The generalised constrained finite strip method for

thin-walled members in shear

Morgan A. Rendall∗, Gregory J. Hancock, Kim J.R. Rasmussen

School of Civil Engineering, University of Sydney, NSW 2006, Australia

Abstract

The constrained finite strip method (cFSM) is an extension of the semi-analytical

finite strip method (SAFSM) of structural analysis of thin-walled members,
where consideration of the displacement fields utilised and of various mechani-
cal criteria allows constraint matrices to be formed. The application of these
constraint matrices to the linear buckling eigenvalue problem of the SAFSM
results in deformation fields that satisfy the considered criteria and, therefore,
isolate particular modes. Through careful selection of the mechanical criteria,
the deformation fields obtained may be restricted to particular buckling modes.
This is referred to as modal decomposition. While the cFSM has been applied
to modal decomposition of thin-walled, prismatic members under the action of
longitudinal normal stresses, it has yet to be applied to such members under the
action of shear stresses. Recent work using the SAFSM to analyse the buckling
behaviour of thin-walled, prismatic members under applied shear stresses, no-
tably by Hancock and Pham, has shown that the issues of potentially indistinct
minima or multiple minima in the signature curve can occur under this loading,
as they did for compression and bending. This paper briefly presents the de-
rivation of a SAFSM that permits coupling between longitudinal series terms
of sines and cosines and also considers membrane instability due both to shear
stresses and transverse normal stresses. It then presents the application of the

∗ Corresponding author.
Email addresses: morgan.rendall@sydney.edu.au (Morgan A. Rendall),
gregory.hancock@sydney.edu.au (Gregory J. Hancock), kim.rasmussen@sydney.edu.au
(Kim J.R. Rasmussen)

Preprint submitted to Thin-Walled Structures April 15, 2017

 cFSM to such a finite strip and results are produced for members under shear
stresses. While the results are presented for members with unrestrained ends
(equivalent to infinitely long members), simplification via removal of the degrees
of freedom not present in typical FSM formulations would allow finite length
members with simply-supported ends to be analysed.
Keywords: Constrained finite strip method, Semi-analytical finite strip
method, Shear buckling analysis

1. Introduction

The finite strip method (FSM) developed by Cheung [1] is a specialisation

of the finite element method that utilises longitudinal regularity of the ana-
lysed member to reduce the dimension of the problem being analysed. First
5 utilised for local buckling analysis of thin-walled members by Przemienicki [2],
before being extended to other forms of buckling by Plank and Wittrick [3], the
FSM has become an indispensable design tool thanks to its ability to generate
a curve showing the critical elastic buckling stress of a section as a function of
the buckling half-wavelength, when only a single longitudinal half-wavelength
10 is considered; this is known as the signature curve of a section and is a concept
that was popularised by Hancock [4]. The ubiquity of the FSM in the analysis
and design of thin-walled, cold-formed steel members has only become more pre-
valent with the development of the Direct Strength Method (DSM) [5] which, in
practice, predicts the ultimate strength of a member by considering the signa-
15 ture curve and the geometric and material properties of the section. Typically,
this process involves taking the buckling stress values of the signature curve at
its two minima as the critical stresses corresponding to local and distortional
buckling, and using these in the DSM strength equations [6, 7]. However, there
are many sections for which the signature curve may not have two minima, or
20 may have more than one minimum for local or distortional buckling [8]. Further,
the buckling modes at the lengths where the signature curve attains its minima
may not be ‘pure’ local or distortional modes. These ambiguities in the sig-

2
 nature curve prompted the development of the constrained finite strip method
(cFSM) [9, 10, 11], which draws on the mechanical assumptions of Generalised
25 Beam Theory (GBT) [12] in order to define pure global, distortional and local
buckling modes. Application of the cFSM, available in the finite strip computer
program CUFSM [13], then allows the critical buckling stresses of these pure
local and distortional buckling modes to be determined. As the DSM is cali-
brated based on signature curves developed for the general (i.e. unconstrained)
30 deformation field of the FSM, application of cFSM to the DSM and further
development of the cFSM are fields of ongoing research [14, 15, 16].
Until recently, the DSM has only been applicable to members under longi-
tudinal normal stresses, i.e. compression and/or bending; it has recently been
extended to C-section members under shear [17]. As for the DSM for com-
35 pression and bending, the DSM for shear requires knowledge of the signature
curve of the section. The first semi-analytical finite strip capable of analysing
members under shear stresses was that of Plank and Wittrick [3], which was
recently revitalised by Hancock and Pham [18, 19] who applied the formulation
to the buckling of C-sections under shear stresses. It should be noted that, as
40 their SAFSM assumes longitudinal regularity, including stresses, the moment
gradient necessary for a section under shear to be in equilibrium cannot be re-
plicated and so the members analysed are under ‘pure’ shear. Their analysis,
as well as subsequent analyses, such as of C-sections with longitudinal stiffe-
ners [20], revealed signature curves that display many of the same ambiguities
45 of those developed for members in compression and/or bending; i.e. indistinct
minima and possible mode coupling.
In light of this, applying the cFSM methodology to members in pure shear
would prove useful, both as a theoretical tool for examining the buckling beha-
viour of such members and for assisting in further development of the DSM for
50 shear. As the cFSM methodology is largely separate from the FSM to which it
is applied, this paper will first present a finite strip that may be applied to mem-
bers in a combined loading state, where all components of the Green-Lagrange
in-plane strains are considered in formulating the stability matrices. Coupling

3
 between longitudinal series terms of different numbers of half-wavelengths is
55 also permitted. Subsequently, the application of the cFSM methodology to this
FSM will be elucidated, using the recently-developed generalised cFSM [21, 22].

2. SAFSM for applied shear

2.1. Linear buckling analysis

The linear buckling eigenvalue problem of the SAFSM is a second-order

60 analysis formulated via the theorem of stationary potential energy. The total
potential energy of the system is the sum of the internal elastic strain energy Ue ,
which is obtained by evaluating the energy stored by the actions of the internal
linear stresses σ L in the linear strains εL , and the potential energy due to the
externally applied stresses, which is obtained as the negative of the work W
65 of the applied stresses σ in the respective non-linear strains εN L . This total
potential energy is as shown in Eq. (1), where the integral is over the volume
of the elements of the system and the linear stresses and strains are related by
σ L = EεL , where E is the relevant constitutive matrix for the problem at hand.

Z   Z  
1 T T 1 T T
Ue − W = σ εL − σ εN L dV = ε EεL − σ εN L dV (1)
V 2 L V 2 L

70 By relating the strains to the degrees of freedom d of the system, the total
potential energy may be rewritten as given in Eq. (2), where KE is the global
elastic stiffness matrix of the system and KG is the global geometric stability
matrix of the system, which scales linearly with the applied stresses and so is
often written as a load factor λ multiplied by KG calculated for some reference
75 stresses. Obtaining the stiffness and stability matrices for the SAFSM is briefly
described in the following sections.

1 T 1 1
Ue − W = d KE d − dT KG d = dT (KE − λKG ) d (2)
2 2 2

By making the total potential energy of Eq. (2) stationary with respect to
each degree of freedom, the linear buckling eigenvalue problem of the SAFSM

4
 is formulated and is as given in Eq. (3), where Λ is a diagonal matrix contai-
80 ning the eigenvalues of the problem and Θ is a matrix whose columns are the
corresponding eigenvectors.

(KE − ΛKG ) Θ = 0 (3)

2.2. Displacement fields

The first SAFSM able to analyse members under shear was that of Plank
and Wittrick [3], who utilised complex degrees of freedom coupled with complex
85 exponentials as defined by Eq. (4), where ξ is proportional to the coordinate in
the longitudinal direction and i is the imaginary unit, to incorporate the phase-
shift of displacements across the strip width that occurs for a member under
shear stresses.
eiξ = cos ξ + i · sin ξ (4)

The displacement fields d were then defined as,

d = Re Nd · eiξ

(5)

90 where N is the vector of transverse shape functions for the current displacement
field, d is the vector of corresponding complex degrees of freedom and ‘Re’ de-
notes the real part of its argument. When evaluating the longitudinal (warping)
displacements, the argument of Eq. (5) was multiplied by i to incorporate the
out-of-phase nature of these displacements with respect to the transverse dis-
95 placements. The utilised longitudinal functions correspond to a finite-length
member with unrestrained ends or, equivalently, a member of infinite length
with supported ends. Due to its complex mathematics, formulation of the stiff-
ness and stability matrices depended on the identity given in Eq. (6), where a
and b are vectors of equal length, G is a square matrix of corresponding size
100 and the bar denotes the complex conjugate. This identity has no direct analog
for the case where the arguments of the two complex exponentials differ and so
coupling between longitudinal series terms is not possible.
Z 2π
Re aT eiξ · G · Re beiξ dξ = π · Re āT Gb
  
(6)
0

5
 One method to overcome this is to explicitly evaluate the ‘realness’ of Eq. (5)
prior to formulating the stiffness and stability matrices. This was done by
105 Mahendran and Murray [23] and resulted in a FSM with twice the usual number
of degrees of freedom. Although this explicitly real displacement field of sines
and cosines makes it possible to consider coupling between longitudinal series
terms, Mahendran and Murray did not do so. The right-handed coordinate
system and degrees of freedom utilised herein are as shown in Fig. 1; the
110 corresponding displacement fields are as given in Eqs. (7) – (11), where cm and
sm are as defined by Eq. (11) and so correspond to m half-wavelengths along the
considered length L, the superscripts ‘r’ and ‘i’ refer to the real and imaginary
components of the complex degrees of freedom of Eq. (5), and q is the number of
series terms considered. Note that, due to the sign of the shape functions used
115 for the rotational degrees of freedom, a positive rotation about the longitudinal
axis is defined by the left-hand rule. The degrees of freedom in Fig. 1 are all
shown at the mid-length of the strip, which would correspond to longitudinal
variation according to a sine of one half-wavelength longitudinally; the degrees
of freedom corresponding to a cosine of one half-wavelength longitudinally would
120 then occur at the ends of the strip.

q h
X ih iT
u= Nu · cm −Nu · sm ur1m ur2m ui1m ui2m (7)
m=1

q h
X ih iT
v=− Nv · sm Nv · cm r
v1m r
v2m i
v1m i
v2m (8)
m=1

q h
X ih iT
w= r r r r i i i i
Nw · cm −Nw · sm w1m θ1m w2m θ2m w1m θ1m w2m θ2m
m=1
(9)
h x   x i
Nu = Nv = 1 − ,
b  b
3x2 2x3 2x x2 3x2 2x3
     2 
x x
Nw = 1− 2 + 3 x 1− + 2 − 3 x −
b b b b b2 b b2 b
(10)

6
  mπy   mπy 
cm = cos , sm = sin (11)
L L
The only difference between the displacement fields of Eqs. (7) – (11) and
125 those of Mahendran and Murray is the negatives on a number of the shape
functions, which arise due to explicitly evaluating the various forms of Eq. (5).
As described earlier, the argument of Eq. (5) for the longitudinal displacements
v was multiplied by i before its real part was evaluated. Although these ne-
gatives are arbitrary and so may be removed without any effect, it was chosen
130 to leave them in. Note that the ‘imaginary’ degrees of freedom (superscript
‘i’) correspond to those of a member with simply-supported ends and so retai-
ning only these degrees of freedom would allow the theory developed herein to
analyse such members, such as in [24].
In the following, the local degree of freedom ordering will be as shown in
135 Eq. (12), which also shows the local degrees of freedom in partitioned form.
The subscripts ‘M’ and ‘B’ refer to membrane and bending/flexural degrees
of freedom respectively. This ordering was chosen as a) it allows the 16x16
local stiffness and stability matrices to be expressed in block diagonal form as
two 8x8 matrices, one each for membrane and flexural components, and b) it
140 is practically identical to the degree of freedom ordering utilised in CUFSM,
resulting in stiffness matrices and stability matrices for applied longitudinal
normal stresses that are directly comparable to those found in [13].
h iT
d = ur1 v1r ur2 v2r w1r θ1r w2r θ2r ui1 v1i ui2 v2i w1i θ1i w2i θ2i
h iT
= drM diM drB diB
(12)

2.3. Development of stiffness and stability matrices

With the displacement fields defined, the degrees of freedom may be rela-
145 ted to the various strains to formulate the local stiffness and stability matrices.
For the stiffness matrix, the linear strains are decomposed into membrane and
bending strains, which follow plane stress assumptions and Kirchoff thin-plate

7
 theory respectively. The utilised constitutive relation between the linear stresses
and strains is Hooke’s Law in two dimensions, allowing for orthotropy. Deri-
150 vation of the local stiffness matrix is given in detail in [25], where it is shown
that the matrix may be expressed in block diagonal form by two 8x8 matrices;
one each for the contributions of the membrane and the flexural displacements
to the elastic strain energy of the system. These 8x8 matrices are identical to
a repetition of the corresponding components of the local stiffness matrix of
155 CUFSM (see Eqs. (9) and (10) of [13], respectively), which are given by 4x4
matrices defined in terms of general longitudinal functions Ym and Yn , on a 2x2
grid, given in Eq. (13), which defines the longitudinal functions to be integrated
in each quadrant for the SAFSM utilised herein.
   
(Ym Yn )11 (Ym Yn )12 c c −cm sn
 = m n  (13)
(Ym Yn )21 (Ym Yn )22 −sm cn sm sn

When the integrals over the strip volume in Eq. (1) are carried out, the pairs of
longitudinal functions in Eq. (13) are integrated over the strip length L. The
values of these integrals for the various m, n that may be analysed are given in
Eq. (14).

Z L Z L L/2, m = n

cm cn dy = sm sn dy = ,
0 0 0, m 6= n


Z L  L · 2m , m + n even

2 2
sm cn dy = π m − n (14)
0 
0, m + n odd

The non-zero values of the integrals of cm sn and sm cn provide coupling between

160 terms of different numbers of half-wavelengths longitudinally. It is this coupling
that makes it possible to analyse members under shear with simply-supported
ends using the SAFSM [24].
For the stability matrix the permitted stress state, consisting of longitudinal
normal stresses that may vary linearly across the strip width, uniform transverse
165 normal stress and uniform shear stress, is illustrated in Fig. 2. Note that the ap-
plied stresses are all assumed to be uniform longitudinally. The nonlinear strains

8
 considered in formulating the stability matrix are the nonlinear components of
the Green-Lagrange strains. In most FSM formulations, all of the nonlinear
terms due to flexural displacements are included, while of the nonlinear terms
170 due to membrane displacements, only those corresponding to the longitudinal
normal strain are included. In Mahendran and Murray [23], for each nonlinear
strain, the terms associated with membrane displacements in the direction(s) of
that strain were neglected. Herein, none of the nonlinear strain components are
neglected. As the linear components of the Green-Lagrange strains have alre-
175 ady been considered in the membrane components of the linear strains utilised
in formulating the local stiffness matrix, the Green-Lagrange strains are fully
incorporated in this SAFSM.
Derivation of the local stability matrices is again given in detail in [25]. As
for the local stiffness matrix, the local stability matrices for the different applied
180 stresses may be expressed in block diagonal form by two 8x8 matrices; one each
for the potential energy stored by the membrane and the flexural displacements
induced by the relevant applied stresses as the member elastically buckles. Each
8x8 matrix may be further partitioned as a 4x4 matrix repeated on the 2x2
grid of Eq. (13). For longitudinal normal stresses, the 4x4 stability matrices
185 derived from the work done by the membrane and the flexural displacements
are identical to those implemented in CUFSM; see Eqs. (11) and (12) of [13]
respectively. For transverse normal stresses, the 4x4 stability matrices derived
from the work done by the membrane and the flexural displacements are as given
in Eqs. (15) and (16) respectively. For shear stresses, the 4x4 stability matrices
190 derived from the work done by the membrane and flexural displacements are
as given in Eqs. (17) and (18) respectively. In Eqs. (15) – (18) a prime
represents differentiation with respect to the longitudinal direction and the term
βm = mπ/L arises due to these differentiations [13]. Note that the addition of
the nonlinear shear strains due to membrane displacements has negligible effect

9
195 on the analyses carried out in Section 4 of this paper.
 
Ym Yn 0 −Ym Yn 0
 
Ym0 Yn0 Ym0 Yn0 
 

Z L 0 0 − 
σx t  βm βn β m βn  dy
 (15)
b 0  −Ym Yn 0 Ym Yn 0 

 
 
 Ym0 Yn0 0
Ym Yn 0 
0 − 0
βm βn βm βn
 6 1 6 1 
−
 5b 10 5b 10 
 1 2b 1 b 
 
 − − Z L
σx t  10 15 10 30  Y Y dy (16)

 6 1 6 1 0 m n

− − − 
 5b 10 5b 10 
1 b 1 2b
 
− −
10 30 10 15
 
−Y Y 0 − Ym0 Yn 0 −Ym Yn0 + Ym0 Yn 0
 m n 
−Ym0 Yn00 − Ym00 Yn0 −Ym0 Yn00 + Ym00 Yn0 
 

Z L 0 0 
τxy t 
 βm βn βm βn 
 dy
2 0   Ym Yn0 − Ym0 Yn 0 Ym Yn0 + Ym0 Yn 0


 
 
 Ym0 Yn00 − Ym00 Yn0 Ym0 Yn00 + Ym00 Yn0 
0 0
βm βn βm βn
(17)

1 b 1 b
 
− (Ym Yn0 + Ym0 Yn ) − (Ym Yn0 − Ym0 Yn ) − (Ym Yn0 − Ym0 Yn ) (Ym Yn0 − Ym0 Yn )
 2 10 2 10 
 2

 b b b 
0 0 0 0 0 0
 10 (Ym Yn − Ym Yn ) 0 − (Ym Yn − Ym Yn ) (Ym Yn − Ym Yn ) 
Z L
τxy t  10 60 
 dy
 1 b 1 b 
0  0 0 0 0 0 0 0 0
 2 (Ym Yn − Ym Yn ) 10
(Ym Yn − Ym Yn )
2
(Ym Yn + Ym Yn ) − (Ym Yn − Ym Yn )
10 
 
b b2 b
 
0 0 0 0 0 0
− (Ym Yn − Ym Yn ) − (Ym Yn − Ym Yn ) (Ym Yn − Ym Yn ) 0
10 60 10
(18)
Once the local stiffness and stability matrices are assembled for a strip,
200 they are transformed to global coordinates by application of the transformation
matrix T, given in Eq. (19), where c = cos (α) and s = sin (α), where α is the

10
 angle about the longitudinal axis between the global coordinate system and the
local strip coordinate system, taken as positive when clockwise. (Note that the
longitudinal axes of the two coordinate systems are parallel). Transforming a
205 local matrix k to global coordinates is achieved by TT kT.
  h i
T1 −T2 T1 = diag c 1 c 1 c 1 c 1
T= , h i (19)
T2 T1 T2 = diag s 0 s 0 s 0 s 0

Once in global coordinates, these matrices are assigned into the global stiff-
ness KE and stability KG matrices, from which the eigenvalue problem of Eq.
(3) is then formulated and solved. Due to the addition of a second set of de-
grees of freedom, the dimensions of the matrices utilised herein are twice that
210 of those obtained in most other SAFSM formulations; the local stiffness and
stability matrices are each 16x16 and the global stiffness and stability matrices
are (8nq)x(8nq), where n is the number of nodal lines in the cross-section model
and q is the number of series terms being considered.

3. Application of the generalised cFSM

215 3.1. Overview of the generalised cFSM [21, 22]

3.1.1. Fundamentals
The basic concept of the constrained finite strip method is that any general
FSM displacement field d may be transformed to a constrained deformation
space by use of a constraint matrix RM , whose columns are base vectors of the
220 constrained space. The original vector and that in the constrained deformation
space (dM ) are related by,
d = RM dM . (20)

The bolded subscript ‘M’ refers to a given deformation space or mode space. By
applying this transformation to the eigenvalue problem of the SAFSM, modal
decomposition is achieved in that the resulting eigenmodes are constrained to
225 the desired deformation space. The resulting eigenvalue problem is as given in
Eq. (21). As the constraint matrices act to reduce the size of the problem, KE,M

11
 and KG,M are reduced-size global stiffness and stability matrices, particular to
the current modal space.

RT T


M KE RM − ΛM RM KG RM ΘM = 0 → (KE,M − ΛM KG,M ) ΘM = 0 (21)

Modal identification is also possible using the constraint matrices of cFSM, in

230 which a general deformation is expressed as a linear combination of modes.
By appropriately combining the base vectors of various constraint matrices, an
alternate basis R may be formed for the FSM deformation space; transforming
a general deformation into this basis is then achieved by,

d = Rc (22)

where c is a vector containing the coefficients of the linear combination of the

235 basis vectors of R which constitute d. The linear combination coefficients are
dependent on the exact form of the basis vectors, including their normalisation
and whether or not they are orthogonal [26]. From these coefficients, participa-
tion factors for individual modes or modal spaces may be obtained by comparing
the norm of the respective coefficients to the norm of c.

240 3.1.2. Definitions of the modal spaces

In the original development of the cFSM [9, 10, 11], four main modal spaces
were observed: global (G), distortional (D), local (L) and combined shear (S)
and transverse extension (T), also denoted ‘other’ (O). Note that the shear
modes are those associated with non-zero membrane shear strains, as opposed
245 to buckling under the action of shear stresses. In the generalised cFSM these
modal spaces are split into subspaces, as given in [21]. A brief description
of these modal spaces and some of the subspaces follows. In addition to the
following, note that the global (G), distortional (D) and local (L) (sub)spaces
all have null transverse extensions (i.e. null transverse normal strains) and null
250 in-plane shear strains, while the shear (S) spaces have null transverse extensions.

• The global space G, which involves rigid-body transverse displacements

and associated warping displacements. Its subspaces are the global axial

12
 space GA , defined by uniform warping displacements only, and the global
bending GB and torsion GT spaces, which respectively contain the pure
255 flexural and torsional buckling modes of flexural-torsional buckling theory.
The mode GT does not exist if the section has closed parts, as torsion
without in-plane shear strains is impossible in such sections.

• The distortional space D, where the section may distort in such a way that
transverse equilibrium is satisfied.

260 • The local space L, where warping displacements are null; deformations are
similar to local-plate buckling.

• The shear space S, characterised by non-null in-plane shear strains. Its

subspaces can be divided into warping-only shear spaces Sw and transverse-
only shear spaces St . These two groups both have subspaces that contain
265 warping or transverse displacements identical to those of the GB , GT and
D spaces, as well as subspaces that contain any remaining modes. Note
that the transverse-only shear torsion mode STt exists regardless of whet-
her or not GT does.

• The transverse extension space T, characterised by zero warping and non-

270 zero transverse extension.

Note that the GB , GT and D spaces are each linearly dependent with their
respective warping-only and transverse-only shear subspaces. As such, in each
trio of subspaces, only two may be selected when constructing an alternate FSM
basis in order to ensure linear independence. Many more details of the various
275 subspaces may be found in the work of Ádány and Schafer [9, 10, 11, 21, 22]
and Ádány [8, 27].
The processes in developing the constraint matrices for the individual modal
(sub)spaces are not covered here, as they are fully described in [21, 22]; the only
difference in this work is that, due to the shape functions utilised, any terms
280 involving rotational degrees of freedom are the negative of what they are in the
equations of [21, 22]. Note that, although the modal spaces are defined in a

13
 very particular way in the generalised cFSM, any mechanical quantity can be
restrained, as long as an appropriate matrix equation of the form Ad = 0 can be
formulated, where the vector d contains the degrees of freedom of the analysed
285 problem.

3.2. Application to an augmented displacement field

Formulation of the constraint matrices requires consideration of the displa-

cement fields utilised in the FSM to which it is being applied. Hence, as the
augmented displacement field of sines and cosines utilised herein differs from
290 that of the traditional SAFSM, which utilises only sines for transverse displa-
cements and only cosines for warping displacements, it should be necessary to
re-derive these matrices for this augmented displacement field. Fortunately, due
to both the orthogonality properties of sines and cosines and that the various
mechanical criteria considered in [21, 22] when deriving the constraint matrices
295 are linear, the process is as simple as applying the constraint process outlined in
[21, 22] separately to the sets of ‘real’ and ‘imaginary’ degrees of freedom. That
is, if one were to carry out the process for the displacement field of the traditio-
nal SAFSM, the resultant constraint matrices may be denoted RiM , as all of the
real degrees of freedom considered herein will be zero and hence the imaginary
300 degrees of freedom will completely define the base vectors. Consequently, the
constraint matrix of the augmented displacement field is simply,
h i
RM = RrM RiM (23)

where RrM is identical to RiM except that the values of the real and imaginary
degrees of freedom are swapped. Similarly, the orthogonality between sines of
different numbers of half-wavelengths (and also of such cosines) means that the
305 constraint matrix for multiple series terms is simply a block diagonal matrix
consisting of constraint matrices developed separately for each number of half-
wavelengths [25].
While it is easy, mathematically, to show the above for most of the mechani-
cal criteria utilised in the formulation of the constraint matrices, one particular

14
310 criterion - the case of transverse equilibrium - warrants a more intuitive expla-
nation, which is now laid out. Consider an augmented displacement field of
sines and cosines that satisfies transverse equilibrium at all points longitudi-
nally. No matter the number of half-wavelengths being considered, there will
be points where one of the sine or cosine is zero yet the other is non-zero. At
315 these points, the non-zero component of the displacement field must then be in
transverse equilibrium, regardless of the other component. For this component
however, only the amplitude of transverse displacements varies along the length,
not their distribution around the section; hence, this component must satisfy
transverse equilibrium at every point longitudinally, independently of the other
320 component, for which the same is also true. As such, transverse equilibrium of
the augmented displacement field, like the other mechanical criteria, is enfor-
ced by applying the corresponding matrix equation separately to the ‘real’ and
‘imaginary’ degrees of freedom.

4. Numerical examples

325 To illustrate the application of the constrained finite strip method to the
analysis of members in shear, a lipped channel (LC) section and a rectangu-
lar hollow flange (RHF) section are analysed under shear loading through the
sections’ respective shear centres and in a direction parallel to the web. The
shear flow distribution around the sections due to the applied shear loading
330 are as shown in Fig. 3; each section has a web depth (upper to lower flange
depth) of 200 mm, flange width of 80 mm, lip depth of 20 mm and uniform
element thickness of 2 mm. In the lipped channel section, the shear flow runs
counter-clockwise around the section from the tip of the upper lip to the tip of
the lower lip. The shear flow around the inner flanges and web of the rectangu-
335 lar hollow flange section are qualitatively similar to that for the lipped channel
section, but is slightly more complicated within the closed loops. The shear flow
for this section runs from the upper right corner of the section, down and left
into the web, before continuing down and right to the lower right corner of the

15
 section. (The point in the upper and lower loops where the shear flow changes
340 direction are actually just to the left of their respective upper right and lower
right corners). Note that, as the longitudinal distribution of the applied loads
is uniform in the utilised SAFSM, it is not possible to replicate the moment
gradient that arises when a section is subjected to a shearing force, as is being
done here. As such, only shear stresses are applied to the sections and they
345 are in a non-equilibrium state of ‘pure’ shear. The lipped channel under shear
loading parallel to the web has been analysed extensively by Pham and Hancock
[28], while hollow flange sections in shear have been analysed by Keerthan and
Mahendran [29, 30] using the finite element method, but have yet to be analysed
using the SAFSM.
350 The uniform shear stress in each strip is generated by calculating the shear
stresses around the section due to the applied loading and then taking the
average across each strip. As such, while the shear stress is uniform within each
strip, it may vary between strips. The signature curves of these sections are
generated by performing analyses for only a single longitudinal half-wavelength
355 (i.e. m = n = 1) and are shown in Fig. 4. Typical values of Young’s modulus
and Poisson’s ratio for steel, 200 GPa and 0.3 respectively, were assumed. For
each section, a traditional FSM signature curve is generated as well as three
modal signature curves that consider only global (G), distortional (D) and local
(L) buckling modes respectively. The global torsion GT mode does not exist
360 for the rectangular hollow flange section, due to the closed loops, and so the
transverse-only shear torsion STt mode has been substituted in its place to
ensure that rigid-body torsional displacements are included, as was done in
[22]. The base vectors for either section are not illustrated in this paper; those
for the rectangular hollow flange section are entirely given in [22]. Those for the
365 lipped channel section have not been presented in the same format as for the
rectangular hollow flange section in [22], but can be found in various forms in a
number of papers (e.g. Figs. 6 – 9 of [31]).
The traditional FSM signature curves are qualitatively similar for each section,
coming to a minimum at a half-wavelength close to the clear width of the web,

16
370 where the buckling mode is local in nature, before rising to a peak with buckling
in distortional modes and then dropping off as longer half-wavelengths are ap-
proached and global modes become critical. Quantitatively, the FSM signature
curves reveal that the critical local buckling shear stress obtained at short half-
wavelengths for the rectangular hollow flange section is approximately 75% gre-
375 ater than that obtained for the lipped channel. Further, the half-wavelength at
which this minimum occurs shifts from 200 mm to 140 mm due to the lesser clear
width of the web for the rectangular hollow flange section and also the greater
torsional restraint provided by the closed loops of the flanges. The critical stress
at the subsequent peak in the FSM signature curve is greater (by approximately
380 188%) for the rectangular hollow flange section than the lipped channel, again
due to the loops of the flanges providing significant torsional resistance. This
relative difference becomes even more pronounced as long half-wavelengths are
approached.
For both sections, the local modal signature curves are practically coincident
385 with the FSM signature curve at short half-wavelengths, with the modal and
FSM curves diverging after the local minimum is achieved. The distortional and
global modal signature curves reveal significantly different behaviours relative
to their respective FSM signature curves. For the lipped channel, the minimum
of the distortional solution is approximately 2.5 times greater than the critical
390 stress of the FSM solution at the same half-wavelength, likely due to the lower
stresses in the flanges relative to the web rendering it more difficult for the
section to buckle in a purely distortional manner. This difference is much more
pronounced than for sections analysed in compression or bending [11, 22] and is
even more noticeable for the rectangular hollow flange section, which displays a
395 monumental difference between the minimum of its distortional solution and the
corresponding critical stress of its FSM solution. The global solution displays
a more standard behaviour for the lipped channel, with the solution becoming
coincident with the FSM solution at long half-wavelengths. The rectangular hol-
low flange section does not experience this, with significant differences between
400 the FSM and global solutions even at very long half-wavelengths.

17
 The modal solution behaviours of the two sections may be made somew-
hat more evident by examining the buckling modes at various half-wavelengths.
The buckling modes obtained at short (200 mm), moderate (900 mm) and long
(3500 mm) half-wavelengths for the lipped channel via the SAFSM are shown
405 in Fig. 5 and are labeled as local, distortional and global buckling modes re-
spectively. The buckling modes at the same half-wavelengths for the rectangular
hollow flange section are shown in Fig. 6, although they are unlabeled as their
behaviour is decidedly more complex at moderate and long half-wavelengths.
Considering Fig. 5b, while the mode is largely a symmetric distortional one,
410 the curvature of the web in particular is greater than would be expected for a
pure distortional mode, implying the presence of local deformations and explai-
ning the large difference between the distortional and FSM solutions for this
section.
From Fig. 6b, it is evident that the buckling behaviour of the rectangular
415 hollow flange section at moderate half-wavelengths still involves a large amount
of local deformations, in conjunction with distortional deformations involving
lateral displacements of the hollow flanges. The long half-wavelength mode of
the rectangular hollow flange section (Fig. 6c) is somewhat similar to that of
the flexural-torsional mode of the lipped channel in Fig. 5c, although there is
420 evidently still some transverse web flexure due to lateral deflection of the hollow
flanges, as the torsional stiffness of the hollow flanges discourages overall torsion
of the section. This distortional mode, involving lateral displacements of the
hollow flanges, may be deemed a lateral distortional mode and arises due to the
addition of the closed loops in the section. As discussed in [11], the global and
425 distortional modes are uniquely defined by their warping profiles; the addition
of the closed loops precludes the existence of a GT mode, however its associated
warping profile may, potentially, be nearly replicated by a shear strain-free mode
if the condition of rigid-body displacements (i.e. null transverse curvature) is
relaxed. This allows the warping profile to be nearly replicated by a distortional
430 mode — in this case, the lateral distortional (DLat ) mode — which does not
exist for the open section. To illustrate, the (normalised) warping displacement

18
 profiles of the GT mode for the lipped channel and the DLat mode for the
rectangular hollow flange section are shown in Fig. 7. For completeness, the
transverse displacements are also shown. The warping profile does not appear
435 to be perfectly replicated, but it is very similar.
Visually identifying the dominant modes involved in the buckling of a mem-
ber in shear may be aided by considering that the utilised SAFSM produces
buckling modes that vary longitudinally according to a combination of sines
and cosines. Since the degrees of freedom corresponding to these two groups
440 of longitudinal variations are explicitly separated in the formulation, it is sim-
ple to decompose the buckling mode into its components corresponding to these
two groups of longitudinal variations. For the signature curves generated, where
m = n = 1, it is not necessary to do this by noting that the cross-sectional trans-
verse displacements of the buckling mode at the ends of the half-wavelength are
445 purely due to the one half-wave cosine, while those at the middle are purely
due to the one half-wave sine. This suggests that, in this case, it is possible
to examine the cross-sectional transverse displacements of the buckling mode
at these locations to visualise the two modes (corresponding to a cosine and
a sine) that constitute the ‘full’ buckling mode. However, as the utilised har-
450 monic functions correspond to examining an internal portion of an infinitely
long member, it is clear that the exact portion of the member to which the
eigenvectors correspond to is arbitrary. That is, the exact half-wavelength of
the buckling modes that is being examined may be phase-shifted up and down
the length of the member without any effect on the eigenvalues/load factors
455 and so the cross-sectional displacements at the ends and middle of the half-
wavelength are not unique. Despite this, it is clear from the nature of a one
half-wave sine and cosine (i.e. they are identical except for a phase-shift of
π/2, or one quarter-wavelength) that the buckling mode for an infinitely-long
member in shear may be pseudo-decomposed into two modes by examining the
460 transverse displacements at cross-sections separated by one quarter-wavelength.
An example of this is shown in Fig. 8, which shows the buckling mode for the
rectangular hollow flange section at a half-wavelength of 2000 mm and the trans-

19
 verse displacements of two cross-sections separated by one quarter-wavelength.
At cross-section “a)” the transverse displacements correspond to that of the
465 lateral distortional mode with a small amount of rigid-body rotation, while at
“b)” the transverse displacements correspond to a symmetric distortional mode
with some additional local buckling in the web, as well as flexural displacements
in a direction perpendicular to the web. Note that these two ‘modes’ are not
the same; indeed, they cannot be for a member in shear, as then it would not
470 be possible for the phase-shift of displacements around the section, which is a
key characteristic of members loaded with shear stresses, to occur. This may be
easily seen by setting the corresponding real and imaginary degrees of freedom
to be equal to one another in Eqs. (7) – (9).
A number of prior observations may be made more explicit by carrying out
475 modal identification. In carrying out the identification process, the base vectors
were normalised such that the greatest magnitude of any degree of freedom was
unity (vector norm) and the participation results obtained for the individual
mode vectors were combined into participation for the five modal spaces using
the l2 -norm (Euclidean norm). These choices for normalisation and norm were
480 made based on the work of Li, Hanna, Ádány and Schafer [26]. As mentioned
previously, three triplets of subspaces are linearly dependent, meaning only two
may be selected from each triplet; here, the shear strain-free subspaces and the
corresponding transverse-only shear subspaces were utilised. The identification
results are given in Fig. 9.
485 The modal identification results for the lipped channel are fairly typical of
open sections analysed using cFSM. At short half-wavelengths, the results indi-
cate essentially pure local participation, which smoothly but rapidly transitions
to more distortional participation as the half-wavelength is increased. At mode-
rate half-wavelengths the distortional participation peaks at approximately 67%,
490 with small contributions of both local and global modes. The contributions of
the local modes were previously indicated in regards to Fig. 5b. This less-than-
complete distortional modal participation explains, to a degree, the discrepancy
between the distortional and FSM solutions. As the half-wavelength increases

20
 further, distortional participation drops off sharply as the modal participation
495 becomes essentially purely global in nature. Shear and transverse extension
participation are negligible at all analysed half-wavelengths.
The results for the rectangular hollow flange section are as expected at
short half-wavelengths, with essentially pure local participation. As the half-
wavelength increases, distortional participation increases and peaks at a longer
500 half-wavelength than for the lipped channel section, before dropping off again.
It remains fairly significant at very long half-wavelengths, with approximately
10% participation. Unlike the lipped channel, the presence of the tubular flan-
ges results in the development of significant shear strains and hence the shear
mode participation is evident over a wide range of half-wavelengths. The dip
505 in the shear mode participation, which coincides with the peak in the distortio-
nal participation, may indicate a switch in which shear modes (warping-only or
transverse-only) are taking part in the overall buckling mode. While the global
participation does increase as the half-wavelength becomes very long, it is less
marked than for the lipped channel section due to the significant distortional
510 and shear participation at these half-wavelengths.
To fully highlight the complexity of the rectangular hollow flange section, it
was found that it was necessary to consider a number of modes for the FSM
and modal solutions to be within 5% of one another at very long (>10000
mm) half-wavelengths: the global bending (GB ) and lateral distortional (DLat )
515 modes and their correpsonding transverse-only shear modes, the transverse-only
shear torsion (STt ) mode, the warping-only shear modes corresponding to the
symmetric and anti-symmetric distortional modes and two more warping-only
shear modes that are the ‘remaining’ modes of the warping-only shear space.
Comparing the identification results at a half-wavelength of 2000 mm to the
520 buckling mode shown in Fig. 8 reveals that the previously identified modes —
flexural (GB ), symmetric and lateral distortional (D), local in the web (L) and
involving rigid-body rotation (S; the mode here is STt as GT does not exist due
to the closed loops) — match with the identification results of Fig. 9, which
have quantified the degree to which the pure modes participate in the overall

21
525 buckling mode.

5. Conclusions

The essentials of a semi-analytical finite strip method (SAFSM) for analy-

sis of members in general loading including shear has been presented, utilising
an augmented displacement field of sines and cosines to allow the phase-shift
530 of displacements due to shear to be incorporated. All of the Green’s strains
were included in formulating the geometric stability matrix, where the nonli-
near membrane shear strains and nonlinear membrane transverse strains are
usually neglected in the formulation. The generalised constrained finite strip
method was then summarised and its application to the developed SAFSM was
535 elucidated. Signature curves were produced for a lipped channel section and a
similarly-sized rectangular hollow flange section, both in shear loading parallel
to the web; typical buckling modes were also shown and elucidated. Modal
identification was then performed for the sections analysed and used to further
explore the nature of the buckling modes obtained by the SAFSM.

540 Acknowledgements

The authors are grateful for the work of Prof. Ben Schafer, Assoc. Prof.
Sándor Ádány, Assoc. Prof. Zhanjie Li and others in developing the open-
source finite strip software CUFSM [13], whose plotting codes were utilised to
produce the cross-sections displayed in this paper. Thanks are also extended to
545 Mr. Song Hong Pham and Mr. Van Vinh Nguyen for providing the graphics
program used to generate the 3D buckling modes in this paper.

References

[1] Y. K. Cheung, Finite strip method in the analysis of elastic plates with
two opposite ends, Proceedings of the Institution of Civil Engineers 40 (1)
550 (1968) 1–7.

22
 [2] J. Przemieniecki, Finite element structural analysis of local instability,
AIAA Journal 11 (1) (1973) 33–39. doi:10.2514/6.1972-354.
URL http://arc.aiaa.org/doi/abs/10.2514/6.1972-354

[3] R. J. Plank, W. H. Wittrick, Buckling under combined loading of thin, flat-

555 walled structures by a complex finite strip method, International Journal
for Numerical Methods in Engineering 8 (2) (1974) 323–339.

[4] G. J. Hancock, Local distortional and lateral buckling of I-beams, Journal

of the Structural Divison, Proceedings of the ASCE 104 (11) (1978) 1787–
1798.

560 [5] B. W. Schafer, T. Peköz, Direct strength prediction of cold-formed steel

members using numerical elastic buckling solutions, Fourteenth Internati-
onal Specialty Conference on Cold-Formed Steel Structures: Recent Re-
search and Developments in Cold-Formed Steel Design and Construction
(1998) 69–76.

565 [6] American Iron and Steel Institute, North American specification for the
design of cold-formed steel structural members AISI S100-2012 (2012).

[7] Standards Australia, AS/NZS 4600 Cold-Formed Steel Structures (2005).

[8] S. Ádány, Buckling mode classification of members with open thin-walled

cross-sections by using the finite strip method, Tech. rep., Johns Hopkins
570 University, Baltimore, Maryland, USA (2004).

[9] S. Ádány, B. W. Schafer, Buckling mode decomposition of single-branched

open cross-section members via finite strip method: Derivation, Thin-
Walled Structures 44 (5) (2006) 563–584. doi:10.1016/j.tws.2006.03.
014.

575 [10] S. Ádány, B. W. Schafer, Buckling mode decomposition of single-branched

open cross-section members via finite strip method: Application and ex-
amples, Thin-Walled Structures 44 (5) (2006) 585–600. doi:10.1016/j.
tws.2006.03.014.

23
 [11] S. Ádány, B. W. Schafer, A full modal decomposition of thin-walled,
580 single-branched open cross-section members via the constrained finite strip
method, Journal of Constructional Steel Research 64 (1) (2008) 12–29.
doi:10.1016/j.jcsr.2007.04.004.

[12] R. Schardt, Verallgameinerte Technische Biegetheorie, Springer-Verlag,

Berlin, 1989.

585 [13] Z. Li, B. W. Schafer, Buckling analysis of cold-formed steel members with
general boundary conditions using CUFSM: Conventional and constrained
finite strip methods, in: 20th International Specialty Conference on Cold-
Formed Steel Structures, 2010, pp. 17–31. doi:10.1016/j.tws.2006.03.
013.

590 [14] Z. Beregszászi, S. Ádány, Application of the constrained finite strip method
for the buckling design of cold-formed steel columns and beams via the
direct strength method, Computers and Structures 89 (21-22) (2011) 2020–
2027. doi:10.1016/j.compstruc.2011.07.003.

[15] Z. Li, B. W. Schafer, Application of the finite strip method in cold-formed

595 steel member design, Journal of Constructional Steel Research 66 (8-9)
(2010) 971–980. doi:10.1016/j.jcsr.2010.04.001.

[16] S. Ádány, Z. Beregszászi, Modal decomposition for thin-walled members

with rounded corners: an extension to cFSM by using elastic corner ele-
ments, in: 8th International Conference on Advances in Steel Structures,
600 Lisbon, Portugal, 2015.

[17] C. H. Pham, G. J. Hancock, Direct strength design of cold-formed C-

sections for shear and combined actions, Journal of Structural Engineer-
ing 138 (6) (2012) 759–768. doi:10.1061/(ASCE)0733-9445(2009)135:
3(229).

605 [18] G. J. Hancock, C. H. Pham, A signature curve for cold-formed channel

24
 sections in pure shear, Tech. Rep. R919, The University of Sydney, Sydney,
Australia (2011).

[19] G. J. Hancock, C. H. Pham, Direct strength method of design for shear of

cold-formed channels based on a shear signature curve, in: 21st Internati-
610 onal Specialty Conference on Cold-Formed Steel Structures, 2012.

[20] S. H. Pham, C. H. Pham, G. J. Hancock, Direct strength method of design

for shear including sections with longitudinal web stiffeners, Thin-Walled
Structures 81 (2014) 19–28. doi:10.1016/j.tws.2013.09.002.

[21] S. Ádány, B. W. Schafer, Generalized constrained finite strip method for

615 thin-walled members with arbitrary cross-section: Primary modes, Thin-
Walled Structures 84 (2014) 150–169. doi:10.1016/j.tws.2014.06.001.

[22] S. Ádány, B. W. Schafer, Generalized constrained finite strip method for

thin-walled members with arbitrary cross-section: Secondary modes, ort-
hogonality, examples, Thin-Walled Structures 84 (2014) 123–133. doi:
620 10.1016/j.tws.2014.06.001.

[23] M. Mahendran, N. W. Murray, Elastic buckling analysis of ideal thin-walled

structures under combined loading using a finite strip method, Thin-Walled
Structures 4 (5) (1986) 329–362. doi:10.1016/0263-8231(86)90029-7.

[24] G. J. Hancock, C. H. Pham, Shear buckling of channel sections with sim-

625 ply supported ends using the Semi-Analytical Finite Strip Method, Thin-
Walled Structures 71 (2013) 72–80. doi:10.1016/j.tws.2013.05.004.

[25] M. A. Rendall, G. J. Hancock, K. J. R. Rasmussen, The generalised con-

strained finite strip method for thin-walled, prismatic members under app-
lied shear, Tech. rep., The University of Sydney, Sydney, Australia (2016).

630 [26] Z. Li, M. T. Hanna, S. Ádány, B. W. Schafer, Impact of basis, orthogo-

nalization, and normalization on the constrained Finite Strip Method for
stability solutions of open thin-walled members, Thin-Walled Structures
49 (9) (2011) 1108–1122. doi:10.1016/j.tws.2011.04.003.

25
 [27] S. Ádány, Decomposition of in-plane shear in thin-walled members, Thin-
635 Walled Structures 73 (2013) 27–38. doi:10.1016/j.tws.2013.06.003.

[28] C. H. Pham, G. J. Hancock, Shear buckling of channel sections with simply

supported ends using the semi-analytical finite strip method, Tech. rep.,
The University of Sydney, Sydney, Australia (2013). doi:10.1016/j.jcsr.
2008.05.015.

640 [29] P. Keerthan, M. Mahendran, Elastic shear buckling characteristics of Li-

teSteel beams, Journal of Constructional Steel Research 66 (11) (2010)
1309–1319. doi:10.1016/j.jcsr.2010.05.005.

[30] P. Keerthan, M. Mahendran, Improved shear design rules of cold-formed

steel beams, Engineering Structures 99 (2015) 603–615. doi:10.1016/j.
645 engstruct.2015.04.027.

[31] N. Silvestre, D. Camotim, Nonlinear generalized beam theory for cold-

formed steel members, International Journal of Structural Stability and
Dynamics 3 (4) (2003) 461–490.

List of figures

650 Figure 1: Local degrees of freedom and axes of a finite strip

Figure 2: Permitted stress state on a strip
Figure 3: Shear flow distribution around the 200x80x20 lipped channel and rec-
tangular hollow flange sections to be analysed
Figure 4: FSM and cFSM modal signature curves for a lipped channel and rec-
655 tangular hollow flange section in shear loading parallel to the web
Figure 5: Buckling modes of a lipped channel section in shear loading parallel
to the web, at half-wavelengths of a) 200 mm, b) 900 mm and c) 3500 mm
Figure 6: Buckling modes of a rectangular hollow flange section in shear loading
parallel to the web, at half-wavelengths of a) 200 mm, b) 900 mm and c) 3500
660 mm

26
 Figure 7: Transverse (subscript t) and warping (subscript w) displacement pro-
files for GT mode of a lipped channel and DLat mode of a rectangular hollow
flange section
Figure 8: Buckling mode of a rectangular hollow flange section in shear loading
665 parallel to the web at a half-wavelength of 2000 mm and cross-section deforma-
tions separated by one quarter-wavelength
Figure 9: Modal identification results for a lipped channel and rectangular hol-
low flange section in shear loading parallel to the web

27