: BC ¼ K B B ΦC
2 1 1 2
section using a joint consisting of four members, three in same
plane and one perpendicular to the others (Fig. 5). Following the where
same methodology as for the 2D joint, the warping deformations 8
>
> ΦB1 ¼ φB' 1 þb1 φB' 2 þ b2 φC' 1 þ b3 φC' 2
are first mapped to the substructure and after static condensation >
>
< ΦB2 ¼ b4 φB' þ b5 φB' þb6 φC' þ b7 φC'
one longitudinal displacement is retained for each face which 1 2 1 2
ð25Þ
represents the warping degree of freedom for that connected face >
> ΦC1 ¼ b8 φB' 1 þ b9 φB' 2 þb10 φC' 1 þb11 φC' 2
>
>
: ΦC ¼ b12 φB' þ b13 φB' þ b14 φC' þ b15 φC'
(u1B1 , u3B2 , u2C 1 and u3C 2 ). 2 1 2 1 2

In Eq. (25), the constants (b1,b2, etc.) are the ratios between stiffness
terms of the Kφ0 -matrix, as obtained from Eq. (13), and the reference
stiffness (K B1 B1 ), e.g. b1 ¼ K B1 B2 =K B1 B1 , b2 ¼ K B1 C 1 = K B1 B1 , etc. The
warping stiffness matrix is symmetric so that b1 ¼ b4 , b2 ¼ b8 , etc.
The 3D joint model, combining the linear springs and con-
Constraint Equation straint equations expressed in Eqs (24) and (25), respectively, is
presented in Fig. 6.
Dummy Node The method has now been demonstrated for sample 2D and 3D
2
Spring joints composed of adjoining I-sections. However, the method is
Rigid link generally applicable to 3D joints of any geometry and members of
1 any type of cross-section. It consists of the following steps:
Member end node
Fig. 4. Joint spring model, 2D. 1. Create a shell finite element model of the joint.
6 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

2. Assuming the number of adjoining members is n, for each features six degrees of freedom corresponding to three transla-
adjoining member, determine the warping function and intro- tions and three rotations at each node. Because there is no match
duce a displacement (ui ; i ¼ 1; …; nÞ to define the magnitude of between the number of degrees of freedom of the beam (7 DOFs)
the warping displacement at the interface between the ith and shell (6 DOFs) elements, no direct method is available to
member and the joint. connect the warping degree of freedom of the beam to the joint
3. In the shell finite element model, construct constraint equa- model consisting of shell elements. However, the proposed joint
tions that allow the warping displacement to be applied for model converts the 3D shell element joint model to a simple
each adjoining member. system of linear springs which can be connected to the beam
4. Use static condensation in the shell finite element analysis to elements as follows.
obtain the n
n stiffness matrix (Ku, see Eq. (9)) for the n In the first step, the substructure is modelled in the same way
warping degrees of freedom (ui). that any structural model may be created using shell elements.
5. Construct the n
n diagonal ω -matrix (as exemplified by The nEQUATION command in ABAQUS is then used to set the
Eqs (10) and (23)) and determine the Kφ0 -matrix using Eq. (13). longitudinal displacements of corner nodes on each face of the
6. Define dummy nodes at joints in the beam element frame joint in accordance with the sectorial coordinates of the nodes.
analysis and introduce spring elements connecting the dummy For sections composed of three plate elements (e.g. I-sections), this
nodes to the end nodes of adjoining elements. requires three constraint equations for each face. Currently, sub-
7. Define the stiffness of the springs (K BB in the two previous structure modelling is not supported by the ABAQUS/CAE pre-
examples) and the deformation of each spring, denoted by Φi in processing user interface; thus a PERL script was written to
Eqs (21) and (25). generate the joint model (PERL is a common programming
8. Construct constraint equations in the beam element frame language that can be run in most computer operating system).
analysis linking the deformation (Φi ) of each spring to the The joint itself is analysed using the ABAQUS nSUBSTRUCTURE
warping degrees of freedom (φj'; j ¼ 1; ::; n) of adjoining mem- GENERATE command and the condensed stiffness matrix is
bers, as exemplified by Eqs (21), (25). generated using the nSUBSTRUCTURE MATRIX OUPTPUT com-
9. Apply rigid links to connect displacements and rotations of mand. This produces the terms of the Ku matrix (Eq. (9)). An
adjoining members, as appropriate. example of a PERL script for creating a shell finite element model
of the corner joint shown in Fig. 2 and an ABAQUS input file for
analysing the joint as a substructure can be found in [35].
4. Implementation of the model Once the warping stiffness matrix (Kφ0 ) is formulated using
Eq. (13), it can be applied to the beam finite element model. This is
The commercial finite element (FE) software ABAQUS [34] is achieved by defining dummy nodes, which are independent of the
used in this study for creating numerical models, although the nodes used to define the structure. Linear spring elements
model is general and can be used in any FE software with the (SPRING2 element of the ABAQUS library) are connected to each
substructuring capability. The S4R shell element is used to model pair of dummy nodes and for each spring element, one dummy
the joint as a substructure. The ABAQUS B31OS beam element is node is fixed while the other is the degree of freedom of the
selected which is a three dimensional element with 7 DOFs of element and is designated by Φ. This spring degree of freedom is
which the seventh is warping. The S4R shell element in ABAQUS then related to the warping degrees of freedom (φ0 ) of the
adjoining members through a constraint equation (e.g. as per
Eqs. (21) and (25)), which is defined in ABAQUS using the
n
EQUATION command. The translational and rotational degrees
of freedom of the adjoining beams are transmitted directly and
rigidly between members connected to the joint using “Beam”
Constraint Equation rigid links. The 2D and 3D joint models using ABAQUS are shown
in Fig. 7.
2 Dummy Node
Spring
3 1
Rigid link
Member end node 5. Verification and illustrative examples

In order to evaluate the performance and capabilities of the

proposed joint model, three thin-walled planar steel frames as
well as three space frames have been analysed using the model in
Fig 6. 3D joint spring model. conjunction with one-dimensional beam elements (B-FEA), and

Fig. 7. ABAQUS joint spring model, (a) 3D, (b) 2D.

 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12 7

the results compared with the results for the same frames from 150UB14 sections (Fig. 8) and a vertical point load P equal
obtained using 3D shell elements (S-FEA). To investigate the to 1000 N is applied to the shear centre at the end of the beam.
performance of the proposed model, buckling and first order A sample ABAQUS indata file for analysing the frame using the
elastic analysis as well as second order inelastic analysis have B-FEA spring model with a box-stiffened joint (Fig. 8(a)) is
been conducted using ABAQUS. All frames are made of steel included in [35].
I-sections with elastic modulus (E) equal to 200 GPa and Poisson's Table 1 shows the spring stiffness, the bimoments in the
ratio equal to 0.3. In order to prevent the overall frame out-of- adjoining members at incipient buckling and the critical buckling
plane buckling of 2D frames, all frames are laterally restrained at loads obtained using the beam element (B-FEA) with spring joint
the beam-column connections. For the shell element joint model, model and the shell (S-FEA) element model for the three joint
a lateral restraint is applied on a surface at the middle of the joint, configurations. In addition, for comparison, the critical buckling
(a surface rather than a node is restrained to avoid large local loads are obtained using the constraint equation model introduced
deformations as can result from applying a restraint at a single by Basaglia et al. [30], referred to as “Equation” in Table 1.
point). In the beam element model, the top of the columns Based on the results shown in Table 1, it can be seen that the
is restrained in the global out-of-plane Z-direction. The column spring stiffness increases as stiffeners are added to the joint.
bases are fully fixed and to avoid any local deformations all cross- The warping transmission is almost complete and inverse for the
sections are chosen as compact. For this study, three dimensional box joint (a) and almost prevented for the box joints with one and
beam elements with 7 DOFs per node (B32OS) and S4R shell two diagonal stiffeners (b and c), as confirmed by Basaglia et al.
elements are used. The shell finite element analyses are based on [30]. Comparing the results of S-FEA and both types of B-FEA, it
discretisations of the flanges and webs into four and six elements can be seen that for joints (a) and (b) the “spring” joint model can
respectively while the length of each element in the beam predict the critical buckling load more accurately than the “con-
analyses is taken as 25 mm. The thicknesses of stiffeners added straint equation” model.
to the joints are taken as equal to the flange thickness of the The maximum discrepancy between the B-FEA spring model
adjoining beams. For the nonlinear analysis, (i) the yield stress of and the S-FEA model is 2.57%, indicating good accuracy. It is worth
all frames is taken as 320 MPa, (ii) the material is assumed to be noticing that the joint configuration influences the critical buck-
linear perfectly plastic, and (iii) the initial imperfection is mod- ling load and that by adding two diagonal stiffeners to the joint,
elled by scaling the first buckling mode by the scale factor of the buckling load is increased by 6% compared to the frame with a
0.00123H, in which H is the total structural height [36]. box-joint without diagonal stiffeners. Fig. 9(a) further illustrates
that excellent agreement is achieved between the load versus out-
5.1. Planar Frames of-plane tip-deflection curves obtained using beam and shell
element nonlinear analyses. The joint stress contours at the
5.1.1. L-shape frame ultimate load are presented in Fig. 9(b) and clearly indicate that
The first study illustrates the elastic buckling behaviour as well the joint is not yielded at the ultimate limit state. The ultimate
as the nonlinear response of the L-shape steel frame shown in frame strength is 910 N which is almost equal to the elastic critical
Fig. 8, which has only one beam and one column connected at the buckling load. Hence, the frame fails elastically in a flexural–
joint. In order to assess the application of the proposed model, torsional buckling mode.
three different types of joints have been considered: (a) box
stiffened joint with web continuity, (b) box-stiffened joint with 5.1.2. One-bay, one -storey frame with an overhanging member
one diagonal stiffener and (c) box-stiffened joint with two diag- A portal frame with an overhanging member has been studied.
onal stiffeners, as shown in Fig. 8. The beam and column are made In addition to the corner joint considered in the previous example,

75 mm

Fixed 5 mm
(a)
150 mm

7 mm 4000 mm
P P
(b)

Y
(c)
X Z
8000 mm

Fig. 8. Configuration and dimension of L-shape plane frame.

Table 1
Spring stiffness, bimoments and critical buckling loads of L-shape frame.

Joint configuration Spring stiffness Bimoment (N mm3) Critical buckling load (N) Error (%) (B-FEA
(N mm3) (spring) and S-
FEA)
(ƒK BB) Beam Column B-FEA B-FEA (Spring) S-FEA
(Equation)

Joint (a) 2.867E+12 860 −725 897 895 872 2.57%

Joint (b) 4.484E+12 1613 −250 972 937 920 1.80%
Joint (c) 6.200E+12 1727 −200 NA 948 931 1.79%
8 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

Fig. 9. (a) Load–deflection curves of nonlinear inelastic analyses, (b) joint stress contours.

the frame features an intermediate joint which connects three Table 2

members. This example has been chosen to show that the method Spring stiffness of joint 2 and critical buckling loads of portal frame with an
is general and does not depend on the number of adjoining overhanging member.
members or joints. The frame configuration and joint types can
Joint Spring Critical buckling Error (%)
be seen in Fig. 10. The same three joint types are considered as in configuration stiffness load (N/mm) (B-FEA
the previous example, where the same type of joint (e.g. box- (joint 2) (spring)
stiffened joint with web continuity) is assigned to both Joint 1 and (N.mm3) and S-FEA)
2. The frame is made from 150UB14 sections and a uniformly (KBB) B-FEA B-FEA S-FEA
(equation) (spring)
distributed load equal to 1 N/mm is applied along the length of
both beams and at the shear centre. Joint (a) 3.322E þ 12 939 922 912 1.08%
The stiffness matrix for joint 2 with non-stiffened box-stiffened Joint (b) 5.038E þ 12 1031 969 960 0.93%
configuration obtained using the substructure analysis in ABAQUS Joint (c) 6.269E þ 12 NA 990 983 0.71%
is presented in Eq. (23) and can be converted to warping stiffness
matrix using Eq. (18). As the size of the beams and column
attached to the joint are the same, several stiffness matrix terms
are equal (i.e. Ku (1, 1) ¼ Ku (2, 2) and Ku (1, 3) ¼ Ku (3, 2), where 5.1.3. One-bay, two-story frame
columns 1, 2 and 3 of Ku refer to u1B1 , u1B2 and u2C , respectively. A two-storey portal frame with box joints is analysed using
2 3 both elastic buckling and nonlinear inelastic analyses. The frame
440; 568  163; 044 251; 363 configuration and loading are presented in Fig. 12. The frame is
6 251; 363 7
Ku ¼ 4  163; 044 440; 568 5 ðN:mm3 Þ ð26Þ made from 310UB40 sections (Fig. 12) and a distributed load (w)
251; 363 251; 363 523; 643 equal to 67.5 N/mm is applied at the shear centre. The elastic
critical buckling load factor obtained using B-FEA and the pre-
The warping stiffness matrix and a set of three constraint equa- sented spring model is 1.18, i.e. the critical buckling value of the
tions for joint 2 are shown in Eqs. (27) and (28) respectively. applied load is 79.65 N/mm, while the frame buckling load
2 3 corresponding to S-FEA is 76.95 N/mm. With the difference of
K BB  0:37 K BB 0:57K BB 3.39% and so again, the spring joint model fairly accurately
6 7
Kφ0 ¼ 4  0:37K BB K BB 0:57 K BB 5 ð27Þ predicts the effect of warping transmission through the joint in
0:57K BB 0:57K BB 1:188 K BB elastic buckling analyses.
In the case of nonlinear analysis, the ultimate load factors
8 obtained using B-FEA and S-FEA analyses are 0.96 and 0.82
> φ '  0:37φB' 2 þ 0:57φC' ¼ B1
< B1 respectively, as shown in Fig. 13(a). It follows from the stress
 0:37φB' 1 þ φB' 2 þ 0:57φC' ¼ B2 ð28Þ contours shown in Fig. 13(b) that the joints are yielded in the
>
: 0:57φ ' þ0:57φ ' þ 1:188φ ' ¼
B1 B2 C C1 ultimate limit state. The discrepancy between the B-FEA and S-FEA
nonlinear analysis results stems from the joint modelling which
The warping stiffness (K BB ) and critical buckling loads are pre- implicitly assumes elastic material behaviour, since the spring
sented in Table 2 for the three Joint 2 configurations. Once again, stiffness terms are obtained from a substructuring method that
excellent agreement is achieved between the critical buckling calls a linear-elastic analysis. Note that if the yield stress of the
loads obtained from the beam finite element analysis using the joints is increased to 450 MPa to prevent yielding at the joints,
spring joint model and the shell element analysis. The largest error it can be seen from Fig. 13(a) that the B-FEA and S-FEA analyses are
is 1.08% which relates to the box joint. Similar to the previous in close agreement. It follows that the proposed model is accurate
example, changing the joint configuration from a box joint to box for nonlinear inelastic analysis as long as the joints remain elastic,
joints with one and two diagonal stiffener increases the critical but may become inaccurate when the frame strength is governed
buckling load by 5% and 7% respectively. For all cases, the spring by yielding in the joints.
model more accurately predicts the elastic buckling behaviour
compared to the constraint equation model [30]. 5.2. Space frames
The load versus out-of-plane deflection curves obtained from
nonlinear analyses of the frame with box joints are plotted in Three space frames have been analysed using beam and shell
Fig. 11(a), indicating excellent agreement between the shell and finite elements to evaluate the applicability of the proposed joint
beam element analyses. The ultimate strength is 950 N/mm which spring model to three dimensional frames with members adjoin-
is close to the elastic critical buckling load. It follows from the ing in different planes. The first frame is a one-bay-, one-storey
stress contours shown in Fig. 11(b) that the failure is elastic. frame with a transverse beam studied by Basaglia et al. [30].
 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12 9

Fixed (a) (b) (c)

w

w 4000
w mm

w Joint 1

6000 mm
Joint 2
Y

X Z 6000 mm

Fig. 10. Configuration and dimensions of plane portal frame with an overhanging member.

Fig. 11. (a) Load–deflection curve of nonlinear inelastic analysis, (b) joint stress contours.

0.5w 0.5w
4000 mm
Fixed

w w 4000 mm

165 mm

304 mm 6.1 mm

6000 mm
10.2 mm

Fig. 12. Configuration and dimensions of two-storey plane portal frame

Fig. 13. (a) Load–deflection curve of nonlinear inelastic analysis, (b) joint stress contours.
10 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

All members have the same I-300
8 cross-sections with a flange the beams is twice the web height of the columns. The cross-
width of 150 mm, web height between flange centrelines of section dimensions can be found Fig. 16. Note that the columns
300 mm and uniform thickness of 8 mm. Five equal point loads have the same cross-section as do the beams in the X- and Z-
P are applied at the top of the columns and the mid-span of the directions. The loads are applied as uniformly distributed loads
transverse beam at the centre. While the column-to-beam joint along the beams.
exhibits web continuity, there is flange continuity in the beam-to- For the second frame, the critical elastic buckling loads obtained
beam ones. Restraints are applied to prevent displacements in the from beam and shell finite element analyses are 350.3 kN and
transverse direction at the joints and mid-span of the transverse 342.8 kN respectively with the difference of 2.13%. For the second
beam. Fig. 14 shows the frame configuration, restraints, loading frame, the critical elastic buckling loads are 14.95 N/mm for beam
and joint types of the space frame for both B-FEA and S-FEA. The analysis and 14.20 N/mm for shell analysis, resulting in a difference
critical buckling loads obtained using shell element (S-FEA) are of 5.02%. The buckling shapes obtained from the two types of analysis
summarized in Table 3 and then compared with the results of are perfectly matched for both frames and shown in Fig. 17. The
B-FEA using the kinematic equation reported by Basaglia et al. [30] beams buckle in flexural–torsional modes which cause pronounced
as well as the joint model using spring. As can be seen from the warping of the joints, so the performance of the joint model is well
results, the proposed joint model can predict the buckling load tested by these examples.
more accurately especially for the case of box-stiffened joint with
a diagonal stiffness. The kinematic equation model assumes the
warping in this type of joint is completely restrained while the
joint model with spring can capture the small amount of warping 6. Conclusions
transmission through the joint.
The second frame is exactly same as the first frame with three This paper outlines an accurate new joint model which can be
transverse beams to verify the joint model for a joint with three incorporated into beam-based finite element analyses to consider
members adjoining in three orthogonal planes. The member the effect of warping and its transmission between connected
dimensions are the same, except for the two transverse end beams members. The proposed model is simple to implement into finite
for which the flange width is 300 mm. The frame features two element models using readily available beam and spring elements,
types of joint: (a) beam-column joint and (b) beam-beam joint. and does not require the beam stiffness matrix to be modified or
Warping through the joints can be modelled by the use of three computationally costly shell finite element modelling to be
springs and three constraint equations for each joint in the beam employed. In the proposed method, the joint itself is modelled
finite element analysis. Fig. 15 shows the frame configuration, as an assemblage of shell elements and analysed separately as a
restraints, loading and joint types of the space frame for both B- substructure to produce the warping stiffness of the joint. The
FEA and S-FEA. proposed model is general and applicable to any arbitrary joint
The third space frame is a one-bay, two-storey frame consisting type and any type of connecting member. Unlike most available
of two different corner joints with three and four connecting methods for modelling joints, which apply to corner joints with
members, as shown in Fig. 16. This example is chosen to show that only two adjoining members, the proposed model can be used
the model is applicable to any type of joint and does not depend regardless of the number of connected members, and applies
on the number of adjoining members. To achieve a more general equally to 2D and 3D frame structures. The model lends itself to
case compared to previous examples, while the connecting beams common steel frame design applications in industry where
and columns have the same flange dimension, the web height of designers look for fast and convenient methods of analysis.

P (a) (b) (c)

P
P

Fixed
10000 mm
5000 mm

Fig. 14. Configuration and joint types of one-bay, one-storey space frame studied by Basaglia et al. [30].

Table 3
Critical buckling loads of portal frame with an overhanging member.

Joint Critical buckling load (KN) Error (%)

configuration
B-FEA (equation- reported B-FEA (spring) S-FEA B-FEA (equation) and B-FEA( (spring) and S-FEA)
by Basaglia et al. [30]) S-FEA(reported by Basaglia et al. [30]

Joint (a) 307.88 308.10 306.47 0.46 0.53

Joint (b) 325.45 329.55 327.99  0.78 0.47
Joint (c) 350.00 339.9 339.2 3.09 0.21
 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12 11

150 mm
150 mm

P
300
Fixed (b) mm
P
P
P

300 mm
150 mm
(a) 10000 m
300
5150 mm
Fixed mm

Fig. 15. Configuration and joint types of one-bay, one-storey space frame.

158 mm
153 mm
0.5w 306 mm
306 mm
Fixed
(b)

w
153 mm

158 mm

Fixed Y

X Z (a) 8000 8000

Fig. 16. Configuration and joint types of the one-bay, two-storey space frame.

Fig. 17. Buckling shapes of space frame (a) B-FEA, Frame 1; (b) S-FEA, Frame 1; (c) B-FEA, Frame 2; (d) S-FEA, Frame 2.
12 S. Shayan, K.J.R. Rasmussen / Thin-Walled Structures 82 (2014) 1–12

The proposed joint model is based on a combination of linear [12] Dinno KS, Merchant W. A procedure for calculating the plastic collapse of I-
springs and linear constraint equations, and as was shown in the sections under bending and torsion. J Struct Eng 1965:219–2143 1965:219–21.
[13] Chaudhary AB. Generalized stiffness matrix for thin-walled beams. J Struct Div
verification examples it can be applied accurately in elastic ASCE 1982;108(3):559–77.
buckling analyses. It can also be applied accurately in nonlinear [14] Kim SE, Kim Y, Choi SH. Nonlinear analysis of 3-D steel frames. Thin-Walled
analyses as long as yielding does not occur in the joints. The Struct 2001;39(6):445–61.
[15] Trahair NS. Elastic stability of continuous beams. J Struct Div ASCE 1969;101
substructuring technique adopted in this paper is based on a linear (7):1497–516.
elastic analysis of the joint and so produces a linear spring model [16] Dinno KS, Gill SS. The plastic torsion of I-sections with warping restraint. Int
which cannot capture the effect of yielding. J Mech Sci 1964;6(1):27–43.
[17] Ojalvo M, Chambers RS. Effect of warping restraints on I-beam buckling.
Five different frames have been analysed using beam elements J Struct Div ASCE 1977;103(ST12):2351–60.
to evaluate between the elastic buckling results of all considered [18] Austin WS, Yegian S, Tung TP. Lateral buckling of elastically end restrained
frames obtained using beam finite element analysis with spring I-beams. Trans, ASCE 1957;122:374–88.
[19] Trahair NS. Elastic stability of propped cantilevers. Inst Eng Aust Civil Eng
joint models and shell finite element analysis. All previous models
Trans 1968;CE10(1):94–100.
for warping transmission including “linear constraint equations”, [20] Tong GS, Yan XX, Zhang L. Warping and bimoment transmission through
as proposed by Basaglia et. al [30], assume fully prevented warping diagonally stiffened beam-to-column joints. J Constr Steel Res 2005;61
for diagonal-stiffened joints. This assumption is accurate when the (6):749–63.
[21] Ettouney MM, Kirby JB. Warping restraint in three-dimensional frames.
diagonal stiffeners are sufficiently rigid to prevent warping. It was J Struct Div ASCE 1981;107(ST8):1643–56.
shown in the first example that even for the case of box-stiffened [22] Yang YB, McGuire W. A procedure for analysing space frames with partial
joints with two diagonal stiffeners there is still a degree of warping restraint. Int J Num Methods Eng 1984;20:1377–90.
[23] Vacharajittiphan P, Trahair NS. Warping and distortion at I-section joints.
warping transmission though the joints and hence, the assump- J Struct Div ASCE 1974;100(ST3):547–64.
tion of fully prevented warping in diagonal/box-stiffened joints [24] Basaglia C, Camotim D, Silvestre N. Global buckling analysis of plane and space
overestimates the stiffness of the frame. The proposed model can thin-walled frames in the context of GBT. Thin-Walled Struct 2008;46
(1):79–101.
predict the elastic behaviour of this type of joint more precisely. [25] Sharman PG. Analysis of structures with thin-walled open sections. Int J Mech
The study of different types of joints shows that the critical Sci 1985;27(10):665–7.
elastic buckling load is related to the joint configuration and [26] Morrell PJB, Riddington JR, Ali FA, Hamid HA. Influence of joint detail on the
flexural/torsional interaction of thin-walled structures. Thin-Walled Struct
progressively increases as the joint becomes stiffer. Stiffeners are 1996;24(2):97–111.
effective in reducing warping and the distortion of cross-sections [27] Masarira A. The effect of joints on the stability behaviour of steel frame beams.
and thereby increase the critical buckling load. J Constr Steel Res 2002;58(10):1375–90.
[28] Basaglia C, Camotim D, Silvestre N. GBT-based local, distortional and global
buckling analysis of thin-walled steel frames. Thin-Walled Struct 2009;47
References (11):1246–64.
[29] Camotim D, Basaglia C, Silvestre N. GBT buckling analysis of thin-walled steel
frames: A state-of-the-art report. Thin-Walled Struct 2010;48(10-11):726–43.
[1] Timoshenko SP. Bull Polytech Inst St Petersburg 1905. [30] Basaglia C., Camotim D., and Silvestre N., Kinematic models to simulate the
[2] Timoshenko SP, Gere JM. Theory of elastic stability. New York: McGraw-Hill; torsion warping transmission at thin-walled steel frame joints. In: Steel &
1961. composite structures – proceeding of the fourth international conference
[3] Wagner, H., Torsion and buckling of open sections, Danzig: Technische 2010: Sydney, Australia.
Hochschule (25th Anniversary Publication); 1936. p. 1904–29. [31] Basaglia C, Camotim D, Silvestre N. Torsion warping transmission at thin-
[4] Vlasov VZ. Twisting, stability and vibrations of thin-walled members. J Appl walled frame joints: kinematics, modelling and structural response. J Constr
Math Mech 1939;3:1. Steel Res 2012;69(1):39–53.
[5] Krajcinovic D. A consistent discrete elements technique for thin-walled [32] MacPhedran, IJ, Frame stability considering member interaction and compat-
assemblages. Int J Solids Struct 1969;5(7):639–62. ibility of warping deformation [Ph.D. thesis]. Edmonton, Alberta: University of
[6] Rajasekaran S, Murray DW. Finite element solutions of inelastic beam Alberta, 2009.
equations. J Struct Div ASCE 1973;99(12):2423–38. [33] Han TY, Abel JF. Substructure condensation using modified decomposition. Int
[7] Yoo CH. Bimoment contribution to stability of thin-walled assemblages. J Numer Methods Eng 1984;20:1959–64.
Comput Struct 1980;11(5):465–71. [34] ABAQUS/STANDARD, User's manual version 6.8, 2009: Hibbit, USA: Karlsson &
[8] Trahair NS. Elastic stability of I-beam elements in rigid-jointed structures. Sorensen.
J Inst Eng 1966:171–8038 1966:171–80. [35] Shayan S, Rasmussen K.J.R., and Zhang H., A new joint model for warping
[9] Kitipornchai S, Trahair NS. Elastic lateral buckling of stepped I-beams. J Struct transmission in thin-walled steel frames. Research report no.922, School of
Div, ASCE 1971;97(10):2535–48. Civil Engineering, The University of Sydney, Australia; 2012.
[10] Tebedge N, Tall L. Linear stability analysis of beam- columns. J Struct Div ASCE [36] Shayan S, Rasmussen K.J.R., and Zhang H., On the modelling of geometric
1973;99(STI2):24–39. imperfections of steel structural members and frames. In: The seventh
[11] Razzaq Z, Galambos TV. Biaxial bending of beams with or without torsion. J international conference on advances in steel structures (ICASS); 2012: China,
Struct Div ASCE 1979;105(ST11):2145–62. Nanjing.