Structures 33 (2021) 3848–3861

Contents lists available at ScienceDirect

Structures
journal homepage: www.elsevier.com/locate/structures

Progressive-models method for evaluating interactive stability of steel box

girders for bridges – Extension of progressive collapse method in
ship structures
Lunhua Bai a, b, Ruili Shen b, Quansheng Yan a, Lu Wang a, b, *, Rusong Miao b, Yinghao Zhao a
a
School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510641, China
b
School of Civil Engineering, Southwest Jiaotong University, Chengdu 610031, China

A R T I C L E I N F O A B S T R A C T

Keywords: The progressive collapse method (PCM) in ship structures, in which local buckling is simplified by average
Ultimate bearing capacity compressive stress-strain curves of stiffened plates, is extended to analyze the ultimate bearing capacity of steel
Progressive-models method box girders for bridges. The PCM for steel bridges is reconstructed as the progressive-models method (PMM).
Steel box girder for bridges
Subsequently, a detailed PMM used for large-scale steel box girders is proposed and implemented through a two-
Local buckling
Shell finite element method
stage analysis, which includes the nonlinear analysis of stiffened plate using shell finite element (FE) model and
Beam-column theory the nonlinear analysis of member using beam-column theory. The PMM is demonstrated by using an example
tested steel box girder of which the ultimate bearing capacity is influenced by local buckling. The traditional
finite element (FE) models of the tested steel box girder, including the refined shell FE model, beam-column FE
model and multi-scale FE model, are established to make a comparison and validate the proposed method. The
results show that the stability of each stiffened plate in the studied steel box girder is different, among them the
mid web, of which the ultimate strength is 0.65 times of the yield strength at the severe defect level, is the most
susceptible to buckling. Compared with results of the test and traditional FE models, the PMM is conservative for
predicting the ultimate bearing capacity. In addition, the comparison of the scale of models shows that the PMM
proposed can significantly reduce the complexity of modeling and improve the computational efficiency, thus the
PMM has potential practical application value. The PMM can be regarded as a new attempt for the ultimate
bearing capacity analysis of large-scale steel bridge members.

1. Introduction design.
In the early years, due to the lack of relevant codes, the local stability
1.1. Background of the SBSM was fully verified by test and theory before their applica­
tions. Taking steel bridges in China as an example, the stability tests and
Stiffened welded box section members (SBSM), depending on many theoretical studies of steel members such as the stiffened plates [1–4],
advantages of stiffened plates, such as convenient welding construction stiffened box section struts [5–8], stiffened steel box girders with
and high strength to weight ratio, are widely used in steel bridges extending flanges [9], wide steel box girder [10], narrow steel box
worldwide. In superstructures, they are used as steel box girders in girder [11,12], steel box arch rib [13], steel bridge tower segments [14],
continuous girder bridges, cable-stayed bridges and suspension bridges, [15] have been carried out. These studies help to bring about the
as steel arch ribs in arch bridges. In vertical supporting structures, they formulation of stability design regulations of the latest Chinese steel
are used as steel piers and steel bridge towers. These members are bridge codes (Ref. [16;17]).
mainly subjected to compression, bending or combined action of However, two points from the above research should be noted: ① in
compression and bending, thus the stiffened plates are locally com­ order to accurately reflect the structural characteristics, the geometric
pressed. The ultimate bearing capacity analysis of members considering similarity ratio for the test model of the SBSM or stiffened plates varies
effects of local buckling has become an important part of bridge safety from 1:2 to 1:5, leading to that the size of test models is over large; ②

* Corresponding author at: School of Civil Engineering and Transportation, South China University of Technology, Guangzhou 510641, China.
E-mail address: [email protected] (L. Wang).

https://doi.org/10.1016/j.istruc.2021.06.061
Received 3 November 2020; Received in revised form 25 February 2021; Accepted 20 June 2021
Available online 8 July 2021
2352-0124/© 2021 Institution of Structural Engineers. Published by Elsevier Ltd. All rights reserved.
L. Bai et al. Structures 33 (2021) 3848–3861

with geometric imperfections and residual stresses of steel plates, the using the fiber beam-column FE model. In the process of modifying the
shell finite element (FE) model can more accurately predict the ultimate uniaxial constitutive model of steel, the original yield stress fy is replaced
bearing capacity of the SBSM and stiffened plates compressed. by the buckling stress σcr of the stiffened plate, which is determined by
If the design of the experimental model can be carried out strictly Japanese code. Thai et al. [32] deduced the ACSSC of plates based on the
according to the similarity ratio, reasonable and reliable results can be energy method, and introduced the curves into the flexible beam-
obtained by tests. However, large-scale test models result in the column FE model. He then used the model to analyze the ultimate ca­
expensive experimental cost, too large test load and difficult experi­ pacity of high-strength steel box section members with no stiffeners, I-
mental operation. The experimental method may not be an optimal beam members and square tube concrete composite section members,
choice in any practical situations. and the proposed model is verified through the ABAQUS shell FE model.
The computational accuracy of the shell FE model is high and It can be concluded from the scattered literatures that the beam-
accepted by engineers. The shell FE model can also consider geometric column models including local buckling are mainly implemented by
imperfections and residual stresses. However, due to the complexity of the effective stress method defined as that local buckling of steel plates is
modeling, low calculation efficiency and difficult convergence, this simplified via the ACSSC. However, to the author’s knowledge, in the
method is difficultly applied to the actual large-scale steel members and field of structural engineering the existing achievements are mainly
bridges. In addition, the ultimate bearing capacity of bridges needs the focused on unstiffened box section members. Only Zhang’s work [31]
global analysis, which should be based on the structural system model involves the SBSM. In his analysis, the inconsistent mechanical behavior
[18,19]. Therefore, the ultimate bearing capacity analysis of large of the steel plate in compression and tension is not considered and de­
bridge steel members by shell FE methods is not fully effective. fects assumed are not clear. Therefore, the beam-column model
Both experimental and shell FE methods have their limitations in including local buckling for the SBSM is not studied systematically, and
analyzing the stability of large-scale bridge steel members, and it is relative practical experiences about using the ACSSC to simplify local
difficult to deduce them to bridge structural systems. Therefore, a few buckling in the SBSM for bridges are also lacked.
literatures adopted the whole bridge full shell FE model or multi-scale To the author’s knowledge, the most mature relative theory using the
FE model to study the ultimate bearing capacity of steel bridges ACSSC to simplify local buckling can refer to the progressive collapse
considering effects of local buckling. For example, Hu et al. [20] and method (PCM) in the field of ship structures [33–37]. The following first
Tang et al. [21] used a bridge full shell FE model to study the ultimate introduces the PCM, then an extended framework of that called the
bearing capacity of steel arch bridges. Ellobody [22] used a bridge shell progressive-models method (PMM) for steel bridges is proposed.
FE model to study interactive-buckling behaviors of railway plate girder Although the PCM in ship structures may not be a very innovative
bridges. Wang et al. [23] used a multi-scale FE model to study the method, the PMM may creatively explore the way of applications of the
interactive buckling ultimate bearing capacity of a self-anchored sus­ PCM in the steel bridge area. There are at least two reasons for studying
pension bridge. However, the influences of geometric imperfections and the PMM. One is that the characteristics of ship hull girders (such as
residual stresses are hardly or only roughly considered in the above scantlings, material) absolutely cannot cover those of large-scale steel
computational models, thus the influence mechanism of local buckling is box members for bridges, the other is that the PCM only calculates the
not completely clarified. Furthermore, one purpose of the existing section but not the member. Thus, the PMM can give new experiences
research is to empirically explain the safety of the bridge through the and insights for these two aspects at least. Under the framework of the
obtained results. In view of this, it is essential to establish an efficient, PMM for the ultimate bearing capacity of steel bridges, the main con­
precise and practical calculation method for the ultimate bearing ca­ tents of the present study will demonstrate the detailed implementation
pacity of large-scale bridge steel members or steel bridge systems, which of the PMM using an example tested steel bridge box girder.
can consider influences of geometric imperfections, residual stresses and
local buckling. This paper aims to do that on the member level. 2. Progressive-models method for steel bridges

1.2. Literature review The static response of a ship hull girder under buoyancy and gravity
loads can be investigated by idealizing the hull girder as a thin-walled
Compared to the experimental and shell FE methods, the beam- box girder. The PCM divides the hull section into many stiffened plate
column method has significant advantages of high computational effi­ elements. An ACSSC is assigned to the corresponding stiffened plate
ciency and convenient modeling. Beam-column models include the element to consider the compressive local buckling effects. All the
beam-column theoretical model and beam-column FE model. However, curves can be made into database, and adopted when calculating the
local buckling cannot be included in traditional beam-column models. ultimate strength of the girder. Then, the ultimate bending moment can
Thus, these beam-column models are reasonable and feasible for steel be obtained through the integration of the section stress. This method is
members without local buckling. These models can be easily found in widely used in the field of ship structures, represented by Smith [33],
nowadays widely used FE software, such as the BEAM188 element Gordon et al. [34], Benson et al. [35,36] and Downes et al. [37], also
model in ANSYS [24], the B31 element in ABAQUS [25]. Some existing known as the Smith’s method. This idea is generally believed to origi­
literatures [26,27] also reported the traditional beam-column models nate from Caldwell’s relevant work in 1965 [38]. Fig. 1 shows the
without considering local buckling. For beam-column models with local analysis flow of the PCM. The symbol M represents bending moment, φ
buckling, works of Chan et al. [28], Lee et al. [29], Paik et al. [30], Zhang represents section curvature, σ represents stress, and ε represents strain.
et al. [31] and Thai et al. [32] are representative. The meaning of “progressive collapse” can be understood as: the
Chan et al. [28] and Lee et al. [29] used the beam-column FE model to failure sequence of the hull girder section can be traced on the
analyze thin-walled box section compressive members, and a three moment–curvature curve, that is, with the increase of the bending
break-line model for the average compressive stress–strain curve moment, different areas of the section fail successively. Fig. 2 shows the
(ACSSC) of the plate to simplify local buckling is adopted in the beam- analysis results of the hull girder by the PCM. Fig. 2 is inspired and
column FE model. Paik et al. [30] used the idealized structural unit reproduced from Ref. [36] for clearer display.
method to deduce the ACSSC of plates in which residual stress is not The progressive-models method is renamed instead of the progres­
considered, and carried out the nonlinear analysis of the thin-walled sive collapse method. It is emphasized that for analyzing different
square box section members with no stiffeners. For the SBSM, Zhang structural hierarchies different computational models should be adop­
et al. [31] considered local buckling effects of stiffened plates by ted, and these models are interconnected. One different point between
modifying the double break-line uniaxial constitutive model of steel, and the PMM and the PCM is that the PMM can be extended to a bridge
analyzed the static and dynamic performance of the steel bridge piers structural system consisting of lots of line elements. The structural

3849
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 1. Flow chat of the progressive collapse method: (a) hull girder section mesh; (b) average compressive stress–strain curve; and (c) M–φ of ship girder section.

Fig. 2. The progressive failure of ship hull girder: (a) M–φ of ship girder section; (b) section state at point 1; (c) section state at point 2; and (d) section state at
point 3.

failure sequence of steel bridges and the applicable scopes of the cor­ the hull girder.
responding model for each structural hierarchy are shown in Fig. 3. For Therefore, the PMM is proposed for the ultimate bearing capacity of
the steel bridge structure with local buckling, the failure may begin with steel bridges. It contains three meanings. The first meaning is that the
local buckling of the basic component of the steel member, namely the refined solid or shell FE model which can precisely consider geometric
plate or the stiffened plate. Then, the local failure of the plate or stiff­ imperfections, residual stresses and local buckling should be more
ened plate may cause the successive failure of other areas of the member suitable for the analysis of the plate or stiffened plate. The second
section, and further leads to the failure of the steel member. When meaning is that for the hierarchy of the member or structure, it is better
several segments are accumulated to fail, the whole bridge collapses. to adopt the beam-column model which is easy to build a structural
The failure of the steel bridge, which follows the failure sequence of model and can be computed highly efficiently. The last meaning is that
plate-stiffened plate-member-system, is similar to the gradual failure of the above two models are interrelated by assuming the ACSSC obtained

3850
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 3. The relationship between failure modes of steel bridge and ultimate bearing capacity analysis models.

from the refined FE model as the uniaxial compressive material consti­

tutive model of steel in the beam-column model. z ∑ n
iπx
vs = vs sin wos,i
This paper only covers hierarchies of the stiffened plate and member, hs i=1 a
but the analysis of the structure system is planned as future work.
πx πy
wc = wc sin sin (3)
a b
3. Progressive-models method for large-scale steel bridge box
girders where, w0pl is the plate imperfection of the subpanel, w0w is the stiffener
web imperfection, wc is the global deflection imperfection, and vs is the
The engineering background of the steel box girder analyzed by the stiffener side imperfection. w0pl , w0w , wc and vs represent the magnitudes
PMM is introduced, then the proposed method and its implementation of these imperfections, respectively. a is the plate length also the spacing
are described. between adjacent diaphragms, b is the subpanel width also the spacing
between adjacent ribs. Table 2 lists the values of these coefficients.
3.1. Objective steel box girder Fig. 5 shows the residual stress distribution of welded plates used in
this study. Residual stresses of welded plates can be expressed as [36]:
⎧
The authors previously conducted a test of steel box girder under
eccentric compression, which is rescaled of a main girder segment of a
σ rcx σ rcy ⎨ − 0.05 slight
= = − 0.15 average (4)
self-anchored suspension bridge (the Egongyan Rail Special Bridge) as a σrtx σrty ⎩ − 0.30 severe
1:4 geometric similar ratio. Fig. 4 shows the general situation of the
tested segment. Table 1 lists properties of steel plates used for fabrica­ where, σ rc is the compressive residual stress, σ rt is the tensile residual
tions of the specimen. The test model has a length of 10 m and 16 box stress and equals to the yield strength fy. The width of the tensile residual
chambers. The spacing between adjacent diaphragms is 625 mm. The stress field can be calculated as:
proportionally increasing eccentric compressive loads are simulated σrc
using 120 prestress strands. The tested girder is finally loaded to 1.81Fd, bt = b (5)
2(σ rc − σrt )
where Fd represents the most unfavorable forces containing of 852 kNm
of the bending moment and 5700 kN of the axial compressive force. The three types of panels are simulated using ANSYS and are meshed
Detailed experimental research can be referred to Ref. [12]. by the 4-node element type SHELL181 which is suitable for moderately-
thick plate or shell structures. Each stiffened plate is simulated as a
(½+½) bay, which is a cantilever model with a half diaphragm spacing
3.2. Stiffened plate analysis – shell finite element model extending to both sides at the diaphragm, as shown in Fig. 6. For di­
aphragms in this girder providing enough stiffness to support the stiff­
The stiffened plates in the steel box girder are divided into three ened plate end, they are simplified as vertical supports in the (½+½) bay
categories owing to their different stability performances, which are the model. Additionally, the (½+½) bay model can consider the continuity
top panel, the web panel and the bottom panel, as shown in Fig. 4. The conditions at the end. The out-plane translations at the diaphragm are
shell FE method is used for the refined simulation of stiffened plates with restrained. Both ends are restrained as symmetrical boundary condi­
geometric imperfections and residual stresses. It is noted that there are tions. The mesh size is taken to about b/10 through a convergent study.
various methods that can be used to analyze stiffened plates, like the The material constitutive model of steel is considered as the elastic
orthotropic plate theory [35] and the idealized unit element method perfect plastic model with two broken lines. The Young’s modulus is
[30]. assumed as 206,000 MPa, and the Poisson’s ratio is assumed as 0.3. The
Geometric imperfections and residual stresses are considered the yield stress is assumed as listed in Table 1. The geometric imperfections
same as those of the PCM, thus the Smith’s defects are adopted. Geo­ are implied through modifying the coordinates of nodes by the com­
metric imperfections are expressed as [36]: mand “NMODIF”, while residual stresses are applied as initial stresses by
(
πx mπx (m + 1)πx
)
πy the command “INISTATE”.
w0pl = w0pl B1 sin + Bm sin + Bm+1 sin sin (1) The nonlinear FE model is solved twice with the New-Raphson
a a a b
method. In the first solution, all degrees of freedom of all nodes are

3851
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 4. The experiment of the steel box segment analyzed: (a) illustration of the experiment; (b) the experiment situation; and (c) the girder section.

restricted to solve the reaction forces of each node. The second solution 3.3. Steel box member analysis – a beam-column model based on
is divided into two load steps. In the first load step, the assumed numerical integration method
boundary conditions of the (½+½) bay model are introduced, and the
node reaction forces obtained from the first solution are loaded to get the The objective is to take the ACSSC obtained from stiffened plate
initial state of the model. Subsequently, in the second load step the FE (½+½) bay FE models as the uniaxial compressive material constitutive
model under longitudinal compression is solved using displacement relationship of steel in the member model. There are various methods
loading. The solved reaction forces of the first solution restrain out-plane capable of analyzing the member, including the force-based beam-col­
displacements initiated by combined actions of geometric imperfections umn FE model [32], displacement-based beam-column FE model [39],
and residual stresses. Since residual stresses adopted in FE models are central difference method [26] and numerical integration method [27].
deduced from the idealized flat plate, geometric imperfections may In this paper, the (½+½) bay FE model results are embedded into the
make plate deformed again. Thus, the solved reaction forces can effec­ numerical integration method [27].
tively reduce secondary deformation caused by the combined action of Fig. 7 shows the calculation model of the member. The numerical
geometric imperfections and residual stresses. integration method for the beam-column model can be clearly found in

3852
L. Bai et al. Structures 33 (2021) 3848–3861

Table 1
L
Steel plates in the steel box girder [12]. le = (6)
n
Classification of Thickness Materials Yield Yield
plates (mm) strength fy strain εy where, L is the member length. The deflection wj and rotation θj at node j
(MPa) (με)
are expressed by the Taylor series:
Top plates (TP) 10 Q345 of which the 349 1694
Bottom plates 8 nominal yield 392 1903 l2e
wj = wj− 1 + le θj− 1 − φ (7)
(BP) stress is 345 MPa 2 j− 0.5

Web plate (WP) 5 447 2170

Diaphragms 3 θj = θj− 1 − le φj− 0.5 (8)
I ribs of TP 8
I ribs of BP 6 378 1835
Web of Tee ribs 3
where, φj− 0.5 is the curvature of the middle node between the j and j-1
Flange of Tee 3 node.
ribs In the calculations, the strain and curvature data of each node is
stored to facilitate the calculation of the axial deformation of each
element:
Table 2 ΔP,j,k = εj− (9)
0.5,k le
Geometric imperfections magnitudes of steel plate in ship structures [36].
( √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ )
Defect m n w0pl Bi wc vs w0w
grades ΔM,j,k = 2 le − l2e − (wj,k − wj− 1,k )2 (10)
2
slight a/ a/ 0.05β t B1 = 0.0002a 0.0004a 0.0002a
b hs 0.8 where, ΔP,j,k is the axial shortening of element j caused by the axial strain
+1 +1 Bm =
generated by load Pk; ΔM,j,k is the axial shortening of element j caused by
0.2
Bm+1 the deflection generated by load Pk; εj− 0.5,k is the axial strain of the
= 0.01 middle node between the j and j-1 node caused by load Pk; and Pk rep­
average a/ a/ 0.1β2t B1 = λs < 0.2: 0.002a 0.001a resents the k level load. The axial shortening of the member under load
b hs 0.8 0.0008a
Pk is obtained by superposition of the above two parts:
+1 +1 Bm = λs < 0.6:
0.2 0.0012a
∑( )
Δk = ΔP,j,k + ΔM,j,k (11)
Bm+1 λs > 0.6:
n
= 0.01 0.0015a
severe a/ a/ 0.3β2t B1 = 0.006a 0.005a 0.0025a Two iterations are needed during the solving process: ① the iteration
b hs 0.8 analysis of the section, which involves the calculation of M-P-φ (bending
+1 +1 Bm =
0.2
moment-axial force-curvature), is to determine the section resistance
Bm+1 according to the given initial curvature and axial strain; ② the iteration
= 0.01 analysis of the member, is conducted from A end to B end (Fig. 7), and
the compatible conditions of B end will be checked. Compared to the
origin numerical integration method in Ref. [27], it is noteworthy that
Ref. [27].
some different details are made to include local buckling effects and
If the component is divided into n equal length elements, the element
improve computational efficiency, as shown in Fig. 8. In the figure, Et is
length is:

Fig. 5. Residual stress distribution of welded plates.

3853
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 6. Shell FE models of stiffened plates: (a) FE model of the top plate; (b) FE model of the web; and (c) FE model of the bottom plate.

Fig. 7. Illustrations of element division in member analysis.

tangent modulus, yi is the coordinate of i fiber corresponding to the same with section 3.2. The mesh size is about 50 mm. A MASS21
section centroid, Ai is the area of i fiber, and Ii is the moment of inertia of element with only one six-freedom-degree node is created at each end
i fiber. The steel box section is meshed as a fiber layer section, shown as section centroid of the test girder. The MASS21 element node is taken as
Fig. 9. The ACSSC is input as the material model of steel instead of the a master node controlling other associated nodes at the end of the shell
original stress-strain curves to consider local buckling of the stiffened FE model, so the ends are simulated as a rigid region by the command
plate. For different fiber layers, the corresponding ACSSC are derived “CERIG”. The constraints are both placed at the MASS21 element nodes,
from different stiffened plate (½+½) bay shell FE models. Thus, the and simple boundary conditions are assumed, as shown in Fig. 11(a).
beam-column model with considering local buckling has a hybrid sec­ Since geometric imperfections can be simulated along the full length of
tion due to the different ACSSC of the fiber layers. the girder, the shell FE model can consider local buckling in the girder
The above calculation method can be programmed by combing completely. The material constitutive model of steel is considered as the
ANSYS and MATLAB: ① the element MESH200 in ANSYS is used to elastic perfect plastic model with two broken lines.
model the section, the section discrete program SEC_GENERATE, which
can output the information of each fiber in the section (yi, Ai, Ii), is 4.2. Beam-column finite element model
compiled and can be directly evoked by the MATLAB main program; ②
the numerical integration method is compiled in MATLAB. Fig. 10 shows The beam-column FE model of the steel girder can be simulated by
the block diagram of the program. the BEAM188 element in ANSYS. The BEAM188 is based on the Timo­
shenko beam theory, which can consider the influence of shear defor­
4. Finite element models for validation mation, and can consider the section shape. The existing research shows
that the BEAM188 element can be used to simulate the compression and
4.1. Shell finite element model for steel box segment bending stability analysis of thin-walled members without considering
local buckling [40,41]. The end boundary conditions are as those of the
The element SHELL181 in ANSYS is used to simulate the tested previous shell FE model, as shown in Fig. 11(b). The material constitu­
girder. Geometric imperfections and residual stresses are adopted as the tive model of steel is considered as the elastic perfect plastic model with

3854
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 8. Comparison of section analysis methods: (a) section analysis of the traditional method [27]; and (b) section analysis of the proposed method.

Fig. 9. Section fiber layers of the specimen.

two broken lines. considered as the elastic perfect plastic model with two broken lines.
The connection between the beam element and shell element is realized
through constraint equations by using the command “CERIG”. The
4.3. Multi-scale finite element model function of the CERIG command is used to generate a rigid domain, there
is one node of the beam element that serves as the master node, and the
The element SHELL181 is used to simulate four box chambers in the other nodes of the shell element act as the slave nodes. This simplifi­
middle area of the tested girder, while BEAM188 is used to simulate the cation follows the plane section assumption, which should satisfy the
rest area. The multi-scale FE model is shown in Fig. 11(c). The element displacement compatibility conditions:
size is consistent with the refined shell FE model. Since geometric im­
perfections can be simulated in the shell element area, the multi-scale FE
model can only consider local buckling of the shell element part of the
girder. Fig. 11(d) and (e) show the modeling of geometric imperfections
and residual stresses. The material constitutive model of steel is

3855
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 10. Flow chart of beam-column theory program.

⎧
⎪
⎪ ui,z = u0,z + yi ⋅θ0,x + xi ⋅θ0,y yield strength, weighted by the steel plate cross-sectional area). The
⎪
⎪
⎪
⎪ ui,y = u0,y + xi ⋅θ0,z statistical results are listed in Table 3. With the increase of the defect
⎨
ui,x = u0,x + yi ⋅θ0,z
(12) level, the ultimate strength of the web decreases most seriously, which is
⎪
⎪
⎪
θi,z = θ0,z 0.65fy under the severe defect, and the ultimate strength of the top plate
⎪
⎪
⎪
⎩
θi,y = θ0,y and bottom plate is 0.85fy and 0.72fy, respectively. Thus, it can be
θi,x = θ0,x concluded that the stiffened plates are sensitive to defects to some
extent.
where, ui,x, ui,y and ui,z represent the translation displacement of the shell
element slave node i, u0,x, u0,y and u0,z represent the translation
displacement of the beam element master node which located at the 5.2. Verification of progressive-models method
center of the section, θi,x, θi,y and θi,z represent the rotations of the shell
element slave node i, and θ0,x, θ0,y and θ0,z represent the rotations of the The corresponding curves in Fig. 12 are assumed as the uniaxial
beam element master node. compressive constitutive models of the 4 stiffened plate regions (The top
panel, two webs and bottom panel). Fig. 13 and Fig. 14 show the load
5. Results and discussion displacement curves calculated by the PMM and other three FE models
under three defect levels, also these curves are compared to the test.
5.1. Results of stiffened plates As seen from Fig. 13 and Fig. 14, in terms of the variation law of the
bending stiffness, axial stiffness and ultimate bearing capacity, results of
Fig. 12 shows ACSSC results calculated by the shell FE models of the the PMM are relatively consistent with the test results, shell FE model
stiffened plates. It can be seen that the ultimate strength of each panel and multi-scale FE model. In additions, with the light defect level, results
decreases with the increase of the defect level. The ultimate strength is from the four calculate models are pretty close. But with the average and
compared with its yield strength (where the yield strength is the nominal severe defect level, results of the PMM, shell FE model and multi-scale
FE model are lower than those of the BEAM188 FE model. Thus, it

3856
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 11. FE models of the steel box girder: (a) shell FE model; (b) beam-column FE model; (c) multi-scale FE model; (d) residual stresses; and (e) geometric
imperfection.

Fig. 12. The compressive stress-strain curves of stiffened plates.

statistical results of the difference are shown in Table 4. The ultimate

Table 3
bearing capacity in the table is expressed by multiple of Fd. It can be seen
Ultimate strength of each stiffened plate.
from the table that the results of the PMM demonstrate that the girder is
Panels Ultimate strength σu(MPa) Nominal yield strength fy(MPa) are more sensitive to the defect level. Under different defect levels, the
slight average severe variation range of ultimate bearing capacity is 1.77–2.11Fd. The PMM
TP 347.38 332.03 298.07 347.87 can figure out more conservative results than the shell or multi-scale FE
WP 377.18 336.52 279.65 426.54 model. The maximum load value of the test is 1.81Fd, so it can be
BP 375.39 329.44 280.90 388.00 inferred that the possible defect level of the test girder is between the
average and severe.
indicates that local buckling effects can lead to the reduction of the
stiffness and ultimate bearing capacity of the girder as the defects worse. 5.3. Failure modes of steel box girder
When considering local buckling, the stiffness degradation of the
member becomes more significant with the aggravation of the defect The failure modes of the test girder calculated by the shell FE model
level, the comparison between the multi-scale FE model and shell FE are clearly shown in Fig. 15. The failure modes of the weak web is also
model shows that when the defect level is severe the degradation of the illustrated through the average level defect situation. The web subpanel
axial stiffness and bending stiffness predicted by the multi-scale FE has locally buckled into four half-waves along the longitudinal direction
model is slower than that of the shell FE model, and the ultimate bearing in the box chamber 7 but has globally buckled in the box chamber 8
capacity of the multi-scale FE model is slightly higher, mainly because influenced by excessive deformation of the T-rib. While the T-rib has
geometric imperfections and residual stresses are only considered in the failed in symmetrical and anti-symmetrical instable modes. These fail­
local area of the girder. ure modes indicate that the out-plane buckling of the web and the
The defects make different influences on different models. The flexural and torsion instability of the web T-rib make the girder reach its

3857
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 13. Comparison of load shortening curves of the girder.

Fig. 14. Comparison of load mid-span deflection curves of the girder.

modeling complexity, computational efficiency and whether local

Table 4
buckling included or not. The complexity of modeling is determined by
Ultimate bearing capacity predicted by different numerical models.
the number of area regions in the model, while the computational effi­
Defect PMM Multi-scale FE Shell FE model Fp/ Fp/ ciency is determined by the type and number of elements.
level Fp model Fm Fs Fm Fs
Results show that the number of elements in the refined shell FE
slight 2.11 2.14 2.18 0.99 0.97 model of the objective girder is far more than that of the other three
average 1.99 2.14 2.12 0.93 0.94
models. The calculation cost of the PMM proposed actually is spent on
severe 1.77 2.14 2.08 0.83 0.85
three stiffened plate shell FE models. It can be predicted that the
computational efficiency of the PMM is faster than that of the multi-scale
ultimate limit state. In order to explore the relationship between the FE model. In additions, the multi-scale FE model can partially reflect
local buckling and the ultimate bearing capacity of the girder, the out- local buckling of the shell FE model, which may lead to unsafe conse­
plane displacement direct stress curve and the out-plane displacement quences when the defects are serious. Effects of local buckling along the
load curve of a failure element located in the chamber 8 (the position is full span can be considered by the PMM method and refined shell FE
marked in Fig. 15) are extracted, as shown in Fig. 16. According to the model. As for the steel box girder analyzed, the PMM can reduce the
severe defect, the failure of this element occurred earlier than the calculation dimension of the member through the two-stage analysis,
moment the global girder reaching its ultimate bearing capacity, and the which reasonably simplifies the analysis of the member considering
load could continue to increase by 0.3Fd after local failure. Thus, the test local buckling.
girder has a post-buckling capacity. It can be concluded that the test
girder mainly fails in the flexural failure mode, while there is obvious 6. Conclusions
local buckling of plates and stiffeners before its failure. Therefore, it is
essential to consider local buckling for the girder. (1) The progressive-models method, shell FE model and multi-scale
FE model are used to restudy the experimental steel box girder
5.4. Comparison of model scale of the Egongyan Rail Special Bridge. This study shows an attempt
to introduce the defect classifications and the progressive
Table 5 summarizes the comparison results of the 4 model scales, and collapse method in ship structures into the evaluation of the ul­
makes a comprehensive analysis on three aspects, including the timate bearing capacity of a steel box girder for bridges. The

3858
L. Bai et al. Structures 33 (2021) 3848–3861

Fig. 15. Failure modes of the girder: distributions of out-plane displacements (unit: mm).

Fig. 16. The relationship between the direct stress and out-plane displacement/the load of the girder: (a) out-plane displacement stress curve; and (b) external load
stress curve.

3859
L. Bai et al. Structures 33 (2021) 3848–3861

Table 5 [3] Di J, Zhou XH, Lv ZD, et al. Experiment on ultimate bearing capacity of u-stiffened
Comparison among the numerical models. plate of orthotropic steel box girder. China Journal of Highway and Transport
2009;22(2):59–64 (in Chinese).
Models Number of area Model scale Local [4] Wu C, Zeng MG, Feng LY. Local part model test for ultimate bearing capacity of
regions buckling orthotropic panel of Sutong Bridge. Bridge. Construction 2006;2):21–23+35 (in
Chinese).
Shell FE model 3548 68,730 shell Yes [5] Yin YP, Fang QH. The stability of compression bar with box section stiffened by ribs
elements on Tianxingzhou Yangtse Bridge. Journal Huazhong University of science and
BEAM188 FE 0 20 beam-column No Technology (Natural Science Edition) 2008;36(4):119–22 (in Chinese).
model elements [6] Su QT, Zhang QL. Mixed finite element method for mechanics analysis of
Multi-scale FE 860 16,327 shell Locally longitudinal stiffened steel box member. Engineering Mechanics 2006;23(10):
model elements included 163–7 (in Chinese).
and 15 beam [7] Liu YJ, Zhang JG, Zhang GX, et al. Experimental research on stability bearing
elements capacity of spliced segments of steel box column. Journal of Building Structures
PMM 86 for TP 20 beam-column Yes 2010;s1):23–27 (in Chinese).
[8] Peng XM. Test of Stability of compressive box member welded with thick plates of
76 for BP elements
Wuhu Yangtze River Bridge. China Railway Science 2001;22(5):101–5 (in
106 for WP and 4613 shell
Chinese).
elements
[9] Gui C Z, Lei J Q, Huang Z W, et al. Experimental investigation on buckling behavior
of stiffened steel box-section beam-column with external flanges. Journal of
Zhejiang University (Engineering Science), 2018, 52(02):333-340+351. (in
results of the example verify the progressive-models method, and Chinese).
prove again that it is feasible to simplify local buckling effects by [10] Li LF, Shao XD, Yi WJ, et al. Model test on local stability of flat steel box girder.
China Journal of Highway and Transport 2007;20(3):60–5 (in Chinese).
average compressive stress-strain curves in the analysis of steel [11] Bai L, Shen R, Wang Lu, Wang Y, Miao R. Experimental and numerical Study on
box girders. Buckling Behavior of a Rigidly Stiffened Plate with Tee Ribs. International Journal
(2) Compared with the multi-scale FE model which connects the of Steel Structures 2018;18(2):582–95.
[12] Shen RL, Bai LH, Zhang SH. Ultimate capacity of narrow type steel box section for
refined shell FE model and beam-column FE model by the railway self-anchored suspension bridge under bias compression. Advanced Steel
displacement compatibility condition, the connection between Construction 2019;15(2):173–84.
the refined shell FE model and beam-column model in the [13] Chen K, Wu Q, Nakamura S, Chen B. Experimental and numerical study on
compressive behavior of convex steel box section for arch rib. Eng Struct 2016;114:
progressive-models method is implemented by the average
35–47.
compressive stress-strain curves. For the test girder with the same [14] Wang CS, Wang X, Yu X, et al. Experiment on local buckling of steel bridge tower
or similar stress along the full span, the progressive-models segment. China Journal of Highway and Transport 2009;22(6):74–81 (in Chinese).
method predicts the lower stiffness and ultimate bearing capac­ [15] Chen KM, Wu QX, Huang HH. Mechanical property of segmental tower of cable-
stayed bridge. Journal of Traffic and Transportation Engineering 2016;16(5):
ity, and amplifies the effects of defects. 17–29 (in Chinese).
(3) New findings of the test steel girder include: the fact that the web [16] JTGD64-2015. Specifications for Design of Highway Steel Bridge. 2015. (in
is the most vulnerable to buckling is clarified, and the influence of Chinese).
[17] TB 10091-2017. Code for Design on Steel Structure of Railway Bridge. 2017. (in
defects on the web is the most significant; The ultimate strength Chinese).
under the severe defect of the web is 0.65fy; the minimum ulti­ [18] Choi D-H, Gwon S-G, Yoo H, Na H-S. Nonlinear static analysis of continuous multi-
mate capacity of the girder under different defect levels is 1.77Fd, span suspension bridges. International Journal of Steel Structures 2013;13(1):
103–15.
indicating that the steel box girder has sufficient safety reserve; [19] Thai HT, Choi DH. Advanced analysis of multi-span suspension bridges. J Constr
and according to the results of the PMM, the defect level of the Steel Res 2013;90(11):29–41.
test girder may be between the average and severe level. [20] Hu X, Xie Xu, Tang Z, Shen Y, Wu Pu, Song L. Case study on stability performance
of asymmetric steel arch bridge with inclined arch ribs. Steel & Composite
(4) The progressive-models method is verified through the test and Structures 2015;18(1):273–88.
analysis of the experimental steel box member . Under the macro [21] Tang Z Z. Research on seismic response evaluation method of steel bridges and
framework of the progressive-models method for steel bridges, seismic damage index of steel material. PhD thesis. Zhejiang University, 2016. (in
Chinese).
the next future work is the establishment and applications of the
[22] Ellobody E. Interaction of buckling modes in railway plate girder steel bridges.
progressive-models method for evaluating the stability of the Thin-Walled Structures 2017;115(6):58–75.
actual steel box girder segment and steel bridge system. [23] Wang CJ, Dai JG, Zang Y, et al. Static stability analysis of self-anchored steel box
girder suspension bridge. Bridge Construction 2019;49(2):47–51 (in Chinese).
[24] ANSYS Inc. ANSYS element reference. Electronic release 8.0. USA: SASIP, Inc.;
1998.
Declaration of Competing Interest [25] ABAQUS Inc. ABAQUS standard user’s manual. Version6.14–2. 2014.
[26] Chen Y. Ultimate strength analysis of stiffened panels using a beam-column
The authors declare that they have no known competing financial method. PhD thesis. Virginia Polytechnic Institute and State University; 2003.
[27] Shen Z-Y, Lu L-W. Analysis of initially crooked, end restrained steel columns.
interests or personal relationships that could have appeared to influence J Constr Steel Res 1983;3(1):10–8.
the work reported in this paper. [28] Chan SL, Kitipornchai S, Al-Bermani FGA. Elastoplastic analysis of box beam
columns including local buckling effects. J Struct Eng 1991;117(7):1946–62.
[29] Lee S-L, Shanmugam NE, Chiew S-P. Thin-walled box columns under arbitrary end
Acknowledgements loads. J Struct Eng 1988;114(6):1390–402.
[30] Paik JK, Kim BJ. Progressive collapse analysis of thin-walled box columns. Thin-
Walled Structures 2008;46(5):541–50.
This paper was financially supported by the China Postdoctoral
[31] Zhang J, Wang CL, Ge H. A simplified method for seismic performance evaluation
Science Foundation (No. 2021M691071), the National Natural Science of steel bridge piers with thin-walled stiffened box sections. Advanced Steel
Foundation of China (No. 51478193, No. 51608207), the Natural Sci­ Construction 2014;10(4):372–84.
[32] Thai HT, Uy B, Khan M. A modified stress-strain model accounting for the local
ence Foundation of Guangdong Province, China (No.
buckling of thin-walled stub columns under axial compression. J Constr Steel Res
2019A1515011941), the Guangdong Basic and Applied Basic Research 2015;111(8):57–69.
Foundation (No. 2020A1515110240) and the China Scholarship Council [33] Smith CS. Influence of local compressive failure on ultimate longitudinal strength
(No. 201806155102, No. 201906155028). of a ship’s hull. In practical design of ships and other floating structures: Tokyo;
1977. p. 73–9.
[34] Gordo JM, Soares CG. Approximate method to evaluate the hull girder collapse
References strength. Mar struct 1996;9(3):449–70.
[35] Benson S, Downes J, Dow RS. Compartment level progressive collapse analysis of
lightweight ship structures. Mar struct 2013;31(2):44–62.
[1] Li LF, Shao XD. Theoretical study and model test on the capacity of orthotropic U-
[36] Benson S. Progressive collapse assessment of lightweight ship structures. PhD
rib stiffened plates. China Civil Engineering Journal 2007;40(6):42–8 (in Chinese).
thesis. University of Newcastle Upon Tyne; 2011.
[2] Zhao Q, Zhai ZS, Chen BC, et al. Experimental study on local stability of u-rib
stiffened plate. Journal of Building Structures 2017;38(7):156–63 (in Chinese).

3860
L. Bai et al. Structures 33 (2021) 3848–3861

[37] Downes J, Tayyar GT, Kvan I, Choung J. A new procedure for load-shortening and [40] Wang Y-B, Li G-Q, Chen S-W, Sun F-F. Experimental and numerical study on the
-elongation data for progressive collapse method. Int J Nav Archit Ocean Eng 2017; behavior of axially compressed high strength steel box-columns. Eng Struct 2014;
9(6):705–19. 58:79–91.
[38] Caldwell JB. Ultimate longitudinal strength. Trans. Royal Inst. Naval Architects [41] Wang Y-B, Li G-Q, Chen S-W, Sun F-F. Experimental and numerical study on the
1965;107:411–30. behavior of axially compressed high strength steel columns with H-section. Eng
[39] Melo P, Vieira RF, Virtuoso FB. A beam-column model for evaluating the effects of Struct 2012;43:149–59.
residual stresses and geometrical imperfections. Comput Struct 2019;223:106090.
https://doi.org/10.1016/j.compstruc.2019.06.007.

3861