Hindawi

Shock and Vibration

Volume 2018, Article ID 3958016, 11 pages
https://doi.org/10.1155/2018/3958016

Research Article
A Copula-Based and Monte Carlo Sampling Approach for
Structural Dynamics Model Updating with Interval Uncertainty

Xueqian Chen ,1,2 Zhanpeng Shen,1,2 and Xin’en Liu1,2

1
Institute of Systems Engineering, China Academy of Engineering Physics (CAEP), Mianyang Sichuan 621999, China
2
Shock and Vibration of Engineering Materials and Structures Key Laboratory of Sichuan Province, Mianyang Sichuan 621999, China

Correspondence should be addressed to Xueqian Chen; cxqdd127@sohu.com

Received 23 March 2018; Revised 29 May 2018; Accepted 4 June 2018; Published 9 July 2018

Academic Editor: Aly Mousaad Aly

Copyright © 2018 Xueqian Chen et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

As the uncertainty is widely existent in the engineering structure, it is necessary to study the finite element (FE) modeling and
updating in consideration of the uncertainty. A FE model updating approach in structural dynamics with interval uncertain
parameters is proposed in this work. Firstly, the mathematical relationship between the updating parameters and the output
interesting qualities is created based on the copula approach and the vast samples of inputs and outputs are obtained by the Monte
Carlo (MC) sampling technology according to the copula model. Secondly, the samples of updating parameters are rechosen by
combining the copula model and the experiment intervals of the interesting qualities. Next, 95% confidence intervals of updating
parameters are calculated by the nonparameter kernel density estimation (KDE) approach, which is regarded as the intervals of
updating parameters. Lastly, the proposed approach is validated in a two degree-of-freedom mass-spring system, simple plates, and
the transport mirror system. The updating results evidently demonstrate the feasibility and reliability of this approach.

1. Introduction In reality, there are always uncertainties in nominally identi-

cal structures, such as the structural parameter uncertainty
Finite element (FE) models that numerically solve various (physical material properties, geometric parameters), the
engineering problems can aid virtual prototyping, reduce assembly joints uncertainty, and the experiment uncertainty
product development cycle, and cut down the cost of per- (measurement noise, modal identification techniques, etc.).
forming the physical tests. However, the reliability of the sim- As a result, the FE model updating approaches with uncer-
ulation results by finite element modeling is not always guar- tainty have received great attentions recently. Studies have
anteed since FE models are the approximations of real world shown that the simulation results are more reliable when the
phenomena based on various assumptions. These assump- uncertainties are taken into account [7], suggesting that it is
tions may detract from the quality and accuracy of simulation necessary to consider the uncertainties during modeling and
results. In order to improve the accuracy of FE simulating simulating [8].
to serve the structural design better, the FE model updating
techniques are needed to develop. In the past few decades, In FE model updating approaches with uncertainty,
various kinds of FE model updating approaches have been the updated parameters are no longer deterministic and
widely investigated based on the actually observed behaviors are described as random variables. Usually, the FE model
of the system. Additionally, experimental modal and vibra- updating approaches with uncertainty can be classified into
tion data are often used in FE model updating in the field two major categories: probabilistic and nonprobabilistic
of structural dynamics [1–6]. approaches. In the earlier works, a probabilistic approach
In most model updating approaches, the simulations are proposed incorporated the measurement noise into model
usually deterministic where each of the updating parameters updating [9]. Subsequently, Bayesian statistical frameworks
is considered to have one “true” value and the purpose of the were adopted to estimate the posterior probabilities of
updating procedure is to provide a deterministic estimation. uncertain parameters [10–12]. However, high computational
2 Shock and Vibration

costs due to a large amount of samples required for a such as ill-condition, nonuniqueness and local optimal solu-
satisfactory estimation greatly restrain the applications of tion, etc. To overcome such inconvenience, an IMU approach
Bayesian updating approaches. As a result, surrogate models is developed in this work based on the copula model and
such as the Gaussian process model with the perturbation MC sampling. In the proposed approach, the copula model
approaches and sensitivity analysis approaches have been between the updating parameters and the interesting qualities
employed in stochastic model updating to improve the is constructed firstly. Then a large amount of samples is
efficiency [13–16]. Though, the surrogate model approaches obtained according to the copula model, and the samples
own the superiority of computational efficiency over Monte are rechosen based on experiment intervals of interesting
Carlo (MC) based methods. Nevertheless, the prerequisite of qualities. Lastly, the updating intervals of parameters are
small uncertainties, together with the Gaussian distribution obtained by estimating on the rechosen samples with kernel
assumption, also limits the applications to complex problems. density estimation (KDE). The remainder of the paper is
Moreover, perturbation based predictions are sensitive to the organized as follows. In Section 2, the copula-based FE model
initial estimates of parameters. Recently, an approach with updating approach and procedure with interval uncertainty
the response surface models and MC simulation has been are presented. In Section 3, three examples are provided
developed, which decomposed a stochastic updating process to validate the accuracy and reliability of the proposed
into a series of deterministic ones [17]. On the other hand, the approach. Conclusions are presented in Section 4.
accuracy of the probabilistic approaches depends on the esti-
mation of the probability distribution characteristics of the 2. Identification of Interval Parameters
structural parameters and the responses. The establishment
of an accurate probability distribution function (PDF) needs The FE model updating problems are classic inverse prob-
lots of experiment data in the probabilistic approaches, which lems in structural mechanics where the standard “forward”
greatly limits its application in engineering. relationship between input and output variables of a model is
In nonprobabilistic approaches, the interval approach has inverted. The key in solving a FE model updating problem
been intensively investigated. By comparison, the experiment is to construct the mathematical relationship between the
samples are not strictly needed in the FE model updating updating parameters and the output interesting qualities. The
with interval analysis as was proposed. In the field of copula function is one of the most effective mathematical
interval model updating (IMU), the inclusion theorem was tools to determine this relationship, which expediently char-
employed to establish an interval inverse problem. And the acterizes the correlation between the marginal functions of
convergence was achieved when measured responses fall multivariables and the joint distribution function.
into numerically predicted intervals [18–22]. Considering
the easy implementation, IMU problems are usually solved 2.1. Brief Introduction of the Copula Function. A copula
within a deterministic framework where the upper and lower function is a general function in statistics to formulate a
bounds of parameters are sought separately. For example, multivariate distribution with various statistical dependence
an IMU problem was decomposed into two deterministic patterns, which was presented by Sklar in 1959 [25]. Formally,
constrained optimization processes where the midpoints a copula is a joint distribution function of standard uniform
and interval radii of parameters were separately estimated random variables. According to the Sklar’s theorem, there
[19]. Alternatively, the vertex solution theorem is effective exists a two-dimensional copula C such that variables 𝑥1 and
and cost-efficient for IMU due to its easy implementation 𝑥2 in a real random space.
[20], particularly in the solution of Eigen value problems
[21]. But the vertex solution was valid only for particular 𝐹 (𝑥1 , 𝑥2 ) = 𝐶 (𝐹1 (𝑥1 ) , 𝐹2 (𝑥2 )) (1)
parameterization of an FE model without the involvement
of eigenvectors, which highly limits its further applications. where 𝐹(𝑥1 , 𝑥2 ) is a two-dimensional distribution
Due to this drawback, global optimization algorithms were function with marginal functions 𝐹1 (𝑥1 ) and 𝐹2 (𝑥2 ) and
taken into account for more general solutions. Surrogate 𝐶(𝐹1 (𝑥1 ), 𝐹2 (𝑥2 )) is the copula cumulative distribution
models such as the Kriging predictor and interval response function (CDF).
surface were used to improve the efficiency of gradient Equation (1) can be spread for m-dimensional variables
computation and facilitate the convergence [23, 24]. So far easily, that is,
most of IMU problems are solved within a deterministic 𝐹 (𝑥1 , 𝑥2 , . . . , 𝑥𝑚 ) = 𝐶 (𝐹1 (𝑥1 ) , 𝐹2 (𝑥2 ) , . . . , 𝐹𝑚 (𝑥𝑚 )) (2)
framework since direct interval arithmetic operations are
difficult to implement during inverse solutions. Therefore the Consequently, the m-dimensional PDF is as follows:
upper and lower bounds of parameters should be sought
separately through a deterministic inverse procedure. Addi- 𝜕𝑚 𝐹 (𝑥1 , 𝑥2 , . . . , 𝑥𝑚 )
𝑓 (𝑥1 , 𝑥2 , . . . , 𝑥𝑚 ) =
tionally, global optimization of interval variables is difficult 𝜕𝑥1 𝜕𝑥2 . . . 𝜕𝑥𝑚
to realize due to the fact that the interval arithmetic is quite
different with the traditional mathematical arithmetic. 𝜕𝑚 𝐶 (𝐹1 (𝑥1 ) , 𝐹2 (𝑥2 ) , . . . , 𝐹𝑚 (𝑥𝑚 ))
= (3)
Though several probabilistic and interval model updating 𝜕𝑥1 𝜕𝑥2 . . . 𝜕𝑥𝑚
approaches have been developed in the past years, most of 𝑚
them are still complicated for implementation. Additionally, = 𝑐 (𝐹1 (𝑥1 ) , 𝐹2 (𝑥2 ) , . . . , 𝐹𝑚 (𝑥𝑚 )) × ∏𝑓𝑖 (𝑥𝑖 )
these approaches with uncertainty suffer from the challenges 𝑖=1
Shock and Vibration 3

where 𝑐(𝐹1 (𝑥1 ), 𝐹2 (𝑥2 ), . . . , 𝐹𝑚 (𝑥𝑚 )) is the copula PDF, Following [28], the kernel density estimator for variable x
𝑓(𝑥1 , 𝑥2 , . . . , 𝑥𝑚 ) is the united PDF for m-dimensional ran- has the form
dom variables, and 𝑓𝑖 (𝑥𝑖 ) is the PDF of the 𝑖th random
1 𝑛 𝑥 − 𝑋𝑖
variable. 𝑓̂ℎ = ∑𝐾 ( ) (6)
𝑛ℎ 𝑖=1 ℎ
At present, the general copula function types include
the Gaussian copula function, t-copula function, and where 𝑛 is the number of observations used to construct
Archimedean copula function [26, 27]. Among them, the the estimate, 𝐾(⋅) is a kernel function, 𝑋𝑖 is the 𝑖th obser-
Gaussian copula is widely utilized because most of the vation, and ℎ is the window width, or bandwidth. A typical
parameters in the engineering satisfy the normal distribution. choice for the kernel 𝐾(⋅) is the standard normal density and
In the study, the Gaussian copula is adopted for the FE model. is implemented here. The choice of the window width ℎ is
Specifically, the Gaussian copula function is constructed usually based on the optimization of some scoring function.
by multidimensional Gaussian distribution and the linear A least-square cross-validation score function is adopted for
correlation parameters, and its distribution function is this work [28].
as follows: The empirical CDF and 95% confidence interval (CI)
[𝑦𝑒− 95 𝑦𝑒+ 95 ]of the random variable 𝑥 can be obtained by KDE
𝐶𝐺𝑎 (𝑢1 , 𝑢2 , . . . , 𝑢𝑚 ; 𝜌) in Matlab that is regarded as the interval of the random
(4) variable in this work.
= Φ𝜌 (Φ−1 (𝑢1 ) , Φ−1 (𝑢2 ) , . . . , Φ−1 (𝑢𝑚 )) Considering the fact that the estimation on the original
intervals of the updating parameters may be inaccurate, the
where Φ𝜌 is the distribution function of the standard reliable intervals are not identified through one copula-based
normal function for d-dimensional with the correlation FE model updating procedure. In order to overcome this
matrix 𝜌, Φ−1 is the inverse function of the distribution problem, the idea of the adaptive response surface technique
function of the standard normal function, and 𝑢𝑖 = 𝐹𝑖 (𝑥𝑖 ), is adopted for this work [29]. That is, in order to get the
𝑖 = 1, 2, . . . , 𝑚. final updating results, multiloop on the copula-based model
updating procedures may be performed.
2.2. Copula-Based Approach for Model Updating with Interval In the FE model updating procedure, the convergent
Uncertainty. Firstly, the original design spaces of updating criterion is that the difference of the updating parameter
parameters 𝑥𝑗 , 𝑗 = 1, 2, ⋅ ⋅ ⋅ , 𝑚, are assumed, and a few sam- intervals between the (𝑖 + 1)th iteration step and the 𝑖th
ples are obtained by design of experiment (DOE) approach iteration step is less than a small value, or the difference of the
and subsequent deterministic FE analysis on samples accord- output interesting quality intervals between the 𝑖th iteration
ing to DOE. Then, the samples of the output interesting step results and the experiment results is less than a critical
qualities 𝑦𝑘 , 𝑘 = 1, 2, ⋅ ⋅ ⋅ , 𝑛, are obtained from the FE analysis value.
results. Secondly, the copula model is constructed according In order to improve the efficiency and the validity of
to the samples of updating parameters and output interesting model updating, the renewal strategy of updating parameters
qualities, and resampling is performed to get large samples is as follows in each iteration step. The current intervals of
with number N for updating parameters and interesting updating parameters are updated according to the results of
response qualities based on the copula model. The samples the previous iteration step, and the interval medians of the
falling into the experiment data space are considered to char- previous step are regarded as the current interval medians,
acterize the input-output relationship of the physical struc- and about 80% of the interval width of the previous step
ture believably, and unuseful samples are needed to remove. is regarded as the current interval width. Also, the Latin
Next, the samples of updating parameters 𝑥𝑗 are rechosen Hypercube Sample (LHS) method is suggested in the DOE,
according to the experiment intervals of interesting qualities, and the number of samples is not less than ten.
as follows: The copula-based model updating procedure is repeated
until the convergent criterion is satisfied. The flow chart for
𝑥𝑗 = {𝑥𝑗 = 𝑓 (𝑦1 , ⋅ ⋅ ⋅ , 𝑦𝑛 ) | 𝑦𝑘 ∈ 𝑅, 𝑦𝑘− ≤ 𝑦𝑘 ≤ 𝑦𝑘+ } , the copula-based FE model updating is outlined in Figure 1.
(5)
𝑗 = 1, 2, ⋅ ⋅ ⋅ , 𝑚, 𝑘 = 1, 2, ⋅ ⋅ ⋅ , 𝑛
3. Case Studies
where 𝑅 is the real space and 𝑦𝑘+ and 𝑦𝑘−
are the upper and 3.1. Example 1: A Two-Degree-of-Freedom Mass-Spring System.
lower bounds of the 𝑘th output interesting quality which can A two degree-of-freedom mass-spring system is shown in
be obtained from the experiment results. Figure 2. The deterministic parameters in the system are 𝑚1 =
In practical model updating, the measured data are only 𝑚2 = 1 kg and 𝑘1 = 1 N/m. The uncertain interval parameters
a few samples in general. Reasonable interval estimation on are 𝑘2 = [0.8 1.2] N/m and 𝑘3 = [0.9 1.1] N/m.
experiment data is the precondition to obtain the reliable For simplicity, it is assumed that the uncertain param-
updated FE model. However, the KDE allows for the capture eters are uniformly distributed. To create such kind of
of the observed distributional structure for the random uncertainty, the LHS method is used to generate twenty
variables, without having to assume a particular parametric experiment samples. Afterwards, the experiment results of
distribution form. the first two natural frequencies are obtained according
4 Shock and Vibration

Table 1: Statistical properties of the measured frequencies of the plates.

f 1 (Hz) f 2 (Hz) f 3 (Hz) f 4 (Hz) f 5 (Hz)

mean 24.12 66.92 77.65 131.97 158.80
interval [23.94 24.4] [66.5 67.53] [76.24 78.91] [131.31 133.03] [156.31 160.94]
Std. 0.11 0.25 0.57 0.42 0.97

Build FE model and specify Assume that the original intervals of 𝑘2 and 𝑘3 are the
updating parameters same as [1.2 2.2] N/m, and the first two natural frequencies
are regarded as the output interesting qualities. The intervals
Specify the original Intervals of 𝑘2 and 𝑘3 are identified according to the copula-based
of updating parameters model updating flow with interval uncertainty in Figure 1.
Renew the intervals of
updating parameters
The updating results are convergent after three iteration steps.
Figure 3 is the scatter map between the updating param-
DOE and obtain the samples eters and the output interesting qualities when the updating
results are convergent, which shows that there is strong cor-
relation between 𝑘3 and 𝑓1 and between 𝑘2 and 𝑓2 and weak
Calculate responses of FE correlation between 𝑘2 and 𝑓1 and between 𝑘3 and 𝑓2 .
model at DOE points
Then, the updated intervals of 𝑘2 and 𝑘3 are ob-
tained by KDE for the resample in Figure 3, which are
Obtain SRQs yk [0.775 1.230] N/m and [0.882 1.133] N/m, respectively. The
comparison between the original uncertain interval and
the updated interval of updating parameters is shown in
Construct the copula model between
Figure 4, which shows that the updated interval matches
the updating parameters and the
SRQs, and resample the real interval better. Because the effect of small samples
on experiment data is considered in the model updating
procedure, the updated interval is bigger than the real interval
Re-choose the samples of updating and the result is reasonable.
parameters according to Eq.(5)
In order to validate the updating results on the interval
uncertain parameters, the copula models are reconstructed
Estimate the intervals of updating according to the original and the updated intervals of 𝑘2
parameters by KDE and 𝑘3 . As a result, 5000 samples of 𝑘2 , 𝑘3 , and the first
two frequencies of the system are resampled by the copula
models constructed just now. The scatter map between the
simulating and the experimental results is shown in Figure 5,
No Satisfy the convergent which indicates that the frequencies of the updated model are
criterion ? agreement with the experimental results better.

Yes 3.2. Example 2: Interval Model Updating in Simple Plates.

End
Impact hammer modal testing with free-free boundary
conditions was conducted on thirty-three nominally iden-
Figure 1: Flow chart of the copula-based FE model updating.
tical steel plates in [30]. The nominal geometric dimen-
sions of the plates are 564 mm(length) × 110 mm(width) ×
1.45 mm(thickness). And the nominal material properties are
Young’s modulus of 210 GPa, the shear modulus of 83 GPa,
k1 k2 k3 and the mass density of 7860 kg/m3 . The statistical properties
m1 m2
of the first five measured natural frequencies of the plates are
given in Table 1.
The FE model of the plate is created by the SHELL181
Figure 2: A two degree-of-freedom mass-spring system.
element in ANSYS, which has 300 shell elements. The
boundary is free-free in the model. The material properties of
the original FE model are the nominal values. The FE model
to the samples. Considering the effect of small sample, and the first five mode shapes of simulation are shown in
the 95% CI of experiment results are estimated by KDE Figure 6.
and the estimated results are [0.9646 1.0338] rad.s−1 and After investigation, the uncertainty of rectangular plates
[1.5789 1.8875] rad.s−1 , respectively, which are regarded as can be characterized by Young’s modulus 𝐸, the shear modu-
the intervals of experimental results. lus 𝐺, and the thick 𝑇 in the FE model. The model updating
Shock and Vibration 5

1.25 1.25

1.2 1.2

1.15 1.15

1.1 1.1
＠1 (rad/s)

＠1 (rad/s)
1.05 1.05

1 1

0.95 0.95

0.9 0.9

0.85 0.85
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Ｅ2 (N/m) Ｅ3 (N/m)

Resample data Resample data

Experiment data Experiment data
(a) 𝑘2 VS 𝑓1 (b) 𝑘3 VS 𝑓1
2 2

1.9 1.9

1.8 1.8
＠2 (rad/s)

＠2 (rad/s)

1.7 1.7

1.6 1.6

1.5 1.5

1.4 1.4
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Ｅ2 (N/m) Ｅ3 (N/m)

Resample data Resample data

Experiment data Experiment data
(c) 𝑘2 VS 𝑓2 (d) 𝑘3 VS 𝑓2

Figure 3: Scatter map between the updating parameters and the output quantities of mass-spring system.

is to identify the interval of these three uncertain parameters the updated interval of the three parameters is shown in
𝐸, 𝐺, and 𝑇 by the experiment results. Figure 7, which shows that the updated intervals of these
Firstly, the intervals of the first five modal frequencies are three parameters are much less than their original uncertain
estimated by KDE according to the experiment data in Ref. interval.
[30]. Secondly, it is assumed that the original intervals of E, G, In order to validate the updating results on the interval
and T are [195 220] GPa, [78 87] GPa, and [1.30 1.60] mm, uncertain parameters, the copula models are reconstructed
respectively, and intervals of E, G, and T are identified accord- on the original and the updated intervals of 𝐸, 𝐺, and 𝑇. The
ing to the copula-based model updating flow with interval 5000 samples of 𝐸, 𝐺, 𝑇, and the first five natural frequencies
uncertainty in Figure 1. There are 20 experimental design of plates are obtained by MC sampling. The scatter plots
data in each iteration step, i.e., 20 determined FE simulations for the simulation and the experimental results the first
in each iteration step. The model updating of the plate is five natural frequencies of plates are shown in Figure 8.
convergent after four iteration steps with the convergent The comparisons on the natural frequencies between the
indices 𝜀1 = 0.002 and 𝜀2 = 0.002. simulation and the experimental results are listed in Table 2.
After updating, the estimated intervals of the It can be seen from Table 2 and Figure 8 that the frequencies
three parameters were 𝐸 = [203.21 205.98] GPa, 𝐺 = of the updated model are in better agreement with the
[82.75 85.20] GPa, and 𝑇 = [1.44 1.46] mm, respectively. experimental results and the mean errors of frequencies
The comparison between the original uncertain interval and decrease from the initial [12.51 12.62]% to [0.72 0.24]%.
6 Shock and Vibration

Table 2: Comparison on the natural frequencies of plates between the simulation and the experimental results.

Mode Experimental interval/Hz Original interval/Hz Error /% Updated interval/Hz Error /%

1 [23.94 24.4] [21.06 27.55] [-12.05 12.92] [23.71 24.44] [-0.95 0.15]
2 [66.5 67.53] [58.61 76.53] [-11.86 13.32] [66.84 67.83] [-0.99 0.44]
3 [76.24 78.91] [65.81 87.80] [-13.68 11.26] [75.93 78.67] [-0.41 -0.30]
4 [131.31 133.03] [116.34 151.62] [-11.40 13.98] [130.37 134.26] [-0.71 0.92]
5 [156.31 160.94] [135.12 179.67] [-11.59 8.55] [155.51 160.95] [-0.51 0.01]
mean [12.51 12.62] [0.72 0.24]

2.4 2.5

2.2

1.8

＠2 (rad.Ｍ-1 )
k3 (N/m)

1.6 2

1.4

1.2

0.8 1.5
0.9 1 1.1 1.2 1.3
0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4
＠1 (rad.Ｍ-1 )
k2 (N/m)

Original interval Original results

Updated results
Updated interval
Test results
True interval
Figure 5: Scatter plots for the first two natural frequencies of mass-
Figure 4: Patch plots for the updating parameters of mass-spring
spring system.
system.

technology fluctuation about the support frame, which

Thus, this experimental validation has also confirmed the
induces the uncertainty of the vibration characteristics. In
feasibility of the proposed IMU approach in estimating
order to study how these uncertain factors affect the natural
interval parameters of steel plates.
frequencies of the transport mirror system, impact hammer
modal testing with fixed boundary conditions was conducted
3.3. Example 3: Interval Model Updating in Transport Mirror on ten nominally identical transport mirror systems. The
System. The ShenGuangIII (SGIII) facility is designed for experiment mount of the transport mirror system is shown
inertial confinement fusion (ICF) high energy experiments in Figure 11. Three natural frequencies in the interesting
with 48 laser beams exactly transported and oriented to frequency range were found to significantly influence the
target. Figure 9 is the view of the beam transport system response of the transport mirror system under the work
in SGIII facility target area. There are 276 transport mirror condition. Then, the first three natural frequencies of the
systems in the facility, and the dynamic response under transport mirror system should be updated before calculating
ambient vibration is a key factor to affect the stability of the response. The corresponding natural frequencies were
the SGIII facility [31]. A classic transport mirror system is obtained by the modal experiments, as listed in Table 3. It is
shown in Figure 10, which consists of mirror component and observed that the frequency variations become more obvious
support frame. with the increase of the mode order.
The material of the support frame is steel with nominal Generally, the bolts are ignored in the structural dynamic
Young’s modulus 200 Gpa, nominal Poisson’s ratio 0.3, and analysis model generally, and the FE model of the transport
the nominal density 7850 kg/m3 . The material of the mirror mirror system is established by the SOLID185 and SHELL181
is K9 class with nominal Young’s modulus 80 Gpa, nominal element in ANSYS, which is shown in Figure 12. A fully
Poisson’s ratio 0.21, and the nominal density 2510 kg/m3 . The fixed bottom of the support frame is used as the boundary
transport mirror system is about 1.32 m in height. condition. The material properties of the original FE model
The uncertainty exists in the transport mirror system are the nominal values. The first three simulation mode
because of the installation fluctuation and the welding shapes are shown in Figure 13.
Shock and Vibration 7

Y MX
X MX
MN Y
Z
Y X
Z
X Z

MN

(a) (b) (c)

MX
MX Y Y
MN
X MX
X Y
Z
Z X
Z
MN

MN

(d) (e) (f)

Figure 6: The FE model and mode shapes of steel plates: (a) FE model; (b) mode 1; (c) mode 2; (d) mode 3; (e) mode 4; (f) mode 5.

95
1.7

90
1.6
G (109 Pa)

85
T (mm)

1.5

80 1.4

1.3
75

1.2
180 190 200 210 220 230 240 180 190 200 210 220 230 240
E (109 Pa) E (109 Pa)

Original interval Original interval

Updated interval Updated interval
(a) 𝐸 vs 𝐺 (b) 𝐸 vs 𝑇

Figure 7: Patch plots for the updating parameters of steel plates.

100 200
90 180
＠3 (Hz)

＠5 (Hz)

80 160
70 140
60 120
90 180
80 30 160 100
70 140 90
60 25 80
120 70
50 20 ＠1 (Hz) ＠4 (Hz) ＠3 (Hz)
＠2 (Hz) 100 60

Original results Original results

Updated results Updated results
Experiment results Experiment results
(a) f1 vs f2 vs f3 (b) f3 vs f4 vs f5

Figure 8: Scatter plots for the first five natural frequencies of steel plates.
8 Shock and Vibration

Table 3: Experiment results of the first three natural frequencies of the transport mirror system.

Case 1 2 3 4 5 6 7 8 9 10
f 1 /Hz 21.573 21.642 21.675 21.472 21.583 21.683 21.657 21.797 21.644 21.462
f 2 /Hz 22.213 22.223 22.31 22.087 22.168 22.293 22.283 22.407 22.232 22.121
f 3 /Hz 41.328 41.432 41.58 41.203 41.416 41.56 41.557 41.807 41.497 41.237

Figure 9: A view of the beam transport system in SGIII facility

target area, which is reproduced from Chen X J et al. (2014).

Figure 11: Experiment setup of the transport mirror system.

Mirror
component

Support
frame

Figure 10: Sketch map of the transport mirror system.

Figure 12: FE model of the transport mirror system.

Theoretical analysis finds that the installation uncertainty

of the transport mirror system can be characterized by model updating flow with interval uncertainty in Figure 1.
Young’s modulus 𝐸1 of the bottom part of the support There are 12 experimental design data in each iteration step,
frame, and the weld uncertainty of the support frame can be i.e., 12 determined FE simulations in each iteration step. The
characterized by Young’s modulus 𝐸2 of the upper part of the model updating of the transport mirror system is convergent
support frame. The goal of FE model updating is to identify after four iteration steps with the convergent indices 𝜀1 =
the interval of 𝐸1 and 𝐸2 . 0.005 and 𝜀2 = 0.005.
Firstly, the intervals of the first three modal frequencies After updating, the estimated intervals of the three
are estimated by KDE according to the experiment data in parameters were E1 = [104.46 114.65] GPa and E2 =
Table 3, which are [21.41 21.83] Hz, [22.03 22.45] Hz, and [104.46 114.65] GPa, respectively. The comparison between
[41.09 41.88] Hz, respectively. the original uncertain interval and the updated interval of the
Secondly, assuming that the original intervals of E1 and E2 updating parameters is shown in Figure 14, indicating that the
are [100 150] GPa and [160 200] GPa, respectively, the inter- updated intervals of the updating parameters are much less
vals of E1 and E2 are identified according to the copula-based than their original uncertain interval.
Shock and Vibration 9

Table 4: Comparison on the natural frequencies of the transport mirror system between the simulation and the experimental results.

Mode Experimental interval/Hz Original interval/Hz Error /% Updated interval/Hz Error /%

1 [21.46 21.80] [22.06 24.24] [2.77 11.20] [21.38 22.91] [-0.36 0.52]
2 [22.09 22.41] [22.68 24.94] [2.69 11.31] [22.00 22.54] [-0.38 0.59]
3 [41.20 41.81] [42.03 45.98] [2.01 9.97] [40.92 41.85] [-0.68 0.09]
mean [2.49 10.83] [0.47 0.40]

(a) Mode 1 (b) Mode 2 (c) Mode 3

Figure 13: Mode shapes of the transport mirror system.

2
46
1.9
45
1.8
44
％2 (1011 Pa)

＠3 (Hz)

1.7 43

1.6 42
41
1.5
40
1.4 25
24 25
24
1.3 23 23
＠2 (Hz) 22
22 21 ＠1 (Hz)
0.9 1 1.1 1.2 1.3
％1 (1011 Pa) Original results
Updated results
Original interval Experiment results
Updated interval
Figure 15: Scatter plots for the first three natural frequencies of
Figure 14: Patch plots for the updating parameters of transport transport mirror system.
mirror system.

the transport mirror system are shown in Figure 15. The com-
In order to validate the updating results on the interval parisons on the natural frequencies between the simulation
uncertain parameters, the copula models are reconstructed and the experimental results are listed in Table 4. It can be
on the original and the updated intervals of E1 and E2 , seen from Table 4 and Figure 15 that the frequencies of the
Next, the 5000 samples of E1 , E2 , and the first three natural updated model agree with the experimental results better
frequencies of the transport mirror system are obtained by and the mean errors of frequencies decrease from the initial
MC sampling. The scatter plots for the simulation and the [2.49 10.83]% to [0.47 0.40]%. As a result, this experimental
experimental results of the first three natural frequencies of validation has also proved the feasibility of the proposed
10 Shock and Vibration

IMU approach in estimating the interval parameters of the Conflicts of Interest

transport mirror system.
The authors declare that they have no conflicts of interest.
4. Conclusions
Acknowledgments
In summary, a copula-based and MC sampling FE model
updating approach in the structural dynamics with interval This work was supported by Science Challenge Project
uncertainty is presented. Because the proposed FE model (Grant no. TZ2018007), the National Natural Science Foun-
updating approach is based on the copula model of input- dation of China (Grant no. 11472256), and the National Key
output parameters and MC sampling, the optimization dif- Research and Development Program of China (Grant no.
ficulty caused by the interval algorithm is avoided in the 2016YFB0201005)
proposed FE model updating approach with interval uncer-
tainty, and the questions of ill-condition, nonuniqueness, Supplementary Materials
and local optimal solution can be avoided also. Unlike other
metal-based model updating approaches, the copula-based Supplementary Table 1. Statistical properties of the measured
model updating approach can deal with the nonlinear rela- frequencies of the plates in example 1. Supplementary Table 2.
tivity between the multi-input qualities and the multioutput Statistical properties of the measured frequencies of the plates
qualities even if the multioutput qualities are correlated. In in example 2. (Supplementary Materials)
order to get a better copula model of input-output parameters
for identifying the interval parameters, the LHS approach References
is suggested in the DOE and the number of samples is not
less than ten in the proposed FE model updating approach. [1] J. E. Mottershead and M. I. Friswell, “Model updating in
Besides, the proposed approach also considers the effect structural dynamics: A survey,” Journal of Sound and Vibration,
of small samples and estimates the interval of experiment vol. 167, no. 2, pp. 347–375, 1993.
samples by KDE in advance, which makes the updated model [2] C.-P. Fritzen, D. Jennewein, and T. Kiefer, “Damage detection
more reliable for the predictions on structural dynamic based on model updating methods,” Mechanical Systems and
responses. Three examples were used to validate the FE model Signal Processing, vol. 12, no. 1, pp. 163–185, 1998.
updating approach proposed in the study. The updating [3] Y. Ren and C. F. Beards, “Identification of ‘effective’ linear joints
results have proved the feasibility and reliability of this using coupling and joint identification techniques,” Journal of
approach. Vibration and Acoustics, vol. 120, no. 2, pp. 331–338, 1998.
The total time consumption of the FE model updating is [4] W. L. Li, “A new method for structural model updating and
mainly the time of the FE model simulation. For example, joint stiffness identification,” Mechanical Systems and Signal
the FE model updating of the transport mirror system was Processing, vol. 16, no. 1, pp. 155–167, 2002.
completed after four iteration steps, and there were 12 FE [5] P. G. Bakir, E. Reynders, and G. De Roeck, “Sensitivity-based
model simulations in each iteration step, so only 48 deter- finite element model updating using constrained optimization
ministic FE simulations were carried out in the whole model with a trust region algorithm,” Journal of Sound and Vibration,
vol. 305, no. 1-2, pp. 211–225, 2007.
updating with interval uncertainty. But, more than hundreds
of FE simulations are needed in a determined model updat- [6] S.-S. Jin and H.-J. Jung, “Sequential surrogate modeling for effi-
cient finite element model updating,” Computers & Structures,
ing without the surrogate model, and the number of FE
vol. 168, pp. 30–45, 2016.
simulations in the model updating with uncertainty is far
[7] W. L. Oberkampf and C. J. Roy, Verification and Validation in
more than that of a deterministic model updating. Then, the
Scientific Computing, Cambridge University Press, Cambridge,
model updating method proposed with interval uncertainty UK, 2010.
is efficient in the paper. Besides, after updating the FE model,
[8] E. Simoen, G. De Roeck, and G. Lombaert, “Dealing with uncer-
the difference of the first few natural frequencies between
tainty in model updating for damage assessment: a review,”
the calculated interval and the experimental interval is less Mechanical Systems and Signal Processing, vol. 56-57, pp. 123–
than 1%. In conclusion, the copula-based and MC sampling 149, 2015.
concept expands the application range of IMU and is useful [9] M. I. Friswell, “The adjustment of structural parameters using
for fast and accurate estimation of interval parameters in the a minimum variance estimator,” Mechanical Systems and Signal
FE model. Processing, vol. 3, no. 2, pp. 143–155, 1989.
[10] J. L. Beck and L. S. Katafygiotis, “Updating models and their
Data Availability uncertainties. I: Bayesian statistical framework,” Journal of
Engineering Mechanics, vol. 124, no. 4, pp. 455–461, 1998.
The data for Example 1 are included in the Supplementary [11] J. L. Beck and S.-K. Au, “Bayesian updating of structural models
Material file associated with this manuscript. The data for and reliability using Markov chain Monte Carlo simulation,”
Example 2 are reported in [30] but are also included in Journal of Engineering Mechanics, vol. 128, no. 4, pp. 380–391,
the Supplementary Information file for completeness. The 2002.
data for Example 3 are included within the article. Other [12] M. C. Kennedy and A. O’Hagan, “Bayesian calibration of
data are available from the corresponding author upon computer models,” Journal of the Royal Statistical Society: Series
request. B (Statistical Methodology), vol. 63, no. 3, pp. 425–464, 2001.
Shock and Vibration 11

[13] H. P. Wan and W. X. Ren, “Stochastic model updating utilizing of simple and complicated structures,” in Proceedings of the
Bayesian approach and Gaussian process model,” Mechanical 10th International Conference on Recent Advances in Structural
Systems and Signal Processing, vol. 70-71, pp. 245–268, 2016. Dynamics, University of Southampton, 2010.
[14] H. H. Khodaparast, J. E. Mottershead, and M. I. Friswell, “Per- [31] X. J. Chen, M. C. Wang, W. K. Wu, and X. H. Que, “Structural
turbation methods for the estimation of parameter variability design of beam transport system in SGIII facility target area,”
in stochastic model updating,” Mechanical Systems and Signal Fusion Engineering and Design, vol. 89, no. 12, pp. 3095–3100,
Processing, vol. 22, no. 8, pp. 1751–1773, 2008. 2014.
[15] N. Abu Husain, H. Haddad Khodaparast, and H. Ouyang,
“Parameter selection and stochastic model updating using
perturbation methods with parameter weighting matrix assign-
ment,” Mechanical Systems and Signal Processing, vol. 32, pp.
135–152, 2012.
[16] X. G. Hua, Y. Q. Ni, Z. Q. Chen, and J. M. Ko, “An improved per-
turbation method for stochastic finite element model updating,”
International Journal for Numerical Methods in Engineering, vol.
73, no. 13, pp. 1845–1864, 2008.
[17] S. E. Fang, W. X. Ren, and R. Perera, “A stochastic model
updating method for parameter variability quantification based
on response surface models and Monte Carlo simulation,”
Mechanical Systems and Signal Processing, vol. 33, pp. 83–96,
2012.
[18] S. S. Rao and L. Berke, “Analysis of uncertain structural systems
using interval analysis,” AIAA Journal, vol. 35, no. 4, pp. 727–
735, 1997.
[19] S. L. Li, H. Li, and J. P. Ou, “Model updating for uncertain
structures with interval parameters,” in Proceedings of the Asia-
Pacific Workshop on Structural Health Monitoring, Yokohama,
Japan, 2006.
[20] C. Jiang, X. Han, and G. R. Liu, “Optimization of structures with
uncertain constraints based on convex model and satisfaction
degree of interval,” Computer Methods Applied Mechanics and
Engineering, vol. 196, no. 49–52, pp. 4791–4800, 2007.
[21] Z. P. Qiu, X. J. Wang, and M. I. Friswell, “Eigenvalue bounds of
structures with uncertain-but-bounded parameters,” Journal of
Sound and Vibration, vol. 282, no. 1-2, pp. 297–312, 2005.
[22] S. Gabriele and C. Valente, “An interval-based technique for FE
model updating,” International Journal of Reliability and Safety,
vol. 3, no. 1–3, pp. 79–103, 2009.
[23] H. H. Khodaparast, J. E. Mottershead, and K. J. Badcock,
“Interval model updating with irreducible uncertainty using the
Kriging predictor,” Mechanical Systems and Signal Processing,
vol. 25, no. 4, pp. 1204–1226, 2011.
[24] S. E. Fang, Q. H. Zhang, and W.-X. Ren, “An interval model
updating strategy using interval response surface models,”
Mechanical Systems and Signal Processing, vol. 60, pp. 909–927,
2015.
[25] M. Sklar, “Fonctions de répartition à n dimensions et leurs
marges,” Publications de l’Institut de Statistique de l’Université de
Paris, vol. 8, pp. 229–231, 1959.
[26] J.-D. Fermanian, “Goodness-of-fit tests for copulas,” Journal of
Multivariate Analysis, vol. 95, no. 1, pp. 119–152, 2005.
[27] C. Genest and L.-P. Rivest, “Statistical inference procedures
for bivariate Archimedean copulas,” Journal of the American
Statistical Association, vol. 88, no. 423, pp. 1034–1043, 1993.
[28] B. W. Silverman, Density Estimation for Statistics and Data
Analysis, Chapman & Hall, London, UK, 1986.
[29] G. G. Wang, “Adaptive response surface method using inherited
Latin hypercube design points,” Journal of Mechanical Design,
vol. 125, no. 2, pp. 210–220, 2003.
[30] N. A. Husain, H. H. Khodaparast, and J. H. Ouyang, “Param-
eter selections for stochastic uncertainty in dynamic models
 International Journal of

Rotating Advances in
Machinery Multimedia

The Scientific
Engineering
Journal of
Journal of

Hindawi
World Journal
Hindawi Publishing Corporation Hindawi
Sensors
Hindawi Hindawi
www.hindawi.com Volume 2018 http://www.hindawi.com
www.hindawi.com Volume 2018
2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Submit your manuscripts at

www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi
Engineering
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Volume 2018
Navigation and
Observation
Hindawi
www.hindawi.com Volume 2018
in Engineering
Hindawi
www.hindawi.com Volume 2018
Engineering
Hindawi
www.hindawi.com Volume 2018
Hindawi
www.hindawi.com www.hindawi.com Volume 2018

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Hindawi Hindawi Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018