Journal of Constructional Steel Research 148 (2018) 295–303

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Use of artificial damping factors to enhance numerical stability for

irregular joints
J.D. Nzabonimpa, Won-Kee Hong ⁎
Department of Architectural Engineering, Kyung Hee University, 1732, Deogyeong-daero, Giheung-gu, Yongin-si, Gyeonggi-do, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Article history: The authors previously proposed a fast and easy assembly method for wide flange steel column-beam frames in
Received 29 August 2017 which the column and beam steel web had skew cuts and the lower flange of the steel beam was partially re-
Received in revised form 8 May 2018 moved to prevent the bottom flange of the steel beam from running into the flanges of the L-shaped guide angles
Accepted 29 May 2018
(which were pre-installed on the column bracket) when they were erected. However, the joint experienced con-
Available online xxxx
vergence difficulties at the irregular structural configuration of the skewed web connections when cyclic finite
Keywords:
element analysis was performed. This study introduced artificial damping factors, implemented during nonlinear
Beam-column joint numerical computation, to balance the internal and external forces. The influence of the damping factors was ex-
Fast assembly plored to ameliorate the numerical instabilities by removing the aforementioned convergence difficulties. It was
Skewed web connection shown that a solution converged while the dissipated stabilization energy was sufficiently small, ensuring that
Damping factor the analysis results were not distorted by the use of damping factors in the range of 1.0 × 10−5 to 5.0 × 10−5.
Numerical instability The best matches with monotonic runs and the convergence for steels with yield strength of 325 MPa were ob-
Precast members tained by applying damping factors of 0.0002 and 0.0005.
© 2018 Elsevier Ltd. All rights reserved.

1. Introduction; previous studies and significance of this study
Extensive analytical and experimental investigations have been car-
ried out in attempts to understand the structural behavior of steel
beam-column joints subjected to either monotonic or cyclic loads [2,
4, 5, 13]. Using a nonlinear finite element package, Muresan and Balc
[9] investigated the behavior of beam-to-column joints. In their study,
all elements in the tension region of the joint were modeled with equiv-
alent T-stubs. It was found that the joint exhibited increased deforma-
tion around its tension flange. In addition, the issues and challenges
facing numerical simulations of steel bolted joints were also reported
[3]. It was concluded that the analysis of bolted joints depends mainly
on the following variables: constitutive relations, step size, element
type, number of integration points in the elements, kinematic descrip-
tions, and discretization. These elements demonstrated nonlinear be-
havior, which required more attention when dealing with finite
element analysis. Accordingly, previous researchers established simpli-
fied methods for bolt modeling. Due to their relative economy and more
rapid construction, prefabricated structural members were gaining pop-
ularity in the construction industry [6, 10, 11, 14]. In earlier study, the
authors proposed an irregular structural configuration of the skewed
web connection for fast and easy assemble of steel frames. A 3D nonlin-
ear finite element model representing the proposed joint was
⁎ Corresponding author.
E-mail address: hongwk@khu.ac.kr (W.-K. Hong).
performed to better understand the structural behavior of the joint.
However, the convergence difficulties have occurred at the irregular
skewed web, and the numerical instability increased in an unbounded
manner when the viscosity and nonlinearity in the numerical solution
process were neglected. In many engineering fields, the artificial
damping terms (ADT) were used in solving problems including
aeroacoustics, wave and fluid flow problems. Sun et al. [12] developed
methods to facilitate stable and accurate numerical solutions of linear-
ized Euler equations, which were often used in solving problems in
computational aeroacoustics. In their works, two new methods that
used artificial damping terms (ADT) were introduced. One of their
methods was constructed to damp the vortical components generated
during the computation. An artificial damping algorithm for solving
the Helmholtz problem was also considered by Kim and Lee [7]. When
the imaginary part of the wave number was small, the problem was
known to be difficult to solve. In their paper, an efficient artificial
damping algorithm which can be viewed as a rational iteration was pro-
posed. Each damped problem was solved incompletely by a nonover-
lapping domain decomposition method. Liu and Nithiarasu [8]
modeled the upper-convected Maxwell (UCM). In addition to allowing
equal-order interpolations for pressure and velocity, they proposed
method along with an appropriate artificial damping scheme which
was able to produce stable solutions for different Deborah numbers
(De). They showed that an additional damping was essential to main-
tain positive definitiveness of the conformation tensor at higher De.
They also demonstrated the need for an additional damping by

https://doi.org/10.1016/j.jcsr.2018.05.034
0143-974X/© 2018 Elsevier Ltd. All rights reserved.
296 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303
analyzing the basic forward time central space scheme applied to the
constitutive equations.
To overcome these convergence problems and to suppress the
growth of instabilities observed in the proposed irregular joints in the
present study, the artificial damping factors were used during nonlinear
numerical computation in order to balance the internal and external
forces for cyclic load applications. At all nodes, the equilibrium state
with added damping factors was found without distorting the analysis
results which was compared with the numerical solutions obtained
with monolithic loads. Finally, sensitivity analysis of the damping fac-
tors was conducted to investigate their influences on the structural be-
havior of the proposed joint.
2. Numerical evaluation of the proposed connections
2.1. Instabilities in the model with irregular cuts
In this study, a fast and easy assembly method was proposed for col-
umn-beam frames having wide flange steels, in which the column and
beam steel web had skew cuts being placed in the L-shaped guide an-
gles until permanent connections were made by bolting stiffener plates
to obtain a double shear connection. The conventional erection of heavy
frames with heavy stiffeners and multiple bolts was replaced by the
erection with novel connections having skew cuts along the column
and beam steel web. The lower flange of the steel column bracket and
beam was also partially removed to prevent the bottom flange of the
steel beam from running into the flanges of the L-shaped guide angles
(which were pre-installed on the column bracket). These helped the
bottom flange of the steel beam to be placed into L-shaped angles,
which were pre-installed on the column bracket, when they were
erected. This assembly method was proposed for use in the steel con-
struction industry; however, it can also be extended to precast frames,
helping to erect and assemble heavy precast frames. Steel frames
Fig. 1. Steel frames encased in precast concrete with heavy stiffeners and multiple bolts.
encased in precast concrete shown in Fig. 1 were proposed to provide
efficient structural resistance to heavy floor loadings over 20 kN/m2.
Numerical analysis based on cyclic analysis of the connection was
performed to evaluate the cyclic hysteretic behavior of the proposed
connections with L-shaped guide angles in order to verify the structural
stability of the proposed connection. Particular attention was paid to the
influence of the irregular web cuts on the flexural capacity of the joints.
The finite element meshes of the web joint with the skew cut, con-
structed for the proposed connections, and the L-type guide angles
and stiffener plate are shown in Figs. 2 and 3. CD8R elements are utilized
for the steel, stiffeners, and bolts. Table 1 presents the material proper-
ties used in the numerical analysis of the proposed joint. Steel sections
with a tensile yield stress (Fy) of 325 MPa and high-strength bolts
with a tensile yield stress of 900 MPa were used in the numerical
study. A 3D FE model of the proposed joint was discretized using the
ABAQUS modeling tool [1]. Two types of elements (C3D8R and R3D4)
were chosen to represent the joint connection. Elements of type
C3D8R, which are referred to as reduced integration elements, were
preferred because they were suitable for nonlinear static analysis. In ad-
dition, these elements demonstrated similar behavior to that of C3D8 el-
ements, with the exception that the eight integration points were
reduced to a single integration point, decreasing the required running
time. These elements (C3D8R) are applied in the ABAQUS model, as il-
lustrated in Fig. 3. Alternatively, elements of type R3D4 are used to
model a rigid body, as shown in Fig. 3. The rigid body was used in this
analysis to locate the reference point at which a cyclic load was exerted.
Two constrained areas (i.e., fixed conditions), as shown in Fig. 3, are de-
fined at both ends of the steel column in order to restrain the movement
of the model during the application of cyclic loads.
In Fig. 4, numerical errors were identified at nodes #1 and #2 in the
neighborhood of the sharp edges, which prematurely terminated the
numerical computation for the cyclic moment-displacement relation-
ship as represented by Legend 3 (refer to Fig. 6) at a stroke around 60
mm. The irregularities of structural configuration such as sharp edges
may cause convergence difficulties and numerical instability to some
degree. The residual force (Rb) shown in Fig. 5 failed to balance the in-
ternal and external forces. As the analysis proceeded, the cyclic analysis
did not converge, as indicated by the residual force, Rb, which failed to
balance the internal and external forces, as shown in Fig. 5. The influ-
ence of the damping factor on the analysis results and convergence is
shown in Fig. 6, where the computation represented by Legend 3 was
terminated at a stroke around 60 mm without a damping factor. The nu-
merical instabilities were resolved by introducing an artificial damping
factor, where a default damping factor was used to overcome the nu-
merical instability, and obtain cyclic moment-displacement relationship
as represented by Legend 4. However, the flexural strength with artifi-
cial damping factor of 0.0002 was over-estimated compared with that
calculated monotonically.
Fig. 2. Finite element meshes of the proposed joint.
 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303 297

Fig. 3. Element types and member sizes of the proposed joint.

Table 1
Material properties of the connection (input for ABAQUS analysis).

Category Material properties

Beam stiffener plates SM490 Tensile yield stress (Fy): 325 MPa Ultimate stress (Fu): 490 MPa Young's Modulus (E):
2-PL (outer): 405 × 199, t = 9 mm 205,000 MPa
2-PL (inner): 405 × 80, t = 9 mm

Column stiffener plates

4-PL: 330 × 190, t = 30 mm
Steel brackets
2-PL (L-type): 405 × 96 × 347, t = 9 mm
H-steel beam
(Size: H-396 × 199 × 7 × 11 mm)
H-steel column
(size: H-400 × 400 × 20 × 35 mm)
Bolt (M20) F10 T Tensile yield stress (fy): 900 MPa Ultimate stress (fu): 1000 MPa Young's Modulus (E):
206,000 MPa

2.2. Application of artificial damping factors to resolve the stability

problems

In this study, an automatic mechanism was introduced to address

the numerical problems through the automatic addition of volume-pro-
portional damping to the model. In Eq. (1), constant damping was

Fig. 4. Convergence difficulties and numerical instability; locations of numerical

instability. Fig. 5. Iteration in ABAQUS [1].
298 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303

Fig. 6. Cyclic results without damping and with the default damping factor.
Fig. 9. Cyclic analysis results with damping ratios of 1.0 × 10−4 to 2.0 × 10−4.

subsequent steps was adopted to remove potential convergence diffi-

culties [1].

P−I−F v ¼ 0 ð1Þ

The applied damping factors varied with time to account for changes
in the analysis depending on numerical instability. Additional artificial
damping had no effect on the viscous forces; therefore, the amount of
viscous energy dissipated was very small when the model was stable.
However, the local velocities increased, as part of the released strain en-
ergy is dissipated by the applied damping when numerical instability
occurs. In ABAQUS [1], the initial default damping factor (with a value
of 0.0002) was used for automatic stabilization. The convergence of

Fig. 7. The cyclic analysis results with damping ratios of 1.0 × 10−6 to 4.0 × 10−6.

applied to stabilize the nonlinear procedure in the model to resolve nu-

merical instabilities. The viscous forces (Fv = cMv) were added to the
global equilibrium equations where c is a damping factor, M is an artifi-
cial mass matrix calculated with unity density, and v is the vector of
nodal velocities (v = Δu/Δt). The vector of nodal velocities was assumed
to add artificial viscous forces to meet equilibrium at the nodes with nu-
merical instabilities.
ABAQUS recalculates new values for the damping factor based on the
declared damping intensity and on the solution of the first increment of
the step. In this study, the damping factors were modified at the begin-
ning of subsequent steps, whereas automatic stabilization in the

Fig. 8. Cyclic analysis results with damping ratios of 1.0 × 10−5 to 5.0 × 10−5. Fig. 10. The most accurately predicted moment-displacement relationship.
 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303
Fig. 11. Comparison of viscous damping (ALLSD) and total strain energy (ALLIE) [1].
the analysis was improved to a stroke of 95 mm at a damping factor of
0.0002, as shown in the curve represented by Legend 4, Fig. 6. The anal-
ysis continued to a stroke of 110–140 mm in the monolithic analysis. In
some cases, additional factors were considered with adaptive automatic
stabilization when the convergence difficulties were not removed after
applying the initial default damping factor. However, Fig. 6 demon-
strates that the cyclic analysis results with a damping factor of 0.0002
(refer to Legend 4) show a substantially greater value compared with
the results without damping (refer to Legend 3). The computed lateral
moment-resisting capability of the section was greater than that of the
capacity obtained monotonically, indicating that the default damping
factor was too high. Therefore, obtaining an optimal value for the
damping factor was a manual process that required the use of trial
and error until a converged solution was obtained. However, the dissi-
pated stabilization energy (ALLSD) should be sufficiently small. Fig. 7
presents the cyclic analysis results with damping factors of 1.0 × 10−6
to 4.0 × 10−6. Computation was extended beyond that without a
damping factor, but the instability of the numerical calculation in the
positive regime did not subside, indicating that the damping factors
should be increased.
Damping factors of 1.0 × 10−5 to 5.0 × 10−5 were implemented as
shown in Fig. 8 so that the numerical instability subsided, whereas the
damping factors determined in such a way that the dissipated energy
for a given increment similar to the first increment was a small fraction
of the strain energy which was applied at nodes.
Fig. 9 exhibits the cyclic analysis results with damping ratios of 1.0 ×
10−4 to 2.0 × 10−4, where the energy dissipated by viscous damping
(ALLSD) increases due to numerical instability. These analysis results
Fig. 12. Deformed mesh and stress distribution of the connection with damping factor of 0.00005.
299
can be distorted when the energy dissipated by the viscous damping
is too large relative to the total strain energy, as shown in the mo-
ment-displacement relationship indicated by Legend 4 of Fig. 6 and by
Legends 4 and 5 of Fig. 9.
The moment-displacement relationship yielded by the last cycle
(highlighted by a solid line represented by Legend 5) with a damping
factor of 5.0 × 10−5 most accurately matched the monotonic moment-
displacement relationship, as shown in Fig. 10(a). Fig. 10(b) illustrates
the meshes that generated errors due to instability during analysis.
The observed errors were (1) maximum contact force error and (2)
maximum residual force error. The maximum contact force error oc-
curred at the interface between the bolt head and the stiffener plate,
as indicated in Fig. 10(b). This type of convergence error was caused
by penetration between two surfaces during contact, meaning that the
bolt head penetrated the stiffener plate. In most cases, the penetration
errors can be addressed by either adjusting the mesh or re-assigning
the contact properties between the elements in contact (bolt-stiffener
plate). Alternatively, the maximum residual force error was caused by
a lack of equilibrium between the external and internal forces. ABAQUS
[1] verifies force equilibrium at every iteration. When force equilibrium
was not achieved within the tolerance, a residual force error was gener-
ated, and the analysis was subjected to abnormal termination. This type
of convergence error can be fixed by issuing an artificial damping factor
to stabilize the analysis.
The damping factors must be increased if the convergence behavior
is problematic, whereas the damping factors should be decreased if they
distort the solution. Post-analysis to compare the energy dissipated by
viscous damping (ALLSD) with the total strain energy (ALLIE) must be
performed to ensure that the dissipated stabilization energy (ALLSD)
is less than the total strain energy (ALLIE), guaranteeing that viscous
force stabilization (VF) is sufficiently low, as shown in Fig. 11. In
ABAQUS [1], viscous damping will stop increasing when a converged so-
lution is obtained.
2.3. Cyclic behavior of the connection with energy dissipation
Fig. 12 shows the deformed mesh and stress distribution of the con-
nection. Bolted stiffeners at the bottom flange of H-beam yielded and
buckled due to the compression while tension force was exerted on
the stiffeners connecting the top flange. Fig. 12 demonstrates the failure
modes of the joints under cyclic loads. The highlighted mesh in red color
illustrates the yielding of the proposed joint. The numerical data were
300 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303

Fig. 13. Stress, strain, and deformation of the joints with damping factor of 0.00005 (including the guide angle, bolts, stiffener plates, flange, and web).
J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303 301

Fig. 14. Stress-strain curves for the selected elements with damping factor of 0.00005.
302 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303
further investigated to comprehend the structural performance of each
element of the proposed joint. The stress, strain, and deformation of the
joints (including the web, flange, L-type stiffener plate, and connecting
bolts) based on the von Mises yield criteria were found to be relatively
small, as shown in Fig. 13. In Fig. 13, all plate stiffeners, bolts, steel
web, and steel flanges (top and bottom) yielded. Large strains as
much as 0.2, which is 126 times the yield strain of the stiffener plate,
were found numerically at stiffener plates for bottom steel flange as
shown in Fig. 13(c) whereas small strains were observed at steel
guide angles shown in Fig. 13(a). The deformation and failure mode of
the bolts connecting bottom steel flanges are illustrated in Fig. 13(b).
The bolt strain and stress were concentrated at the edge of bolt head
as shown in Fig. 13(b). This is because the bottom stiffener plates (see
Fig. 13(c)) buckled and exerted a compression force on the bolt head,
causing stress concentration and large strains at bolt head. The strain
observed at bolt head was found numerically as much as 0.27 which is
170 times the yield strain of the steel material used in this analysis (εy
= 0.00158). Fig. 13(d) depicts the strain and stress found at steel web
and steel flanges of H-beam. Large deformation was observed at bottom
steel flange at the vicinity of the cut between two bottom flanges (see
Fig. 13(d)).
2.4. Stress-strain relationships of the stiffener plates, steel flange, skew web,
and bolts
Fig. 14(a) to (f) illustrates the stress-strain relationships for the se-
lected elements of the proposed steel joint. It was clear that the selected
components (i.e., the stiffener plates, steel flange web, guide plates, and
bolts) yielded and underwent severe strain hardening with large plastic
strains. The yield stresses for the materials employed in this analysis
were described in Table 1. All of the graphs in Fig. 14 were plotted at
the end of the analysis. Fig. 14(f) compares the strains recorded when
a monotonic load and a cyclic load were exerted at the steel joint. It
should be noted that large strains were observed during the cyclic
analysis.
2.5. Application of damping factors to steels with high yield strength
The influence of damping factors on the flexural performance of
steels with high yield strength is demonstrated in Fig. 15, in which con-
vergence was enhanced with damping factors of 0.0002 and 0.0005.
However, the numerical results without a damping factor represented
by Legend 3 displayed premature numerical termination prior to
reaching the maximum strength predicted by the monolithic numerical
analysis. Application of damping factors larger than 0.0002 to the frame
analysis with premature numerical instability provided better conver-
gence while the flexural strength increased substantially as shown in
the moment-displacement relationships represented by Legends 4, 5
Fig. 15. Application of damping factors to steels with high yield strength.
and 6 of Fig. 15. It was also demonstrated that the damping factors
larger than 0.001 improved convergence, but with the nominal moment
strength too high. It was recommended that the results with damping
factors of 0.0002 provided best match with monotonic run up until to
the maximum load limit state. The damping factor can be used for a
wide range of yield strengths when significant convergence difficulties
occur.
3. Conclusions
An easy and fast installation method for wide flange steel beam sec-
tions were introduced by the authors, where the web was cut in a
skewed manner to prevent the bottom flange of the steel beam from
penetrating the L-shaped guide angles (which were pre-installed on
the column bracket) during assembly. The skewed web of the steel
beam sections was placed in the L-shaped guide angles until permanent
connections were made by bolting stiffener plates to obtain a double
shear connection. However, numerical instability can occur due to the
beam-column steel joint (with skew web cuts) shown in the proposed
connection. This study used artificial damping factors to overcome the
numerical instabilities caused by the irregular cut. The finite element
models presented in this study balanced the internal and external
forces, while the artificial damping had a minimal effect on the struc-
tural behavior of the joints; therefore, the viscous forces and the amount
of viscous energy dissipated were small. Damping terms were found to
equilibrate by balancing the internal and external forces at all nodes sta-
bilizing the nonlinear procedure in the model. The potential conver-
gence difficulties were, then, suppressed by using damping factors
which were modified at the beginning of subsequent steps until the
equilibrium of subsequent steps at all nodes was met. The use of
damping factors in the range of 1.0 × 10−5 to 5.0 × 10−5 were explored
via trial and error approaches until a converged solution was obtained;
the dissipated stabilization energy was sufficiently small, ensuring that
the analysis results were not distorted. The stresses and strains obtained
by numerical evaluation also showed that the structural behavior of the
joint elements was predictable and stable during the assembly of pre-
cast frames with heavy loads. The cyclic energy dissipation capability
of the proposed connection also indicated that the connections pos-
sessed sufficient ductility and resistance to heavy construction loads.
Numerical stability was ensured for the frames carrying heavy loads
with rapid assemblage using skewed web connections. The proposed
connection method will be used for the assembly of not only steel
frames, but also steel frames encased in structural concrete with
heavy vertical loads.
Acknowledgement
This work was supported by the Basic Science Research Program
through the National Research Foundation of Korea (NRF) funded by
the Ministry of Education (Grant No. NRF-2016R1D1A1A02937558).
This work was also supported by the Business for Cooperative R&D be-
tween Industry, Academy, and Research Institute, funded by the Korea
Small and Medium Business Administration in 2016 (Grant No.
C0441547).
References
[1] ABAQUS, ABAQUS Analysis User's Manual 6.14-2, Dassault Systèmes Simulia Corp.,
Providence, RI, USA, 2018.
[2] K.M. Ang, G.A. Morris, Analysis of three-dimensional frames with flexible beam–col-
umn connections, Can. J. Civ. Eng. 11 (2) (1984) 245–254.
[3] O.S. Bursi, J.P. Jaspart, Basic issues in the finite element simulation of extended end
plate connections, Comput. Struct. 69 (3) (1998) 361–382.
[4] G. Culache, M.P. Byfield, N.S. Ferguson, A. Tyas, Robustness of beam-to-column end-
plate moment connections with stainless steel bolts subjected to high rates of load-
ing, J. Struct. Eng. 143 (6) (2017), 04017015. .
[5] Dongzhi Guan, Zhengxing Guo, Quandong Xiao, Yongfeng Zheng, Experimental
study of a new beam-to-column connection for precast concrete frames under re-
versal cyclic loading, Adv. Struct. Eng. 19 (3) (2016) 529–545.
 J.D. Nzabonimpa, W.-K. Hong / Journal of Constructional Steel Research 148 (2018) 295–303
[6] S.H. Ju, C.Y. Fan, G.H. Wu, Three-dimensional finite elements of steel bolted connec-
tions, Eng. Struct. 26 (3) (2004) 403–413.
[7] S. Kim, M. Lee, Artificial damping techniques for scalar waves in the frequency do-
main, Comput. Math. Appl. 31 (8) (1996) 1–12.
[8] C.B. Liu, P. Nithiarasu, An artificial-dissipation-based fractional step scheme for
upper-convected Maxwell (UCM) fluid flow past a circular cylinder, Int. J. Numer.
Methods Fluids 57 (9) (2008) 1171–1187.
[9] I.C. Muresan, R. Balc, Finite element analysis of an extended end-plate connection
using the T-stub approach, in: T.E. Simos, C. Tsitouras (Eds.), AIP Conference Pro-
ceedings, 1648, No. 1, AIP Publishing 2015, March, p. 850091.
[10] D.G. Proverbs, G.D. Holt, P.O. Olomolaiye, Factors impacting construction project du-
ration: a comparison between France, Germany and the UK, J. Build. Environ. 34
(1999) 197–204.
303
[11] V. Saberi, M. Gerami, A. Kheyroddin, Comparison of bolted end plate and T-stub con-
nection sensitivity to component thickness, J. Constr. Steel Res. 98 (2014) 134–145.
[12] Y. Sun, S. Zhong, X. Zhang, J. Gill, X. Chen, Artificial Damping Methods for Stable
Computations with Linearized Euler Equations, 22nd AIAA/CEAS Aeroacoustics Con-
ference 2016, May, p. 2971.
[13] Z. Tao, M.K. Hassan, T.Y. Song, L.H. Han, Experimental study on blind bolted connec-
tions to concrete-filled stainless steel columns, J. Constr. Steel Res. 128 (2017)
825–838.
[14] N.S. Trahair, M.A. Bradford, D. Nethercot, L. Gardner, The Behaviour and Design of
Steel Structures to EC3, CRC Press, 2007.