Thin-Walled Structures 185 (2023) 110618

Contents lists available at ScienceDirect

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

Full length article

Seismic risk assessment of reticulated shell structures considering multiple

uncertainties
Wen Hua, Jihong Ye ∗
Jiangsu Key Laboratory Environmental Impact and Structural Safety in Engineering, China University of Mining and Technology, Xuzhou 221116, China
Xuzhou Key Laboratory for Fire Safety of Engineering Structures, China University of Mining and Technology, Xuzhou, 221116, China

ARTICLE INFO ABSTRACT

Keywords: Seismic risk assessment can quantitatively predict the damage caused by earthquakes. This method has been
Reticulated shell structures widely used in many structures, but it is seldom used in reticulated shell structures, and the uncertainty factors
Two-parameter damage criterion are not fully considered. First, a two-parameter damage criterion suitable for reticulated shell structures is
Multiple uncertainties
proposed based on the Park-Ang damage model. Second, multiple uncertainty analysis is carried out. Based
Seismic risk assessment
on the traditional method of considering only the uncertainties of record-to-record and structural design
parameters, the uncertainties of performance indicators are obtained by the Latin hypercube sampling (LHS)
method according to the probability distribution of the parameters in the two-parameter damage criterion.
The modelling uncertainty is obtained by the comparison results of the tests and numerical simulations of
the reticulated shells. For the first time, the evaluation concept of the construction uncertainty of reticulated
shells is proposed by considering node errors. Finally, seismic risk assessments of two large-scale shaking
table reticulated shell test models are carried out, and the influence of each uncertainty on the exceedance
probability of the reticulated shells in 50 years is analysed. The results show that the two-parameter damage
criterion can effectively evaluate the damage to reticulated shell structures and that the structural damage
index values D at the four performance points are 0.3, 0.6, 1.0, and ∞. The modelling uncertainty has the
greatest impact on the 50-year exceedance probability of the reticulated shells, and the exceedance probabilities
in 50 years are 3 to 4.9 times the probabilities determined when only considering the uncertainties of record-
to-record, structural design parameters, and performance indicators. The 50-year exceedance probabilities
considering these five types of uncertainties are 3 to 9 times the probabilities determined when only considering
the uncertainties of record-to-record and structural design parameters, indicating that the traditional method
can underestimate the potential risks of reticulated shell structures. The two-parameter damage criterion and
the multiple uncertainty analysis method proposed in this paper can be used for the seismic risk assessment
system of reticulated shell structures.

1. Introduction 12]. However, unlike the traditional structural form of multistory build-
ings, the reticulated shell structure is a complex structure composed
Reticulated shell structures are widely used in urban landmark of numerous members in a certain topological relationship, and the
buildings; however, as emergency shelters in the event of an earth- special characteristics and complexity of the force mechanism make it
quake, damage or collapse of a reticulated shell structure will result less studied in the field of seismic risk assessment [13–18]. In addition,
in serious casualties and economic losses. Seismic risk assessment is most risk assessment studies only consider the uncertainties of record-
based on performance-based seismic design, and it is a quantitative to-record and structural design parameters, ignoring the effects brought
assessment of the possibility of various damage scenarios caused by about by performance indicators, modelling, and construction, which
earthquakes at the proposed site. It can provide the data basis for may underestimate the potential risks of reticulated shell structures,
the prediction of human casualties and economic losses and has been with serious consequences. Therefore, it is necessary to establish a
widely used in many structures, such as steel frame structures [1– comprehensive scientific seismic risk assessment system for reticulated
5], masonry structures [3,4], reinforced concrete structures [3,4,6–9], shell structures and to fully consider the uncertainty factors throughout
light-frame wood structures [10], and cold-formed steel structures [11, the seismic risk assessment.

∗ Corresponding author at: Jiangsu Key Laboratory Environmental Impact and Structural Safety in Engineering, China University of Mining and
Technology, Xuzhou 221116, China.
E-mail address: [email protected] (J. Ye).

https://doi.org/10.1016/j.tws.2023.110618
Received 1 December 2022; Received in revised form 19 January 2023; Accepted 6 February 2023
Available online 24 February 2023
0263-8231/© 2023 Elsevier Ltd. All rights reserved.
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Seismic risk assessment requires appropriate criteria to measure and snow load are sampled by the Latin hypercube sampling (LHS)
structural damage. In 1985, Park and Ang [19] proposed a two- method to calculate the uncertainties of structural design parameters.
parameter damage model suitable for reinforced concrete members Based on the probability distribution of the parameters, an evalua-
under earthquakes based on the test results of a large number of rein- tion method for the uncertainties of performance indicators of the
forced concrete column members. The model used a linear combination two-parameter damage criterion is established. The literature related
of normalized maximum displacement and normalized hysteretic en- to the comparison results of the tests and numerical simulations for
ergy. Since then, many scholars have modified this model and applied reticulated shell structures is referenced to calculate the modelling
it to other components or structures [20–23]. In terms of reticulated uncertainty. The concept of construction uncertainty is first proposed
shell structures, Du [24] proposed the damage criterion for single-layer in seismic risk assessment of reticulated shell structures by considering
spherical reticulated shell structures based on the Park-Ang damage the node errors, and the critical stability bearing capacity is used for
model and reflected the contribution of displacement and energy reflection. Finally, the two large-scale shaking table reticulated shell
dissipation to damage at a ratio of 1:1. However, the criterion has the test models of Ye [29] are taken as examples to draw the vulnerability
problem that the physical meaning of the parameters is not clear, and curves. The ‘‘square-root-sum-of-squares (SRSS)’’ method is selected to
there is a lack of experimental verification. conduct seismic risk assessment considering multiple uncertainties and
In addition, reasonably considering the uncertainties in the struc- analyse the influence of each uncertainty on the exceedance probability
tural system is also the key to seismic risk assessment. The uncertainties in 50 years. The two-parameter damage criterion and the analysis
in the process of structural seismic risk assessment can be classified method of multiple uncertainties proposed in this paper can be used
into two types [25]: aleatoric uncertainty and epistemic uncertainty. for the performance-based seismic design and seismic risk assessment
The former is an inherent attribute that cannot be eliminated, such as of reticulated shell structures.
the uncertainties of record-to-record, structural design parameters, and
performance indicators. The latter is caused by a lack of knowledge 2. Two-parameter damage criterion of reticulated shell structures
and can be eliminated, such as modelling uncertainty. In the early the-
ory of seismic risk assessment, record-to-record and structural design 2.1. Two-parameter damage model of reticulated shell structures
parameters were often taken as uncertain factors. With the continu-
ous development of research, modelling quality has begun to be an In 1985, Park and Ang proposed a damage model suitable for
uncertain factor in seismic risk assessment [5,7,10–12,26]. Terzic [26] reinforced concrete members under earthquake loading based on the
conducted a blind prediction contest of a full-scale reinforced-concrete test results of a batch of reinforced concrete column members [19]:
bridge column exposed to six consecutive unidirectional ground mo- 𝑑𝑀 𝑏
𝐷𝑃 𝑎𝑟𝑘-𝐴𝑛𝑔 = + 𝑑𝑒𝑝 (1)
tions of different intensities; this study provided important quantitative 𝑑𝑈 𝑞𝑦 𝑑𝑈 ∫
data on reinforced-concrete bridge column modelling uncertainty in where 𝐷𝑃 𝑎𝑟𝑘-𝐴𝑛𝑔 represents the damage index; 𝑑𝑈 represents the ulti-
seismic risk assessment. In addition, Jiang [11] used the maximum mate displacement of the member under monotonic loading; 𝑞𝑦 repre-
interstory drift angle of many test results for cold-formed steel walls sents the yield shear; 𝑑𝑀 and ∫ 𝑑𝑒𝑝 represent the maximum deformation
as an index and utilized the logarithmic standard deviation of different and the accumulative hysteretic energy of the member under earth-
index values under the same damage state as the uncertainty of the quake, respectively; and b represents the energy dissipation factor
performance indicators. related to the shear span ratio, axial compression ratio and stirrup ratio
However, most seismic risk assessments for reticulated shell struc- of the member.
tures only consider the uncertainties of record-to-record and struc- The denominator representing energy dissipation in this model is
tural design parameters, and the effect of uncertainties in modelling the product of yield shear 𝑞𝑦 and ultimate displacement under mono-
and performance indicators is still unclear. Furthermore, the complex tonic loading 𝑑𝑈 . It does not represent hysteretic energy dissipation,
structures, node positioning, and hoisting installation during construc- and its physical meaning is unclear. In addition, the model assumes
tion will cause node errors and initial imperfections. The reticulated that the contribution of maximum deformation and the accumulative
shell structures are sensitive to initial imperfections, and the D value hysteretic energy to damage is linear. However, through the author’s
(structural damage index) will be affected by construction errors. The analysis of a large number of examples, it was found that the contribu-
uncertainty of the analysis results caused by construction errors can be tion of the displacement and energy dissipation of the reticulated shell
defined as construction uncertainty; it belongs to epistemic uncertainty, structure to the damage is not linear. The 𝐾8,6 single-layer spherical
and no scholar has considered it in the seismic risk assessment system reticulated shell structure with a span of 40 m is used as an example
of reticulated shells. (as shown in Fig. 1), and the response of the displacement and energy
Hence, in this paper, a two-parameter damage model suitable for consumption of the reticulated shell structure with the development
reticulated shells is proposed based on the Park-Ang damage model. of ground motion is observed. The rise-span ratio of the structure is
According to the classification habits of Chinese GB 50011-2010 [27], 1/3, the periphery of the structure is supported by three-way fixed
the performance state of the reticulated shells is divided into five states: hinges, the circumferential members are 95 mm × 3.5 mm, and the
intact, slightly damaged, moderately damaged, severely damaged, and oblique and radial members are 102 mm × 3.5 mm. The structural
collapsed. The division basis is proposed according to the dynamic char- material is Q235 steel, the elastic modulus is 2.06 × 105 MPa, and the
acteristics of the reticulated shells and previous research results [13, ideal elastic–plastic model is used. Rayleigh damping is calculated by
24]. The two-parameter damage criterion and classification indexes using the natural period of the first and second modes (0.28 s), and the
of the reticulated shell structures are fitted by the results of 216 sets damping ratio is 0.02. The load on the structure is 60 kg/m2 .
of comprehensive calculation examples with different spans, rise-span The structure is analysed with ANSYS finite element analysis soft-
ratios, roof loads, member sizes, and seismic effects. Next, the five types ware, and the members are simulated by PIPE20 (three-dimensional
of uncertainty analyses of record-to-record, structural design parame- plastic 2-node pipe element). The static elastic–plastic full-process anal-
ters, performance indicators, modelling, and construction are carried ysis is carried out on the structure, and the maximum nodal dis-
out through multiple sets of randomly selected reticulated shell exam- placement 𝛿𝑈 and the cumulative plastic energy dissipation 𝐸𝑝′ of the
ples. According to the record selection requirements of the Chinese GB static process are obtained when the structure reaches the critical
50011-2010 code [27], the ground motion records that can represent load of static stability. In the dynamic analysis, the load is simulated
most of the site characteristics in China are selected from ATC-63 [28] by MASS21 (three-dimensional mass element), a one-way El-Centro
to calculate the record-to-record uncertainty. Five structural design seismic wave is input to the structure, incremental dynamic analysis
parameters of yield strength, elastic modulus, damping ratio, dead load, (IDA) is performed with 100 cm/s2 as the peak ground acceleration

2
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

plastic members P are used as the criterion for judging the performance
states of reticulated shell structures. The performance states are divided
into five states: intact, slightly damaged, moderately damaged, severely
damaged, and collapsed, as shown in Table 1. Before the structure
enters the plastic state, the structure is considered to be ‘‘intact’’,
which can be judged according to P = 0, and the maximum nodal
displacement 𝛿𝑀 is less than 𝛿𝑃 at this time. When the dynamic equi-
librium equation of the structure cannot obtain a convergent solution
under earthquake action, the displacement diverges, and the structure
is considered to be ‘‘collapsed’’, which can be judged according to
𝛿𝑀 = ∞. Considering that the failure modes of reticulated shells are
dynamic instability and strength failure, the failure modes of most
Fig. 1. 𝐾8,6 single-layer spherical reticulated shell structure. reticulated shells include the above two. When dynamic instability is
the main failure mode, the displacement response is more obvious,
and the plastic development is not large; when strength failure is
the main failure mode, the plastic development is more obvious. To
render the obtained criterion universal, the three performance states of
‘‘slightly damaged’’, ‘‘moderately damaged’’ and ‘‘severely damaged’’
need to be judged from the two aspects of displacement and plastic
development. When the displacement and plastic development of the
structure satisfy the two failure states, the most unfavourable case is
taken. The judgement criterion for each performance point is shown in
Table 2.

2.3. The calculation parameters of reticulated shell structural models

Fig. 2. The development of the displacement term and energy dissipation term of the
reticulated shell structure with the intensity of ground motion. To obtain the values of 𝛼, 𝛽, and 𝛾 in Eq. (2) and the value of D at
each performance point, it is necessary to perform statistical analysis
on the displacement and energy dissipation of the reticulated shell
(PGA) step, and the maximum nodal displacement 𝛿𝑀 and the cumu- structures at each performance point through numerical simulation
lative plastic energy dissipation 𝐸𝑝 are recorded under each working examples. The values of D calculated at LS-4 are all infinite, so they
condition. The ratio of the maximum displacement of the structure are not counted here.
under earthquake action 𝛿𝑀 to the maximum displacement under static The single-layer spherical reticulated shell structure is a represen-
load 𝛿𝑈 is called the displacement term (𝛿𝑀 ∕𝛿𝑈 ), and the ratio of the
tative structure of the reticulated shells. This structure can reflect
cumulative plastic energy dissipation of the structure under earthquake
the basic mechanical characteristics of the reticulated shells and has
action 𝐸𝑝 and the cumulative plastic dissipation under static load 𝐸𝑝′ is
simple force characteristics. Therefore, according to JGJ 7-2010 [30],
called the energy consumption term (𝐸𝑝 ∕𝐸𝑝′ ). It can be seen from Fig. 2
K6 and K8 single-layer spherical reticulated shell structures with dif-
that with increasing ground motion intensity, the displacement term
ferent spans, rise-span ratios, and roof loads are selected for numerical
increases approximately linearly, while the energy consumption term
increases roughly in a power function form after the structure enters simulations. To ensure the universality of Eq. (2), the parameters
plasticity. are selected within the usual range of engineering. The model pa-
Therefore, based on the Park-Ang damage model and the response rameters are shown in Table 3. Three-way fixed hinged supports are
characteristics of the displacement and energy dissipation of reticulated located at the periphery of the structures. The structural material is
shells under earthquake action, a two-parameter damage model suit- Q235 steel, the yield strength is 235 MPa, and the elastic modulus
able for reticulated shell structures is proposed. As shown in Eq. (2), is 2.06 × 105 MPa. The ideal elastic–plastic constitutive model is
the undetermined parameters of 𝛼, 𝛽, and 𝛾 are introduced so that the adopted. The basic natural vibration period of 216 examples is in the
displacement term and the energy dissipation term can be used as a range of 0.28∼0.57 s, and the Rayleigh damping ratio is 0.02. Each
form of nonlinear combination. type of reticulated shell adopts two member sizes. The circumferential
𝛿 𝐸𝑝 members and the oblique members adopt a smaller size. The radial
𝐷 = 𝛼 ⋅ 𝑀 + 𝛽 ⋅ ( ′ )𝛾 (2) members adopt a larger size. The selected examples of reticulated shells
𝛿𝑈 𝐸𝑝
have been calculated for static stability, and the stability safety factor K,
where D represents the structural damage index; 𝛿𝑀 and 𝛿𝑈 represent
the stress strength of the members, and the structural deflection of each
the maximum nodal displacement response of the structure under
example meet the requirements of JGJ 7-2010 [30]. The models are
earthquake and static loads, respectively; 𝐸𝑝 and 𝐸𝑝} represent the
analysed by ANSYS Simulation Software. The members are simulated
accumulative plastic energy dissipation of the structure under earth-
by the BEAM188 element (three-dimensional two-node Timoshenko
quake and static loads, respectively; and 𝛼, 𝛽 and 𝛾 are undetermined
parameters. beam element), and each member is divided into three elements. In
the dynamic analysis, the load is simulated by the three-dimensional
2.2. Definition and division of the performance states for reticulated shell mass element MASS21. One-way ground motion records with four
structures different wave types (Northridge-01, Imperial Valley-06, Duzce, and
Hector Mine) are selected from ATC-63 and input into the structures.
According to GB 50011-2010 [27] and the research results of pre- To improve the calculation efficiency, a 95% energy duration [15] is
vious scholars [13,24], the maximum nodal displacement response selected. A total of 216 sets of calculation models are employed. The
under an earthquake 𝛿𝑀 , the maximum displacement response when IDA analysis with a PGA step size of 100 cm/s2 is performed on each
the structure begins to enter the plastic state 𝛿𝑃 , and the proportion of calculation model.

3
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 1
Performance states and division basis of reticulated shell structures.
Performance state The response of the structure Post-earthquake measures Division basis
Intact The structure is in the elastic state, the It can be used without repair after an 𝑃 =0
members are intact, and the earthquake. and 𝛿𝑀 < 𝛿𝑃
displacement development is very small.
Slightly damaged A small number of members begin to It can be used without repair or a little 0 < 𝑃 ≤ 15% or
enter the plastic state, and the repair after an earthquake. 𝛿𝑃 ≤ 𝛿𝑀 < 2.5𝛿𝑃
displacement development is not large.
Moderately damaged Many members enter the plastic state, It needs to be repaired, and some 15% < 𝑃 ≤ 45% or
and the displacement develops larger. reinforcement measures should be taken 2.5𝛿𝑃 ≤ 𝛿𝑀 < 10𝛿𝑃
before use after an earthquake.
Severely damaged Most of the members have entered the Major repairs should be carried out for 45% < 𝑃 < 100%
plastic state, and the displacement has risk elimination after an earthquake. or 𝛿𝑀 ≥ 10𝛿𝑃
developed greatly.
Collapsed The structure loses its bearing capacity. The structure needs to be demolished 𝛿𝑀 = ∞
and rebuilt.

Note: 𝑃 is the proportion of plastic members; 𝛿𝑀 is the maximum nodal displacement response of the structure under an earthquake; and 𝛿𝑃
is the maximum displacement response when the structure begins to enter the plastic state.

Table 2
Judgement criterion for each performance point.
Performance point LS-1 LS-2 LS-3 LS-4
Judgement criterion Start to appear Start to appear Start to appear Start to appear
𝑃 >0 𝛿𝑀 ≥ 2.5𝛿𝑃 or 𝑃 > 15% 𝛿𝑀 ≥ 10𝛿𝑃 or 𝑃 > 45% 𝛿𝑀 = ∞

Note: LS-1, LS-2, LS-3, and LS-4 are the four performance points that divide the five performance states; P is the proportion of plastic members;
𝛿𝑀 is the maximum displacement response of the structure under an earthquake; 𝛿𝑃 is the maximum displacement response when the structure
begins to enter the plastic state; and each judgement criterion is a critical state.

Table 3
Model parameters of K6 and K8 single-layer spherical reticulated shells.
Structure parameters Parameter values
Span-grid latitudinal number 40 m-6; 50 m-7; 60 m-8
Rise-span ratio 1/3; 1/5; 1/7
Roof load 60 kg/m2 ; 120 kg/m2 ; 180 kg/m2
Seismic effect Northridge-01; Imperial Valley-06; Duzce; Hector Mine
Member size(mm) 𝜙 95 × 3.5, 𝜙 102 × 3.5, 𝜙 114 × 3.5, 𝜙 121 × 3.5, 𝜙 127 × 3.5, 𝜙
133 × 3.5, 𝜙 140 × 3.5, 𝜙 146 × 3.5, 𝜙 152 × 3.5, 𝜙 159 × 4, 𝜙 168 × 4

2.4. Fitting of the two-parameter damage criterion

To render the D at each performance point more concise, first, the

D at the performance point LS-3 is set as 1, and the results of the
examples are fitted according to Eq. (2) to obtain 𝛼 = 0.2, 𝛽 = 0.4,
and 𝛾 = 0.1. Thus, the two-parameter damage criterion for reticulated
shell structures is:
𝛿 𝐸𝑝
𝐷 = 0.2 ⋅ 𝑀 + 0.4 ⋅ ( ′ )0.1 (3)
𝛿𝑈 𝐸𝑝

Then, the calculation results of the 216 sets of examples at LS-1

and LS-2 are substituted into Eq. (3), and their D values are shown
in Fig. 3. The mean values of the results of the calculation examples
at performance points LS-1 and LS-2 are 0.3 and 0.6, respectively. Fig. 3. The D value results corresponding to all levels of performance points.

The variances are both 0.002, indicating that the fitting is accurate.
Therefore, Eq. (3) can be used as the two-parameter damage criterion
for reticulated shell structures. In addition, it can be seen from Fig. 3 and seismic hazard analysis. In light of the total probability theorem,
that the three sets of data points all approach their mean values, so it can be expressed as [1,2,5–7,10–12,32]
0.3, 0.6, and 1 can be taken as the division indicators for each level of [ ] ∑ [ ] [ ]
performance points, as shown in Table 4. The universality and validity 𝑃 𝐿𝑆𝑖 = 𝑃 𝐿𝑆 ||𝛼𝐼𝑀 = 𝑥𝑖 ⋅ 𝑃 𝛼𝐼𝑀 = 𝑥𝑖 (4)
of the damage criterion (Eq. (3)) and division indexes (Table 4) have
Since the actual ground motion intensity is continuous, Eq. (4) can be
been well verified by two large-scale shaking table reticulated shell test
expressed as a continuous form:
models [31].
[ ]
𝑃 𝐿𝑆𝑖 = 𝐹𝑅 (𝑥) ⋅ |𝑑𝐻 (𝑥)| (5)
∫
3. Seismic risk assessment methodology
In Eq. (5), 𝐹𝑅 (x)=P[LS| 𝛼𝐼𝑀 = x 𝑖 ] represents the seismic vul-
Seismic risk assessment can calculate the annual probability P[LS𝑖 ] nerability function, which can measure the relationship between the
for a structure that exceeds the ith performance point under different intensity of ground motion and structural damage from the proba-
possible seismic conditions. It includes seismic vulnerability analysis bilistic perspective and is usually described by a lognormal function:

4
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 4
Performance state divisions of the two-parameter damage criterion for reticulated shell structures.
Performance state Intact Slightly damaged Moderately damaged Severely damaged Collapsed
Division criterion 𝛿𝑀 < 𝛿𝑃 𝛿𝑃 ≤ 𝛿𝑀 < 2.5𝛿𝑃 2.5𝛿𝑃 ≤ 𝛿𝑀 < 10𝛿𝑃 𝛿𝑀 >10 𝛿𝑃 𝛿𝑀 = ∞
𝑃 =0 0 < 𝑃 ≤ 15% 15% < 𝑃 ≤ 45% 45% < 𝑃 < 100%
D 0 ≤ D < 0.3 0.3 ≤ D < 0.6 0.6 ≤ D < 1.0 D ≥ 1.0 D = ∞

Note: P is the proportion of plastic members; 𝛿𝑀 is the maximum displacement response of the structure under earthquake action; 𝛿𝑃 is the
maximum displacement response when the structure begins to enter the plastic state; and D is the structural damage index.

where 𝛽𝐷 , 𝛽𝐶 , 𝛽𝑆 , 𝛽𝑀 , and 𝛽𝑄 represent the logarithmic standard devia-

[ ( )]
ln 𝑥∕𝑚𝑅 tion due to record-to-record, structural design parameters, performance
𝐹𝑅 (𝑥) = 𝛷 (6) indicators, modelling, and construction, respectively.
𝛽𝑅

In Eq. (6), 𝑚𝑅 and 𝛽𝑅 represent the mean value and the logarithmic 4.1. Record-to-record uncertainty
standard deviation of the seismic vulnerability function, respectively.
In Eq. (5), H (x) is the seismic hazard function, which represents the The record-to-record uncertainty at the site of the structure is
annual probability of an earthquake with a ground motion intensity related to the site information, the spectral characteristics of ground
𝛼𝐼𝑀 ≥ 𝑥 and is generally expressed as a power function: motions, and the peak acceleration. Nine randomly selected K6 and
[ ] K8 single-layer spherical shells considering different spans, stand-span
𝐻 (𝑥) = 𝑃 𝛼𝐼𝑀 ≥ 𝑥 ≈ 𝑘0 𝑥−𝑘 (7)
ratios, loads, and other factors are used as the research objects of
In Eq. (7), 𝑘0 and k are shape parameters that can be obtained by the this section, and the basic information is shown in Table 5. (i) Ex-
design-based earthquake (DBE) and maximum considered earthquake ample 1 uses the fixed supports around the structure and the bilinear
(MCE) [32]: strengthening constitutive model with a strain hardening rate of 0.02.
( ) (ii) Example 2 and Example 3 are the large-scale shaking table test
ln 𝑣𝐷𝐵𝐸 ∕𝑣𝑀𝐶𝐸
𝑘= ( ) (8) models of Ye [29], with a test scale ratio of 1:3.5. The distribution of
ln 𝑎𝐼𝑀,𝑀𝐶𝐸 ∕𝑎𝐼𝑀,𝐷𝐵𝐸 members was obtained through the similarity ratio design and the full
( ) [ ( )𝑘 ]
ln 𝑘0 = ln 𝑣𝐷𝐵𝐸 𝑎𝐼𝑀,𝑀𝐶𝐸 stress design. The overall stiffness of Example 2 (Model 1) is uniform
( ) ( ) ( ) ( ) (Fig. 4(a)). Example 3 (Model 2) is based on Example 2 (Model 1) with
ln 𝑎𝐼𝑀,𝐷𝐵𝐸 ⋅ ln 𝑣𝑀𝐶𝐸 − ln 𝑎𝐼𝑀,𝑀𝐶𝐸 ⋅ ln 𝑣𝐷𝐵𝐸
= ( ) (9) two weak zones in the large dynamic response region (Fig. 4(b)). The
ln 𝑎𝐼𝑀,𝐷𝐵𝐸 ∕𝑎𝐼𝑀,𝑀𝐶𝐸 support constraints are shown in Fig. 4(c). The material parameters
In Eq. (9), DBE corresponds to fortified earthquakes (moderate earth- obtained from the experimental measurements are used to simulate
quakes), the probability of exceedance is 10% in 50 years, and the the material constitutive model by the piecewise linear plastic model.
return period is 475 years; MCE corresponds to rare earthquakes (severe Further details of the shaker testing and dynamic analysis can be found
earthquakes), the probability of exceedance is 2% in 50 years, and in the literature [29,31,33,34]. (iii) For Examples 4∼9, three-way fixed
the return period is 2475 years; 𝛼𝐼𝑀,𝐷𝐵𝐸 and 𝛼𝐼𝑀,𝑀𝐶𝐸 represent the hinged supports are located at the periphery of these structures, and
ground motion intensity corresponding to DBE and MCE; and 𝑣𝐷𝐵𝐸 the ideal elastic–plastic constitutive model is used.
and 𝑣𝑀𝐶𝐸 represent the average annual probability of exceedance Several reticulated shell calculation examples (Example 1, Example
corresponding to DBE and MCE, respectively. 4∼Example 9) and two test models (Example 2 and Example 3) are
Eqs. (6) and (7) are substituted into Eq. (5) to obtain the annual analysed in the finite element analysis software ANSYS and ANSYS/LS-
probability of exceedance: DYNA, respectively. The members are simulated by using the three-
[ ] ( ) ( )2 dimensional two-node Timoshenko beam element, with each member
𝑃 𝐿𝑆𝑖 = 𝑘0 𝑚−𝑘 𝑅 exp[ 𝑘𝛽𝑅 ∕2] (10) divided into three elements; the roof loads are applied to the nodes
Then, the probability of exceedance in 50 years can be obtained: in the form of equivalent concentrated loads and are simulated by the
[ ] three-dimensional mass element. This is a common modelling method
𝑃𝐿𝑆,50 = 1 − (1 − 𝑃 𝐿𝑆𝑖 )50 (11) used by scholars [13–18] in the field of reticulated shell seismic resis-
tance. The correctness of the numerical simulation method has been
Therefore, the values of 𝑚𝑅 , 𝛽𝑅 , k, and 𝑘0 are the key parameters in
verified by the experimental results [29]. Q235 steel is selected as the
the seismic risk assessment. 𝑚𝑅 and 𝛽𝑅 can be determined by seismic
material, the yield strength is 235 MPa, and the elastic modulus is
vulnerability analysis (Eq. (6)), and k and 𝑘0 can be determined by
2.06 × 105 MPa. The basic natural vibration period of each example
seismic hazard analysis (Eqs. (8) and (9), respectively).
is shown in Table 5, and the damping ratio is 0.02.
4. Multiple uncertainty analysis The representative ground motion records of the site need to be
selected for the record-to-record uncertainty analysis. Since China has
The traditional seismic risk assessment of reticulated shell structures not yet built a complete ground motion record database and earthquake
only takes record-to-record and structural design parameters as uncer- disaster maps, there are few strong earthquake records, and most of
tain factors. However, in an actual reticulated shell project, the perfor- them are far-field records beyond 200 km. However, the selection
mance indicators to measure the damage of structures, the modelling criteria for far-field ground motion records in the ATC-63 report [28]
quality of the designers, and the errors caused by the construction will are relatively comprehensive and universal. Therefore, the 22 groups
also affect the evaluation results. Therefore, in this paper, five types of of far-field records in the ATC-63 report are used as the basis for
uncertainties, namely, record-to-record, structural design parameters, record selection. Since 𝛽𝐷 should have universality, the selected ground
performance indicators, modelling, and construction, are considered. motion records should be able to represent most of the sites in China.
Using the method by Celik [7], Yin [10], and Jiang [5,11,12], we Therefore, the I1 , II, and III sites in the fortified areas of 6, 7, 8,
assume that the uncertainties from different sources are independent and 9 degrees in China are considered (do not consider the I0 and IV
and uncorrelated. Then, the logarithmic standard deviation can be sites because they are too strong or weak). The selection of ground
calculated by using the SRSS: motion records is carried out according to the principle of GB 50011-
√ 2010 in China [27], which is ‘‘the maximum error between the mean
𝛽𝑅 = 𝛽𝐷 2 + 𝛽2 + 𝛽2 + 𝛽2 + 𝛽2 (12) value of the acceleration response spectrum of the input multiple
𝐶 𝑆 𝑀 𝑄

5
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 5
Examples of record-to-record uncertainty.
Number Type Span Rise-span Roof load Member sizes Basic natural
(m) ratio (kg/m2 ) (mm) vibration period (s)
1 𝐾8,8 40 1/3 157.5 Circumferential 0.28
members: 𝜙 89 × 3
Radial and oblique
members: 𝜙 76 × 3
2 𝐾6,20 Model: 23.4 1/2 Prototype: 115 Model: 𝜙 23 × 1; 𝜙 0.55
(Model 1) Prototype: 81.9 38 × 2;
𝜙 63.5 × 3.5; 𝜙 114 × 4
3 𝐾6,20 Model: 23.4 1/2 Prototype: 115 Model: 𝜙 18 × 1; 𝜙 0.55
(Model 2) Prototype: 81.9 23 × 1;
𝜙 38 × 2; 𝜙 63.5 × 3.5;
𝜙 114 × 4
4 𝐾6,6 40 1/3 60 𝜙 114 × 3.5 0.28
5 𝐾6,7 50 1/5 60 𝜙 140 × 3.5 0.31
6 𝐾6,8 60 1/7 90 𝜙 168 × 3.5 0.42
7 𝐾8,8 40 1/3 60 𝜙 102 × 3.5 0.24
8 𝐾8,7 50 1/7 90 𝜙 121 × 3.5 0.35
9 𝐾8,8 60 1/5 60 𝜙 168 × 3.5 0.26

Note: Example 2 and Example 3 are the large-scale shaking table test models, Model 1 and Model 2, respectively. The structural models and
the distribution of weakened zones are shown in Fig. 4.

Fig. 4. Large-scale shaking table test models [29].

sets of ground motion records and the site’s strong earthquake design The 20 selected ground motion records are input into the structures,
response spectrum is not more than 20% at the basic natural vibration the results are fitted by linear regression, and the logarithmic standard
period point of the structure [27]’’. Referring to Jiang [5,11,12] for the deviation is obtained:
√
number of selected records and according to the selection requirements ∑20 ( ( ))2
of the Chinese code [27], 20 ground motion records are selected from 𝑖=1 ln (𝐷) − ln 𝑚𝐷
𝛽𝐷 = (15)
the 22 groups of far-field ground motion records in ATC-63 [28] to 20 − 2
analyse the nine calculation examples in Table 5. The selection results Fig. 5 shows the seismic demand regression curves and 𝛽𝐷 calcu-
of ground motion records are shown in Table 6. Group-1 is input to the lation results of nine sets of reticulated shell examples. The record-to-
K6 reticulated shells (Example 2∼Example 6), and Group-2 is input to record uncertainty 𝛽𝐷 is in the range of 0.712∼0.911. It can be seen
the K8 reticulated shells (Example 1, Example 7∼Example 9). that the record-to-record uncertainty of the reticulated shell structure is
The record-to-record uncertainty can be determined by the seismic relatively high, and similar conclusions can be obtained in the research
demand uncertainty [11]. The seismic demand parameter characterizes of Zhong [16,17]. To ensure the universality and safety of the seismic
the response of the structures under the earthquake, such as displace- risk assessment results, it is recommended to take 0.9 as the index value
ment, interstory drift angle, and damage index. In this paper, D in 𝛽𝐷 of the record-to-record uncertainty.
Eq. (3) is used to measure the damage of the reticulated shell structures,
so it is taken as the seismic demand parameter. Peak ground acceler- 4.2. Uncertainties of the structural design parameters
ation (PGA) and spectral acceleration are commonly used as ground
motion intensity parameters. PGA is used by the Chinese GB 50011- The selected examples cover K8 and K6 single-layer spherical shells,
2010 code [27] to characterize ground motion intensity, and there is a considering different spans, rise-span ratios, loads, and other factors,
corresponding relationship between PGA and fortification intensity, so as shown in Table 7. The parameters and modelling methods for
PGA is taken as the ground motion intensity parameter in this paper. Examples 1∼3 are the same as those for Examples 1∼3 in Section 4.1.
It is generally assumed that the seismic demand parameters obey the Examples 4 and 5 are surrounded by three-way fixed hinge supports.
lognormal distribution and that there is a power function relationship The material is Q235 steel, the yield strength is 235 MPa, and the
between the mean value 𝑚𝐷 and the ground motion intensity parameter elastic modulus is 2.06 × 105 MPa. The ideal elastic–plastic constitutive
PGA [1,2,5–7,10–12] model is used. The damping ratio is 0.02. The finite element analysis
software ANSYS is used for the analysis. The members are simulated
𝑚𝐷 = 𝑎(𝑃 𝐺𝐴)𝑏 (13) by BEAM188 (three-dimensional two-node Timoshenko beam element),
Taking the logarithm of both sides of Eq. (13) to obtain Eq. (14): each divided into three members. The roof loads are applied to the
nodes in the form of equivalent concentrated loads and are simulated
( )
ln 𝑚𝐷 = ln 𝑎 + 𝑏 ln (𝑃 𝐺𝐴) (14) by the three-dimensional mass element MASS21. This is a common

6
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 6
Two sets of ground motion records (20 records per group).
Group-1-K6 Group-2-K8
Number RSN Event/Station Name PGA/g Number RSN Event/Station Name PGA/g
1 68 San Fernando/LA-Hollywood Stor FF 0.23 1 68 San Fernando/LA-Hollywood Stor FF 0.23
2 125 Friuli Italy-01/Tolmezzo 0.36 2 125 Friuli Italy-01/Tolmezzo 0.36
3 169 Imperial Valley-06/Delta 0.35 3 169 Imperial Valley-06/Delta 0.35
4 174 Imperial Valley-06/El Centro Array #11 0.37 4 174 Imperial Valley-06/El Centro Array #11 0.37
5 721 Superstition Hills-02/El Centro Imp. Co. Cent 0.36 5 721 Superstition Hills-02/El Centro Imp. Co. Cent 0.36
6 725 Superstition Hills-02/Poe Road (temp) 0.48 6 725 Superstition Hills-02/Poe Road (temp) 0.48
7 752 Loma Prieta/Capitola 0.51 7 752 Loma Prieta/Capitola 0.51
8 767 Loma Prieta/Gilroy Array #3 0.56 8 767 Loma Prieta/Gilroy Array #3 0.56
9 829 Cape Mendocino/Rio_Dell_Overpass-FF 0.39 9 848 Landers/Coolwater 0.42
10 848 Landers/Coolwater 0.42 10 900 Landers/Yermo Fire Station 0.25
11 900 Landers/Yermo Fire Station 0.25 11 953 Northridge-01/Beverly Hills-14145 Mulhol 0.49
12 960 Northridge-01/Canyon Country - W Lost Cany 0.47 12 960 Northridge-01/Canyon Country - W Lost Cany 0.47
13 1111 Kobe, Japan/Nishi-Akashi 0.48 13 1111 Kobe, Japan/Nishi-Akashi 0.48
14 1116 Kobe, Japan/Shin-Osaka 0.23 14 1116 Kobe, Japan/Shin-Osaka 0.23
15 1148 Kocaeli, Turkey/Arcelik 0.21 15 1148 Kocaeli, Turkey/Arcelik 0.21
16 1158 Kocaeli, Turkey/Duzce 0.36 16 1158 Kocaeli, Turkey/Duzce 0.36
17 1244 Chi-Chi, Taiwan/CHY101 0.40 17 1244 Chi-Chi, Taiwan/CHY101 0.40
18 1485 Chi-Chi, Taiwan/TCU045 0.47 18 1602 Duzce, Turkey/Bolu 0.81
19 1602 Duzce, Turkey/Bolu 0.81 19 1633 Manjil, Iran/Abbar 0.21
20 1633 Manjil, Iran/Abbar 0.21 20 1787 Hector Mine/Hector 0.33

Note: RSN is the ground motion record number of the US Pacific Earthquake Research Center (PEER) NGA-West 1 database. Seismic record Group-1 is input into K6 reticulated
shells, and seismic record Group-2 is input into K8 reticulated shells.

Fig. 5. Seismic demand regression curves and 𝛽𝐷 .

7
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 7
Examples of uncertainties in the structural design parameters.
Number Type Span Rise-span Dead load Snow load Ground motion Member sizes Basic natural
(m) ratio (kg/m2 ) (kg/m2 ) record (mm) vibration period (s)
1 𝐾8,8 40 1/3 143.48 28.02 RSN-68 Circumferential 0.28
members: 𝜙 89 × 3
Radial and oblique
members: 𝜙 76 × 3
2 𝐾6,20 Model: 23.4 1/2 Model: 90 Model: 50 RSN-68 Model: 𝜙 23 × 1; 𝜙 0.55
(Model 1) Prototype: 38 × 2;
81.9 𝜙 63.5 × 3.5; 𝜙
114 × 4
3 𝐾6,20 Model: 23.4 1/2 Model: 90 Model: 50 RSN-68 Model: 𝜙 18 × 1; 𝜙 0.55
(Model 2) Prototype: 23 × 1;
81.9 𝜙 38 × 2;
𝜙 63.5 × 3.5;
𝜙 114 × 4
4 𝐾6,7 50 1/5 95 50 RSN-900 Radial members: 𝜙 0.42
140 × 3.5
Circumferential and
oblique members: 𝜙
133 × 3.5
5 𝐾8,8 60 1/7 75 50 RSN-1111 Radial members: 𝜙 0.43
168 × 4
Circumferential and
oblique members: 𝜙
159 × 4

Note: Examples 2 and 3 are the large-scale shaking table test models, Model 1 (normal design) and Model 2 (weak), respectively; and RSN is the ground motion record number
of the US PEER NGA-West 1 database.

Table 8
Probability distribution models and distribution parameters of the structural random parameters.
Parameter Distribution Mean value Coefficient of References
(unit) variation
𝑓𝑦 (MPa) Lognormal 235 0.08 Lv [37] and Dai [38]
E (MPa) Lognormal 2.06 × 105 0.03 JCSS [39] and GB
50017-2017 [40]
𝜁 (–) Normal 0.02 0.4 Cao [41]
DL (kg/m2 ) Normal 143.48 (Example 1) 0.07 GB 50011-2010 [42]
80.71 (Examples 2, 3) and Gao [36]
95 (Example 4)
75 (Example 5)
SL (kg/m2 ) Gumbel Extreme 28.02 (Example 1) 0.229 GB 50011-2010 [42]
value type I 23.55 (Examples 2, 3) and Wang [43]
50 (Examples 4, 5)

√
√ 5
modelling method used by scholars [13–18] in the field of seismic √∑
resistance of reticulated shells. The basic natural vibration period of 𝛽𝐶 = √ 𝛽𝑐−𝑖
2 (17)
𝑖=1
each example is shown in Table 7.
According to the results of Zhong [16,35] and Gao [36], the steel where i (𝑖 = 1, 2, 3, 4, 5) represents the parameter; j (𝑗 = 1, 2, 3, . . .
yield strength 𝑓𝑦 , elastic modulus E, damping ratio 𝜁, dead load DL, 20) represents the number of sampling frequencies; 𝛽𝑐−𝑖 represents the
and snow load SL are considered uncertain factors for the structural logarithmic standard deviation due to the uncertainties of the structural
design parameters. The probability distribution models and distribution design parameters; 𝑃 𝐺𝐴𝑖,𝑗 represents the PGA value of the 𝑗th sampling
parameters are shown in Table 8. for parameter i at a certain performance point; and ln(𝑃 𝐺𝐴𝑖,𝑗 ) repre-
Many scholars, such as Vamvatsikos [44], Celik [7], and Yin [10],
sents the mean value of ln(𝑃 𝐺𝐴𝑖,𝑗 ) in the 20 samples for parameter i
believe that each structural design parameter can be approximately
at a certain performance point.
independent. In addition, according to the research of Dolsek [45],
when the number of samples is more than twice the number of vari- Table 9 shows the calculation results of 𝛽𝐶 for each example. The 𝛽𝐶
able parameters, the number of samples has little effect on the mean range at performance point LS-1 is larger (0.094∼0.513), while the 𝛽𝐶
value and logarithmic standard deviation of the structural response. at the other performance points is in the range of 0.1∼0.3. When the
Therefore, to save calculation costs, for each reticulated shell structure, structure has a failure mode dominated by dynamic instability (such
only one ground motion record is input for IDA. The corresponding as in Examples 2 and 3), the stochasticity of the material parameters
ground motion records in Table 7 are input into each example for has little effect on 𝛽𝐶 , while when the structure has a failure mode
IDA analysis, and the LHS method is used to select 20 sets of samples dominated by strength failure (such as in Examples 1, 4, and 5), 𝛽𝐶 is
for each structural parameter in Table 8. When analysing 𝛽𝑐−𝑖 , other greatly affected by random factors [35]. Since the performance point
parameters are taken as the mean value [11,12], and finally, the SRSS LS-1 is in the critical state of elasticity and plasticity, the 𝛽𝐶 range
method, recommended by many scholars [5,10–12], is used to calculate
obtained here is larger. Therefore, considering the universality and
𝛽𝐶 :
√ safety of the seismic risk assessment results, it is recommended to take
∑20 2 0.3 as the index value 𝛽𝐶 of the uncertainties of the structural design
𝑗=1 (ln(𝑃 𝐺𝐴𝑖,𝑗 ) − ln(𝑃 𝐺𝐴𝑖,𝑗 ))
𝛽𝑐−𝑖 = (16) parameters.
20 − 1

8
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 9
Calculation results of the uncertainties of the structural design parameters 𝛽𝐶 .
Performance point Example 1-𝛽𝐶 Example 2-𝛽𝐶 Example 3-𝛽𝐶 Example 4-𝛽𝐶 Example 5-𝛽𝐶
LS-1 (𝐷 = 0.3) 0.385 0.099 0.094 0.513 0.469
LS-2 (𝐷 = 0.6) 0.197 0.116 0.130 0.255 0.288
LS-3 (𝐷 = 1.0) 0.129 0.266 0.121 0.163 0.232
LS-4 (D = ∞) 0.107 0.258 0.111 0.293 0.316

Note: D represents the index of structural damage.

Table 10
Probability distribution model and distribution parameters of 𝛼 and 𝛽.
Parameter Distribution Mean value Standard deviation
𝛼 0.2 0.233
Lognormal
𝛽 0.4 0.142

4.3. Uncertainties of the performance indicators

In this paper, the two-parameter damage criterion (Eq. (3)) is

adopted as the performance index. However, due to the difference
in size, load, and other factors of the reticulated shells, at the same
performance point, different displacement parameters 𝛼 and energy dis- Fig. 6. Variation curves of the uncertainties of performance indicators 𝛽𝑆 with the
number of samples N.
sipation parameters 𝛽 can be obtained (taking 𝛾 = 0.1 as a fixed value),
thus reflecting the uncertainties of performance indicators (Eq. (3)).
By fitting the 216 sets of comprehensive examples in Section 2.3, it Table 11
Calculation results of the uncertainties of performance indicators 𝛽𝑆 .
is verified by the K–S test [46] that the parameters 𝛼 and 𝛽 at perfor-
Performance point LS-1 (𝐷 = 0.3) LS-2 (𝐷 = 0.6) LS-3 (𝐷 = 1.0)
mance points LS-1, LS-2, and LS-3 obey the lognormal distribution. The
probability distribution parameters are shown in Table 10. Example 1-𝛽𝑆 0.065 0.368 0.332
Example 2-𝛽𝑆 0.266 0.451 0.216
The five reticulated shell examples of Table 7 in Section 4.2 are Example 3-𝛽𝑆 0.159 0.545 0.257
taken as research objects, and the corresponding ground motion records Example 4-𝛽𝑆 0.224 0.448 0.235
in Table 7 are input into each example for IDA analysis. The LHS Example 5-𝛽𝑆 0.167 0.507 0.265
method is used to sample the parameters 𝛼 and 𝛽; it is assumed that Note: D represents the index of structural damage. Since the values of D at performance
the parameters are independent of each other. When analysing 𝛽𝑠−𝑖 , the point LS-4 are infinite and are not affected by the parameters 𝛼 and 𝛽, the uncertainties
other parameters are taken as the mean values, and the SRSS method of the performance indicators are not considered here.

is finally used to calculate 𝛽𝑆 :

√
√ 2
√∑ 4.4. Modelling uncertainty
𝛽𝑆 = √ 𝛽𝑠−𝑖 2 (18)
𝑖=1
√ There are errors between the results of numerical simulations and
∑𝑁 2
𝑗=1 (ln(𝑃 𝐺𝐴𝑖,𝑗 ) − ln(𝑃 𝐺𝐴𝑖,𝑗 )) the actual structures, and the uncertainty of errors caused by different
𝛽𝑠−𝑖 = (19) simulation qualities is called modelling uncertainty. Since reticulated
𝑁 −1
shell structures are complex and difficult to model, it is necessary to
where i (𝑖 = 1, 2) represents the parameter; j (𝑗 = 1, 2, 3, . . . N )
calculate the modelling uncertainty based on the case data. Therefore,
represents the sampling frequency; 𝛽𝑠−𝑖 represents the logarithmic stan-
this paper sorts out the literature related to the comparison of dynamic
dard deviation due to the uncertainties of the performance indicators;
and static tests and numerical simulations of reticulated shell structures
𝑃 𝐺𝐴𝑖,𝑗 represents the PGA value of the 𝑗th sampling for parameter i at
and performs statistical analysis on the error values (as shown in
a certain performance point; and ln(𝑃 𝐺𝐴𝑖,𝑗 ) represents the mean value
Table 12). The modelling uncertainty 𝛽𝑀 = 0.75 can be obtained by
of ln(𝑃 𝐺𝐴𝑖,𝑗 ) in the N samples for parameter i at a certain performance
Eq. (20).
point. √
√
With respect to the uncertainties of performance indicators (two- √ ∑15 ( )2
√
parameter damage criterion), there is no literature to refer to regarding √ 𝑖 ln(𝑒𝑖 ) − ln(𝑒𝑖 )
𝛽𝑀 = (20)
the number of samples. Therefore, Example 1 in Table 7 is taken as the 15 − 1
research object, and 10 up to 100 sets of samples for the parameters 𝛼 where 𝑒𝑖 (𝑖 = 1, 2, 3, … 15) represents the error value in Table 12 and 𝑒𝑖
and 𝛽 are taken to calculate 𝛽𝑆 at performance points LS-1, LS-2, and represents the mean value of error samples in 15 groups (Table 12).
LS-3. As shown in Fig. 6, after N ≥30, the development of 𝛽𝑆 at each
performance point tends to be a straight line. Therefore, when solving 4.5. Construction uncertainty
for the uncertainties of performance indicators, it is sufficient to take
30 sets of samples. Node position deviation is the main error in construction, and
Table 11 shows the calculation results of each example. At LS- node spatial position control is also an important link in the con-
1, 𝛽𝑆 is in the range of 0.065∼0.266; at LS-2, 𝛽𝑆 is in the range of struction process. The critical stability-bearing capacity of reticulated
0.368∼0.545; at LS-3, 𝛽𝑆 is in the range of 0.216∼0.332. It can be seen shell structures is particularly sensitive to initial imperfections, and
that the calculated results have little fluctuation at each performance a small imperfection can cause a significant reduction in the critical
point. Therefore, considering safety and universality, it is recommended stability-bearing capacity. Therefore, in the seismic risk assessment
that at performance points LS-1, LS-2, and LS-3, the uncertainties of of reticulated shells, the uncertainty caused by construction node er-
performance indicators 𝛽𝑆 can be set to 0.2, 0.5, and 0.3, respectively. rors of the critical stability-bearing capacity is called the construction
uncertainty 𝛽𝑄 . Construction uncertainty can be gradually eliminated

9
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 12 5. Seismic risk assessment

Errors between the reticulated shell tests and numerical simulations.
Type Error Reference 5.1. Seismic vulnerability curves
19.35% Nie [47]
19.1%, 13.6%, 19% Ye [48]
Two shaking table test models of Ye [29] are utilized as examples
28.57% Zhang [49]
Shaking table test of 18.5% Ding [50] (Fig. 4), and the recommended value of each uncertainty index in
reticulated shells 25% Liu [51] Section 4 is used to carry out the seismic risk assessment of reticulated
28.58% Zhang [52] shell structures considering multiple uncertainties. PGA is used as the
30%, 23.17% Lu [29] ground motion intensity parameter. Twenty sets of ground motion
7.5% Wang [53]
records in Group-1 (Table 6) are input into the two models for IDA
1.66% Ma [54]
to obtain 𝑚𝑅 , and the step is 100 cm/s2 . The two-parameter damage
Static test of reticulated shells 25%, 15% Luo [55]
8.4% Xiong [56]
criterion in Section 2.4 is used as the division criterion. Table 13
shows the key parameters to describe the vulnerability curves based
on Eq. (12). Fig. 8 shows the vulnerability curves of the two models.
It can be seen from Fig. 8 that with increasing uncertain factors, the
vulnerability curve will rotate clockwise around the mean value 𝑚𝑅 .
Because Model 2 had two weakened zones, the damage of Model 2
is more severe than that of Model 1 under the same ground motion
intensity, which is in line with the original test design and test results.
In addition, the area of ‘‘severely damaged’’ in the vulnerability curves
of the two models is very small. That is, with increasing PGA, when
the performance state exceeds the ‘‘moderately damaged’’ state, it can
easily collapse over the ‘‘severely damaged’’ state, reflecting the failure
mode dominated by dynamic instability.

5.2. Seismic hazard analysis and the 50 year probabilities of exceedance

The models are located at the 8 (0.2 g) degree zone, class I1 site,
Fig. 7. Variation curve of the construction uncertainty 𝛽𝑄 with the number of samples
group 2. According to Chinese GB 50011-2010 [27], 𝑘 = 2.38 and
N.
𝑘0 = 4.56×10−5 are obtained by Eqs. (8) and (9), respectively. Then, the
50 year probability of exceedance 𝑃𝐿𝑆,50 can be obtained by Eq. (11)
and the calculation results are shown in Fig. 9. It can be seen that these
through the improvement of both the technical level of construction
five types of uncertainties have different exceedance probabilities at the
personnel and the construction technology equipment. This type of
four performance points and that the failure probabilities increase with
uncertainty belongs to epistemic uncertainty.
increasing uncertain factors.
The node errors are considered by adding random parameters to
After considering the uncertainties of structural design parameters,
the node coordinates. The positioning deviation of each node in the X,
the 50 year probabilities of exceedance of the two models at each
Y, and Z directions can be regarded as obeying a normal distribution. performance point are 1.3 times the probabilities determined when
The maximum installation deviation is R, the mean value is 0, and only considering the record-to-record uncertainty. The effect of uncer-
the standard deviation is R/3 (according to the ‘‘3𝜎’’ principle). It tainty in structural design parameters on the 50 year probabilities of
is assumed that the random variable of the installation deviation for exceedance for reticulated shell structures is relatively small. Similar
each node is independent [57]. Static stability full-process analysis is findings have been found in steel frame structures [2] and cold-formed
performed on N reticulated shell samples with different errors in three steel structures [11]. However, if the uncertainties of structural design
directions of each node, and N stable critical bearing capacities are parameters are ignored, then unsafe seismic risk assessment results will
obtained. Then, the construction uncertainty 𝛽𝑄 is obtained by Eq. (21): be evident [10].
After considering the uncertainties of performance indicators, the
√
√
√ ∑𝑁 ( )2 50 year probabilities of exceedance of the two models at each perfor-
√
√ 𝑖 ln(𝐹𝑖 ) − ln(𝐹𝑖 ) mance point are 1.1∼2 times the probabilities determined when only
𝛽𝑄 = (21) considering the uncertainties of record-to-record and structural design
𝑁 −1
parameters. It is necessary to consider the uncertainties of performance
where 𝐹𝑖 (𝑖 = 1, 2, 3, … 𝑁) represents the critical stability-bearing ca- indicators in seismic risk assessment.
pacity; 𝐹𝑖 represents the mean value of the critical stability-bearing After considering the modelling uncertainty, the 50 year proba-
capacity; and N represents the number of samples. bilities of exceedance of the two models at each performance point
JGJ 7-2010 [30] stipulates that the node error cannot exceed L/300, are 3 to 4.9 times the probabilities determined when only considering
where L is the span of the reticulated shell. Therefore, 𝑅 = 𝐿∕300 is the uncertainties of record-to-record, structural design parameters, and
taken, and 𝑁 = 1000 is temporarily taken. Example 1 of Table 7 in performance indicators. Compared with other uncertain factors, the
Section 4.2 is taken as the research object, and the 𝛽𝑄 value obtained modelling uncertainty has the greatest impact on the 50 year proba-
by Eq. (21) is plotted in Fig. 7. As the number of samples N increases, bility of exceedance for reticulated shell structures, which is related to
the logarithmic standard deviation of construction uncertainty 𝛽𝑄 is the complex structural form of reticulated shells, difficult modelling,
between 0.150 and 0.154 after 𝑁 = 300, so 300 samples are sufficient and different model quality. However, modelling uncertainty is a type
in this part. of epistemic uncertainty and can be eliminated by human factors.
Then, Examples 1, 4, and 5 in Table 7 in Section 4.2 are taken as re- Therefore, designers should strive to improve the technical level of
search objects, and the calculation results are 0.151, 0.070, and 0.059, reticulated shell structure modelling.
respectively. To ensure the safety and universality of the seismic risk After considering the construction uncertainty, the 50 year proba-
assessment results, it is recommended that the construction uncertainty bilities of exceedance of the two models at each performance point are
𝛽𝑄 be 0.1. 1.03 times the probabilities determined only considering the aleatoric

10
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Table 13
Vulnerability parameters of reticulated shells with multiple uncertainties.
𝑚𝑅 (g) 𝑚𝑅 (g) Different combinations of 𝛽𝑅
Performance point
Model 1 Model 2
𝛽𝐷 𝛽𝐷 , 𝛽𝐶 𝛽𝐷 , 𝛽𝐶 , 𝛽𝑆 𝛽𝐷 , 𝛽𝐶 , 𝛽𝑆 , 𝛽𝑀 𝛽𝐷 , 𝛽𝐶 , 𝛽𝑆 , 𝛽𝑀 , 𝛽𝑄
LS-1 (𝐷 = 0.3) 0.5 0.4 0.9 0.949 0.970 1.226 1.230
LS-2 (𝐷 = 0.6) 1.8 1.7 0.9 0.949 1.072 1.309 1.312
LS-3 (𝐷 = 1.0) 2.5 2.2 0.9 0.949 0.995 1.246 1.250
LS-4 (D = ∞) 2.6 2.3 0.9 0.949 0.949 1.209 1.214

Note: D represents the damage index of reticulated shell structures; 𝑚𝑅 represents the mean value of the seismic vulnerability function; 𝛽𝑅
represents the logarithmic standard deviation of the seismic vulnerability function; and 𝛽𝐷 , 𝛽𝐶 , 𝛽𝑆 , 𝛽𝑀 , and 𝛽𝑄 represent the logarithmic
standard deviation due to record-to-record, structural design parameters, performance indicators, modelling, and construction, respectively.
Each uncertainty value is the recommended value in Section 4.

Fig. 8. Vulnerability curves.

uncertainties (record-to-record, structural design parameters, perfor- (2) Considering most sites in China, PGA, which can correspond to
mance indicators) and modelling uncertainty. the fortification intensity of China, is used as the ground motion in-
The 50 year seismic risk assessment results obtained by using the tensity parameter. The record-to-record uncertainty index 𝛽𝐷 obtained
method proposed in this paper are 3∼9 times the probabilities deter- from 9 sets of randomly selected reticulated shell examples is in the
mined by only considering the uncertainties of record-to-record and range of 0.712∼0.911, and 0.9 is recommended as the record-to-record
structural design parameters. This shows that traditional seismic risk uncertainty value 𝛽𝐷 .
assessment method underestimates the seismic risk of reticulated shell (3) The yield strength, elastic modulus, damping ratio, dead load,
structures. In addition, because Model 2 had two weakened zones, the and snow load are considered uncertain factors of the structural design
50 year exceedance probability at each performance point is greater parameters. The range of 𝛽𝐶 obtained by 5 sets of randomly selected
than that of Model 1, i.e., approximately 1.2 to 1.6 times that of Model reticulated shell examples is between 0.1 and 0.5 at performance point
1. LS-1 and between 0.1 and 0.3 at performance points LS-2, LS-3, and LS-
4. Considering the universality and safety of the seismic risk assessment
6. Conclusions results, the recommended value of 𝛽𝐶 is 0.3.
(4) The uncertainties of performance indicators 𝛽𝑆 obtained by five
sets of randomly selected reticulated shell examples are 0.065∼0.266,
This paper proposes a two-parameter damage criterion suitable for
0.368∼0.545, and 0.216∼0.332 at performance points LS-1, LS-2, and
reticulated shell structures and establishes a multi-uncertainty evalu-
LS-3, respectively, and the corresponding recommended values are 0.2,
ation system for reticulated shell structures based on the traditional
0.5, and 0.3, respectively. When solving the uncertainties of perfor-
method. The uncertain factors include aleatoric uncertainties (record-
mance indicators, 30 sets of samples for each parameter are sufficient.
to-record, structural design parameters, and performance indicators)
(5) According to the literature involving the comparison of the tests
and epistemic uncertainties (modelling, construction). Then, through
and numerical simulations of reticulated shell structures, the modelling
multiple sets of randomly selected reticulated shell examples, the
uncertainty 𝛽𝑀 of the reticulated shell structure is 0.75. The modelling
ranges and recommended values for each uncertainty index are ob- uncertainty has the greatest impact on the 50 year exceedance prob-
tained. Finally, two reticulated shell shaking table test models are ability of the reticulated shell structures, and its probabilities are 3
used to carry out seismic risk assessment analysis considering multiple to 4.9 times the probabilities determined when only considering the
uncertainties and to study the influence of each uncertainty on the aleatoric uncertainties (record-to-record, structural design parameters,
50 year probability of exceedance. The following conclusions can be and performance indicators). The designers should strive to improve
drawn the modelling techniques.
(1) Considering the response characteristics of reticulated shells (6) By considering the node errors of construction, the concept of
under earthquakes, based on the Park-Ang damage model and 216 uncertainty in construction is proposed for the first time in the seismic
comprehensive examples of reticulated shells, a two-parameter damage risk assessment of reticulated shell structures. In the process of solving
criterion suitable for reticulated shells is proposed. The index values D construction uncertainty, it is sufficient to take 300 sets of samples for
at performance points LS-1, LS-2, LS-3, and LS-4 are 0.3, 0.6, 1.0, and each structure. The construction uncertainty 𝛽𝑄 obtained by three sets
∞, respectively. The criterion has good accuracy, takes into account the of randomly selected examples of reticulated shells is in the range of
contribution of displacement and energy dissipation to the damage of 0.059∼0.151, and it is recommended to take 0.1 as the index value 𝛽𝑄 .
the reticulated shells, and covers the failure forms of reticulated shell (7) After considering these five uncertainties, the 50 year ex-
structures under earthquake action. ceedance probabilities of reticulated shell structures are 3 to 9 times

11
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

Fig. 9. 50 year probabilities of exceedance of Model 1 and Model 2 with multiple uncertainties.

the probabilities determined only considering the uncertainties of References

record-to-record and structural design parameters. This shows that the
traditional method underestimates the potential risks of the reticulated [1] K. Kinali, B.R. Ellingwood, Seismic fragility assessment of steel frames for
consequence-based engineering: A case study for memphis, TN. Steel Constr. 29
shell structure in the life cycle, leading to unsafe results. (6) (2007) 1115–1127.
The research results of this paper could provide a design basis for [2] B.R. Ellingwood, K. Kinali, Quantifying and communicating uncertainty in
the performance-based seismic design of reticulated shell structures seismic risk assessment, Struct. Saf. 31 (2) (2009) 179–187.
[3] I. Iervolino, A. Spillatura, P. Bazzurro, Seismic reliability of code-conforming
and have the function of risk guidance and data basis. In addition,
Italian buildings, J. Earthq. Eng. 22 (sup2) (2018) 5–27.
this paper gives recommended values for each uncertainty index based [4] G.J. O’Reilly, R. Monteiro, A.M.B. Nafeh, T.J. Sullivan, G.M. Calvi, Displacement-
on the consideration of universality and safety. Designers can select based framework for simplified seismic loss assessment, J. Earthq. Eng. 24 (sup1)
appropriate parameter values within the range of each index value (2020) 1–22.
[5] L.Q. Jiang, Z. Hong, Y. Hu, Effects of various uncertainties on seismic risk of
according to their construction level and environment, or they can add steel frame equipped with steel panel wall, Bull. Earthq. Eng. 16 (12) (2018)
more test and numerical analysis data to expand or update each index 5995–6012.
in real time. [6] O.C. Celik, B.R. Ellingwood, Seismic risk assessment of gravity load designed
reinforced concrete frames subjected to mid-America ground motions, J. Struct.
Eng. 135 (4) (2009) 414–424.
CRediT authorship contribution statement [7] O.C. Celik, B.R. Ellingwood, Seismic fragilities for non-ductile reinforced concrete
frames-role of aleatoric and epistemic uncertainties, Struct. Saf. 32 (1) (2010)
1–12.
Wen Hua: Writing – original draft, Validation, Software, Method- [8] D. Shahnazaryan, G.J. O’Reilly, R. Monteiro, On the seismic loss estimation of
ology, Investigation, Formal analysis, Conceptualization. Jihong Ye: integrated performance-based designed buildings, Earthq. Eng. Struct. Dyn. 51
Writing – review & editing, Supervision, Resources, Funding acquisi- (8) (2022) 1794–1818.
[9] A. Orumiyehei, T.J. Sullivan, Displacement-based seismic assessment of the
tion, Data curation, Conceptualization.
likelihood of failure of reinforced concrete wall buildings, Buildings 11 (7) (2021)
1–25.
Declaration of competing interest [10] Y.J. Yin, Y. Li, Seismic collapse risk of light-frame wood construction considering
aleatoric and epistemic uncertainties, Struct. Saf. 32 (4) (2010) 250–261.
[11] L.Q. Jiang, J.H. Ye, Seismic risk assessment of a 2-storey steel-sheathed CFS
The authors declare that they have no known competing finan- building considering different sources of uncertainty, Structures 16 (2018)
cial interests or personal relationships that could have appeared to 347–357.
[12] L.Q. Jiang, J.H. Ye, Quantifying the effects of various uncertainties on seismic
influence the work reported in this paper. risk assessment of CFS structures, Bull. Earthq. Eng. 18 (1) (2020) 241–272.
[13] X.D. Zhi, G.B. Nie, F. Fan, S.Z. Shen, Vulnerability and risk assessment of single-
layer reticulated domes subjected to earthquakes, J. Struct. Eng. 138 (12) (2012)
Data availability
1505–1514.
[14] G.B. Nie, X.D. Zhi, F. Fan, J.W. Dai, Seismic performance evaluation of single-
The data that has been used is confidential. layer reticulated dome and its fragility analysis, J. Construct. Steel Res. 100
(2014) 176–182.
[15] G.B. Nie, C.X. Zhang, X.D. Zhi, Dai J.W., Damage quantification, damage
Acknowledgements limit state criteria and vulnerability analysis for single-layer reticulated shell,
Thin-Walled Struct. 120 (2017) 378–385.
[16] J. Zhong, X.D. Zhi, F. Fan, Sensitivity of seismic response and fragility to
This research is sponsored by the National Key R&D Program of parameter uncertainty of single-layer reticulated domes, Int. J. Steel Struct. 18
China (Gran No. 2017YFC1500702). (5) (2018) 1607–1616.

12
W. Hua and J. Ye Thin-Walled Structures 185 (2023) 110618

[17] J. Zhong, J.P. Zhang, X.D. Zhi, F. Fan, Probabilistic seismic demand and capacity [38] G.Q. Dai, L.C. Li, Statistics and analysis of new material properties parameters
models and fragility curves for reticulated structures under far-field ground of building structure steel, Build. Struct. 4 (2000) 31–32 (in Chinese).
motions, Thin-Walled Struct. 137 (2019) 436–447. [39] T. Vrouwenvelder, The JCSS probabilistic model code, Struct. Saf. (1997).
[18] Z.W. Yu, C. Lu, H.H. Ma, F. Kong, Probabilistic seismic vulnerability assessment [40] GB 50017-2017, Standard for Design of Steel Structures, China Construction
of aluminium alloy reticulated shells with consideration of uncertainty, Eng. Industry Press, Beijing, 2017 (in Chinese).
Struct. 195 (2019) 288–298. [41] Z. Cao, X.D. Xue, X.S. Wang, Y.C. Liu, Selection of earthquake waves and values
[19] Y.J. Park, H.S. Ang, Mechanistic seismic damage model for reinforced concrete, of damping ratio for space structures in seismic analysis, Spat. Struct. 14 (3)
J. Struct. Eng. 111 (4) (1985) 722–739. (2008) 3–8 (in Chinese).
[20] Y.H. Chai, K.M. Romstad, S.M. Bird, Energy-based linear damage model for [42] GB50009-2012, Load Code for the Design of Building Structures, China
high-intensity seismic loading, J. Struct. Eng. 121 (5) (1995) 857–864. Construction Industry Press, Beijing, 2012 (in Chinese).
[21] J.R. Jiang, J.J. Sun, Seismic failure model of brick structure, Earthq. Eng. Eng. [43] D. Wang, Seismic Fragility Analysis and Probabilistic Risk Analysis of Steel Frame
Vib. 7 (1) (1987) 20–34 (in Chinese). Structures, Harbin Institute of Technology, Harbin, 2006, pp. 18–19 (in Chinese).
[22] J.P. Ou, D.T. Niu, G.Y. Wang, Fuzzy dynamic reliability analysis and design of [44] M. Fragiadakis, D. Vamvatsikos, Fast performance uncertainty estimation via
multi-layer nonlinear seismic steel structure, Earthq. Eng. Eng. Vib. 10 (4) (1990) pushover and approximate IDA, Earthq. Eng. Struct. Dyn. 39 (6) (2010)
27–37. 683–703.
[23] D.T. Niu, L.J. Ren, Improved seismic failure model of reinforced concrete [45] M. Dolsek, Simplified method for seismic risk assessment of buildings with
structure with double parameters, Earthq. Eng. Eng. Vib. 4 (1996) 44–54 (in considering of aleatory and epistemic uncertainty, Struct. Infrastruct. Eng. 8 (10)
Chinese). (2012) 939–953.
[24] W.F. Du, B.Q. Gao, S.L. Dong, Study on types and characters of dynamical failure [46] J.F.J. Massey, The Kolmogorov–Smirnov test for goodness of fit, J. Amer. Statist.
for single layer latticed shells, Eng. Mech. 26 (7) (2009) 39–46 (in Chinese). Assoc. 46 (253) (1951) 68–78.
[25] K.A. Der, O. Ditlevsen, Aleatory or epistemic? Does it matter? Struct. Saf. 31 (2) [47] G.B. Nie, Failure Mechanism and Performance Evaluation of Reticulated Shells
(2009) 105–112. Under Severe Seismic Motion Based on Damage Cumulative Constitutive Model,
[26] V. Terzic, M.J. Schoettler, J.I. Restrepo, S.A. Mahin, Concrete column blind Harbin Institute of Technology, Harbin, 2012, pp. 106–107 (in Chinese).
prediction contest 2010: Outcomes and observations, PEER Rep. 2015 (01) [48] J.H. Ye, R. Pan, Shaking table test on collapse process of single-layer spherical
(2015) 1–114. shells, J. Build. Struct. 34 (4) (2013) 81–90 (in Chinese).
[27] GB 50011-2010, Code for Seismic Design of Buildings, China Construction [49] M. Zhang, Study on Seismic Behavior and Failure Criterion of Reticulated Shells
Industry Press, Beijing, 2016 (in Chinese). Based on Energy, Harbin Institute of Technology, Harbin, 2014, pp. 28–30 (in
[28] ATC-63, Quantification of Building Seismic Performance Factors, Applied Chinese).
Technology Council, Redwood City, 2008. [50] Y. Ding, Z.T. Chen, L. Zong, J.B. Yan, A theoretical strut model for severe
[29] M.F. Lu, Optimization of Domes Against Instability Considering Joints’ Me- seismic analysis of single-layer reticulated domes, J. Constr. Steel Res. 128 (2017)
chanical Performance, Southeast University, Nanjing, 2020, pp. 42–58, (in 661–671.
Chinese). [51] Y. Liu, S.D. Xue, G.X. Wang, H.Y. Li, Shaking table test of single-layer latticed
[30] JGJ 7-2010, Technical Specification for Space Frame Structures, China cylindrical shell considering soil–structure interaction and its numerical analysis,
Construction Industry Press, Beijing, 2010 (in Chinese). J. Central South Univ. Sci. Technol. 48 (1) (2017) 223–232 (in Chinese).
[31] W. Hua, J.H. Ye, Study on two-parameter criterion of reticulated shell structures [52] M. Zhang, C.L. Xie, X.D. Zhi, Shaking table tests for a small-scale single-layer
under earthquake action based on Park-Ang damage model, Eng. Mech., http: latticed cylindrical shell and its dynamic response probability model, J. Vib.
//dx.doi.org/10.6052/j.issn.1000-4750.2021.05.0375 (in Chinese). Shock 39 (15) (2020) 8–17 (in Chinese).
[32] D.G. Lv, X.H. Yu, Theoretical study of probabilistic seismic risk assessment based [53] Y. Wang, Q.B. Li, Study on the collapse resistant mechanism of K6 spherical
on analytical functions of seismic fragility, J. Build. Struct. 34 (10) (2013) reticulated shells under member removal conditions, Low Temp. Archit. Technol.
41–48 (in Chinese). 42 (10) (2020) 83–85 (in Chinese).
[33] L.L. Xu, J.H. Ye, DEM algorithm for progressive collapse simulation of single- [54] H.H. Ma, F. Fan, P. Wen, H. Zhang, S.Z. Shen, Experimental and numerical
layer reticulated domes under multi-support excitation, J. Earthq. Eng. (2017) studies on a single-layer cylindrical reticulated shell with semi-rigid joints,
1–28. Thin-Walled Struct. 86 (2015) 1–9.
[34] J.H. Ye, M.F. Lu, Optimization of domes against instability, Steel Compos. Struct. [55] W. Luo, W.D. Lu, X.Y. Sun, Y.N. Zhou, Overall stability test of semi-rigid single-
28 (4) (2018) 427–438. layer spherical reticulated shells, J. Nanjing Technol. Univ. Nat. Sci. Ed. 38 (5)
[35] J. Zhong, Probabilistic Seismic Fragility Analysis of Reticulated Shells, Harbin (2016) 81–86 (in Chinese).
Institute of Technology, Harbin, 2016, pp. 54–60 (in Chinese). [56] Z. Xiong, X.N. Guo, Y.F. Luo, S.J. Zhu, Y.P. Liu, Experimental and numerical
[36] G.Y. Gao, Research of Seismic Probabilistic Risk Assessment for Single-Layer studies on single-layer reticulated shells with aluminium alloy gusset joints,
Reticulated Domes, Harbin Institute of Technology, Harbin, 2012, pp. 11–27 (in Thin-Walled Struct. 118 (2017) 124–136.
Chinese). [57] S. He, Z.R. Jiang, J. Cai, Investigation on simulation methods of initial geo-
[37] D.G. Lv, X.P. Li, P. Zhang, G.Y. Wang, FEM reliability method for seismic metric imperfection distribution in elasto-plastic stability analysis of single-layer
vulnerability analysis of civil engineering structures, J. Basic Sci. Eng. 5 (2006) reticulated shells, KSCE J. Civ. Eng. 22 (4) (2018) 1193–1202.
264–271 (in Chinese).

13