Professional Documents
Culture Documents
1 s2.0 S1270963821001449 Main
1 s2.0 S1270963821001449 Main
a r t i c l e i n f o a b s t r a c t
Article history: For the dispersity of material properties, geometry dimensions and other parameters, uncertainty is
Received 5 November 2020 unavoidable introduced into the solution of structural stability problem. The paper presents a new
Received in revised form 18 February 2021 model for the evaluation of structural buckling loads with unknown-but-bounded parameters. Regarding
Accepted 2 March 2021
unknown-but bounded parameters as interval variables, the elastic stiffness matrix, geometric stiffness
Available online 17 March 2021
Communicated by Jang-Kyo Kim
matrix and eigenvalue with uncertain parameters are divided into deterministic part and perturbation
part taking use of perturbation theory. The deterministic part of eigenvalues is calculated by Finite
Keywords: Element method utilizing the deterministic part of elastic stiffness matrix and geometric stiffness
Uncertain buckling matrix. And the interval uncertain part of eigenvalues is derived by interval arithmetic and first order
Interval parameters perturbation theory. Eventually, the upper bound and lower bound of structural buckling eigenvalues then
Eigenvalue buckling can be easily obtained by summing the deterministic part and uncertain part of buckling eigenvalues.
Perturbation theory Comparing the results with traditional Monte Carlo simulation, two numerical examples are given
to demonstrate the accuracy and efficiency of the proposed method for the solution of eigenvalue
buckling problems with unknown-but-bounded parameters. The results also present the coincidence of
the proposed method with probabilistic method.
© 2021 Elsevier Masson SAS. All rights reserved.
https://doi.org/10.1016/j.ast.2021.106634
1270-9638/© 2021 Elsevier Masson SAS. All rights reserved.
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
Nomenclature
the uncertainty models and found that the plate width has negli- Considering the inherent objective existence of uncertainties
gible influence on the buckling strength uncertainty. and insufficient information, the purpose of this paper is to de-
However, as the sufficient information regarding uncertainties is velop a technique based on perturbation method to analyze the
often difficult to obtain [21–23], many problems exist in the uncer- structural eigenvalue buckling with interval parameters. In sec-
tain stability problem using probabilistic theory such as the choice tion 2, the eigenvalue problem of buckling analysis and the per-
of uncertainty model, the inaccuracies in the tail of probabilistic turbation method for the eigenvalue are introduced. The technique
distribution function and so on. Indeed, a small error in proba- for interval analysis of buckling loads with unknown-but-bounded
bilistic data may cause a large error in the evaluation of the un- parameters utilizing perturbation method is proposed in section 3.
certainty problems [24,25]. Since the formulation of set-theoretical To verify the coincidence of interval theory and probabilistic the-
models of uncertainty usually requires far less information than ory, the uncertain buckling analysis using probabilistic method is
probabilistic models, in order to overcome the shortcomings of introduced in section 4. Section 5 demonstrates three numerical
previous probabilistic methods, some non-probabilistic approaches examples to illustrate the efficiency and accuracy of the proposed
such as convex model and interval analysis have aroused wide con- technique, and some conclusions are drowned in the last section.
cern as an important complement to the probabilistic theory. Many
achievements have been obtained since the non-probabilistic the- 2. Problem statement
ory was proposed. Only the bounds of uncertain parameters are
required, the non-probabilistic methods including non-probabilistic Considering the elastic eigenvalue buckling function as
convex models [25] and interval analysis methods [26–28] could
predict the structural response without knowing the probabilistic K e − qi K g ϕi = 0 (1)
distributions of uncertainties.
where K e is n × n elastic stiffness matrix, K g is n × n geomet-
Based on set-theoretical models, Ben-Haim and Elishakoff [29]
ric stiffness matrix. Eigenvalue q i and eigenvector ϕ i are the i-th
presented a non-probabilistic, set-theoretical treatment of the
buckling load factor and corresponding buckling mode, respec-
buckling of shells with uncertain initial geometrical imperfections.
tively. It can be known from finite element theory that the elastic
Lindberg [30] discussed the application of convex models proposed
stiffness matrix K e and geometric stiffness matrix K g are positive
in Ref. [25] and pointed out that the non-probabilistic unknown-
definite [39].
but-bounded model based on the idea of set membership and
Based on the generalized eigenvalue theory, Eq. (1) has eigen-
convex modeling is a useful alternative in modeling uncertainty
value qi , i = 1, 2, · · · , n and their corresponding eigenvector ϕ i ,
of imperfections. By the use of convex model, Elishakoff et al.
[31] developed an analytical tool to incorporate the uncertain- i = 1, 2, · · · , n. Taking use of the orthogonality of eigenvector [40],
ties in elastic moduli into analysis firstly. Both the upper and we can obtain
lower bounds of the buckling load are derived in the paper. Pan- T 1, i= j
ϕi K g ϕ j = δi j = (2)
telides [32] examined the bucking and post-buckling behavior of 0, i = j
thin-walled stiffened elements under uniform compression with
geometric and material uncertainties by using convex model to Considering the property of generalized eigenvalue, Eq. (1) is
model the uncertainties in geometry and material properties. Ex- equivalent to
pressing stiffness properties as an interval quantity, Modares et al.
[33] extended the interval finite element methods (IFEM) to solve
det K e − qi K g = K e − qi K g = 0 (3)
the possible values of loads that will result in structural stability In engineering practice, the uncertainties of structural elastic
failure. Based on interval mathematics and Taylor series expansion, stiffness matrix K e caused by the uncertain structural parameters
Qiu et al. [34] presented a method performed on interval math- and material properties introduce uncertainty into structural crit-
ematics and Taylor series expansion to solve the buckling load of ical load and corresponding buckling mode. In this condition, the
composite laminate with uncertainties, the high accuracy and sim- influence of uncertainties on buckling behavior is necessary to be
ple calculation demonstrate that the presented method is easily studied. The paper mainly deals with the structural elastic eigen-
accepted and applied in engineering. value buckling problem under interval uncertainty.
In addition to convex models and interval analysis methods, With unknown-but-bounded parameters, the elements in struc-
perturbation method is a relatively simple and efficient means to tural elastic stiffness matrix K e are fluctuated in a range and can
deal with non-probabilistic uncertain problems. Utilizing an arti- be expressed as [41]:
ficial small parameter for numerical analysis of complex systems
with unknown-but-bounded parameters, the perturbation method K e ≤ Ke ≤ Ke (4)
has been applied to structural static response analysis [35], dy-
namic response problem [36], acoustic field prediction [37] and where K e and K e are the lower bound and upper bound of elastic
structural-acoustic problems [38] with uncertain parameters, but stiffness matrix. K e and K e can be calculated by interval propa-
has been rarely applied in the structural buckling problem with gation methods such as Taylor expansion method and Polynomial
unknown-but-bounded parameters. Chaos Expansion method. Based on interval mathematics [42], the
2
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
structural buckling function Eq. (1) and Eq. (2) then can be trans- be regarded as perturbed system. Then the structural eigenvalue
formed to buckling problem can be solved by perturbation theory. Using per-
turbation theory, K e can be written as [41]:
( K e ) I − qi K g ϕ i = 0 (5)
K e = K e0 + ε K e (18)
ϕ iT K g ϕ i = 1 (6)
The constant ε in Eq. (20) is a scalar quantity much less than
where
unity and is called the perturbation factor. The eigenvalue of elastic
stiffness matrix and geometric stiffness matrix after perturbation
( K e ) I = [ K e , K e ]. (7)
K e0 + ε K e should satisfy
As K e is interval variables, the bucking eigenvalue q i is an in-
terval variable. The solution of interval structural buckling problem K e0 + ε K e − (qi ) K g ϕi = 0 (19)
is the solution of boundary of buckling factor:
T
1, i= j
ϕ 0i Kg ϕ 0j = δi j = , i , j = 1, 2, · · · , n (20)
I
(qi ) = qi , qi (8) 0, i = j
To solve the problem, we mainly study the buckling problem Similarly, the structural buckling load q0i and its corresponding
with the constraint of Eq. (4). The following sections of this paper buckling modes are
derive the bounds of buckling loads.
qi = q0i + ε qi (21)
0
3. The interval perturbation method for eigenvalue buckling ϕ i = ϕ + ε ϕ ii (22)
The structural elastic stiffness matrix K e can be transformed to Substituting Eq. (21), (22) into Eq. (19), we can get
be function of stiffness matrix of nominal stiffness and uncertain
K e0 + ε K e − q0i + ε qi K g ϕ 0i + εϕ i = 0. (23)
stiffness:
As K e and K e0 are all positive definite [39], K e is semi-positive Substituting Eq. (25) and Eq. (26) into Eq. (17), the upper bound
definite [43]. For nominal structure, its eigenvalue problem can be and lower bound of structural buckling eigenvalue can be calcu-
expressed as: lated by
K e0 − q0i K g ϕ 0i = 0, i = 1, 2, · · · , n (12)
T
qi = q0i + ϕ 0i K e ϕ 0i (27)
T 1, i= j
0 0
ϕ K g ϕ = δi j = , i , j = 1, 2, · · · , n (13)
i j 0, i = j and
0 T
When the elastic stiffness matrix is interval matrix, the eigenvalue qi = q0i 0
− ϕi K e ϕi , (28)
also be an interval vector. By means of interval extension of in-
terval mathematics [42], the interval eigenvalue namely interval respectively.
critical load can be expressed as:
4. The uncertain buckling analysis with random variables
(q) I = q0 + (q) I (14)
where Utilizing the probabilistic perturbation method, the mean value
and standard deviation of buckling eigenvalue can be expressed as
(q) I = [q, q], (q) I = [−q, q] (15) [44]:
By virtue of interval operation, we can get
E { p i } = E { p 0i } = p 0i (29)
I 0 0 0
(q) = [q, q] = q + [−q, q] = [q − q, q + q] (16) And
⎛ ⎞
Based on the necessary and sufficient condition for the equal of
l
2
interval, we can get ⎝
ϕi0 n ⎠
Ke 2 2
σ ( pi ) = σmn ϕi0 m (30)
q = q0 + q, q = q0 − q (17) m,n=1
Taking use of the perturbation theory, structural elastic stiff- where σ ke and σ k g are the standard deviation of elastic stiffness
ness matrix K e0 can be regarded as nominal system, while K e can matrix and geometric stiffness matrix. Let k be a positive integer,
3
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
p = p 0i − kσ (qi )
i
⎛ ⎞
l
2 (32)
= p 0i − k⎝ ϕi0 n ⎠
K 2 2
σmne ϕi0 m
m,n=1
4
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
⎡ ⎤
0 0 0 0 0 Table 2
⎢0 0 0 −12n/l2 −6n/l ⎥
⎢ ⎥ The eigenvalue under different α.
K e2,1 = ⎢
⎢0 0 0 −6n/l 2n ⎥ ⎥, α q q1
⎣0 0 0 0 0 ⎦
1
Interval Monte Carlo Interval Monte Carlo
0 0 0 0 0 perturbation simulation perturbation simulation
⎡ ⎤
12/h2 0 6/h −12/h2 6/h 0.01 1.205 1.20503 1.22935 1.22933
⎢ 0 12n/l2 −6n/l 0 0 ⎥
0.02 1.19283 1.19285 1.24152 1.2415
⎢ ⎥ 0.03 1.18066 1.18068 1.25369 1.25355
K e2,2 = ⎢
⎢ 6/h −6n/l 4 + 4n −6/h 2 ⎥ ⎥ 0.04 1.16849 1.16856 1.26586 1.26584
⎣ −12/h2 0 −6/h 12/h2 −6/h ⎦ 0.05 1.15632 1.1564 1.27803 1.278
6/h 0 2 −6/h 4 0.06 1.14414 1.14429 1.2902 1.29016
0.07 1.13197 1.13202 1.30238 1.30231
0.08 1.1198 1.11989 1.31455 1.31439
5.1.2. The solution of eigenvalue buckling based on proposed model 0.09 1.10763 1.10767 1.32672 1.32652
Assuming the elastic modulus has perturbation of E 1 and 0.1 1.09546 1.09563 1.33889 1.33853
E 2 in E 1 and E 2 respectively, the disturbance of elastic stiffness
matrix and geometric stiffness matrix in Eq. (24) are
K e,1 K e,2 Table 3
K e = (36) The error of the perturbation with different α.
K e,2 K e,1
α Error of lower bound (%) Error of upper bound (%)
12 E 1 I 1 /h3 6 E 1 I 1 /h2
where K e,1 = , K e ,2 = 0.01 0.00218 0.0012
6 E 1 I 1 /h2 4 E 1 I 1 /h + 4 E 2 I 2 /l
0.02 0.00145 0.0016
0 0 0.03 0.00184 0.01125
. 0.04 0.00654 0.00152
0 2 E 2 I 2 /l
0.05 0.00731 0.00242
Submitting Eq. (36) into Eq. (25) and Eq. (26), we can get the
0.06 0.01233 0.0031
lower bound and upper bound of the plane-frame structure which 0.07 0.00398 0.00509
can be expressed as 0.08 0.00787 0.0121
T 0.09 0.00356 0.01474
q = q1,0 + K e ϕ 1,0
ϕ 1,0 0.1 0.0162 0.02725
1
⎡ ⎤ ⎡ ⎤
−1.4894 T
−1.4894
⎢ 0.2595 ⎥ K e,1 K e,2 ⎢ 0.2595 ⎥
= 1.2172 + ⎢
⎣
⎥ ⎢ ⎥
⎦ K e,2 K e,1 ⎣ −1.4894 ⎦
−1.4894
0.2595 0.2595
and
T
q = q1,0 − K e ϕ 1,0
ϕ 1,0
1
⎡ ⎤ ⎡ ⎤
−1.4894 T
−1.4894
⎢ 0.2595 ⎥ K e,1 K e,2 ⎢ 0.2595 ⎥
= 1.2172 − ⎢
⎣
⎥ ⎢ ⎥
⎦ K e,2 K e,1 ⎣ −1.4894 ⎦
−1.4894
0.2595 0.2595
Then we can get the upper bound and lower bound of struc-
tural buckling eigenvalue with different uncertain parameter α .
The upper bound and lower bound of buckling eigenvalue with
uncertain parameter from 0.01 to 0.1 calculated by the proposed
method and Monte Carlo Simulation are given in Table 2. 106
samples are implemented in Monte Carlo simulation. In order to
demonstrate the accuracy of the interval perturbation method, the
relative errors are listed in Table 3. In order to investigate the
impact of uncertainty, the bounds of uncertain parameter with dif- Fig. 2. Interval bounds calculated by proposed method and Monte Carlo simulation.
ferent uncertain parameter α are plotted in Fig. 2. (For interpretation of the colors in the figure(s), the reader is referred to the web
It can be seen from Table 1 and Table 2 that the buckling loads version of this article.)
obtained from the proposed method are very close to those ob-
tained from Monte Carlo simulation, the largest error 0.02725%
appears in the upper bound when uncertain parameter α = 0.1, buckling eigenvalue with the proposed method and probabilistic
which means that the interval perturbation has a high accuracy in method. In this paper k is set as 3.
the interval analysis of buckling loads. Besides, it can be known The results obtained by proposed interval method and proba-
from Fig. 2 the interval width between the bounds of buckling bilistic method are plotted in Fig. 3. It can be deduced that the
loads presents an increasing tendency with the increasing of un- width of the buckling eigenvalue obtained by the interval pertur-
certain parameter α . bation method is larger than that by the probabilistic approach
for structures with uncertain-but-bounded structural parameters.
5.1.3. The coincidence verification with probabilistic method This result indicates that the region determined by the interval
In order to verify the coincidence of the interval perturbation perturbation method contains that predicted by the probabilistic
method with probabilistic method, we calculate the interval of perturbation approach.
5
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
Table 5
The lower bound of eigenvalue under different α.
Fig. 3. Interval bounds calculated by interval method and probabilistic method. α q q1
1
Interval Monte Carlo Interval Monte Carlo
perturbation simulation perturbation simulation
0.01 3.995969 3.99597 4.076695 4.076695
0.02 3.955605 3.955607 4.117059 4.117055
0.03 3.915242 3.915245 4.157422 4.157416
0.04 3.874879 3.874882 4.197785 4.197778
0.05 3.834515 3.83452 4.238149 4.238139
0.06 3.794152 3.794157 4.278512 4.2785
0.07 3.753789 3.753795 4.318875 4.318862
0.08 3.713425 3.713432 4.359239 4.359223
0.09 3.673062 3.67307 4.399602 4.399585
0.1 3.632699 3.632708 4.439965 4.439946
per bound and lower bound of buckling load of the Mindlin plate
5.2. The uncertain eigenvalue buckling analysis of Mindlin plate with different uncertain parameter α calculated by the proposed
method, Monte Carlo simulation and colocation method are given
Consider a simply-supported square Mindlin plate under uni- in Table 5 and plotted in Fig. 6. In order to investigate the accuracy
axial initial stress shown in Fig. 4 whose geometry dimension of the interval perturbation method, the relative errors comparing
and elastic parameter are given in Table 4. The buckling loads with Monte Carlo simulation and colocation method are given in
of the plate with uncertainty shall be calculated by the proposed Table 6.
method. From the calculation results, it can be seen that the upper
The buckling problem of this structure could be described bounds and lower bounds obtained from interval perturbation
as: method are very close to those of Monte Carlo simulation, while
the computational cost is reduced. The accuracy of the proposed
K e − pi K g ϕ i = 0, i = 1, 2, . . . , n (37) method is higher than that of colocation method. This numerical
6
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
Table 6
The upper bound of eigenvalue under different α. Table 7
The parameters of the Mindlin plate.
α Error of lower bound (%) Error of upper bound (%)
Parameters Value
0.01 2.20E-05 1.05E-05
0.02 4.49E-05 9.33E-05 Modulus of elasticity (E) 210 × 109
0.03 6.80E-05 0.000138 Poisson ratio (v) 0.30
0.04 9.17E-05 0.000183 Radius (R) 0.5
0.05 0.000116 0.000226 Radius of crossing section (d) 0.05
0.06 0.000141 0.000269
0.07 0.000166 0.000311
0.08 0.000191 0.000352 Table 8
0.09 0.000218 0.000393 The lower bound of eigenvalue under different α.
0.1 0.000245 0.000432
α q
1
q1
Interval Monte Carlo Interval Monte Carlo
example demonstrates the feasibility of the interval perturbation perturbation simulation perturbation simulation
for solving plate buckling problems with interval uncertain param- 0.001 -100454263.5 -100438193.4 -96045556.27 -96020590.92
eters. 0.002 -102835078.8 -102813488 -93984397.42 -93961846.58
0.003 -105343888.4 -105309136.4 -92023239.89 -91994721.53
5.2.1. The coincidence verification with probabilistic method 0.004 -107987682.7 -107936952.3 -90165606.37 -90114241.25
0.005 -110762483.4 -110703810.4 -88366091.99 -88313986.74
In this example, we also calculate the buckling eigenvalue in-
terval with both interval perturbation method and probabilistic
method, the results are plotted in Fig. 7. E I = E, E = Ec − α Ec, Ec + α Ec (40)
From this numerical example, we can see that the region of
the buckling eigenvalue with uncertain-but-bounded parameters and
obtained by the interval analysis method contains that produced
by the probabilistic approach. In other words, it is seen that the v I = [v , v̄] = v c − α v c , v c + α v c (41)
present interval analysis method yields larger bounds. The lower where α is the uncertain parameter. To verify the accuracy of the
bounds within the present interval analysis method are smaller proposed with different uncertain parameter, we calculate the in-
than those predicted by the probabilistic approach. Likewise, the terval of buckling load when α = 0.001 to α = 0.005. The results
upper bounds furnished by the present interval analysis method are given in Table 8 and plotted in Fig. 8, the relative errors com-
are larger than those yielded by the probabilistic approach. This paring with Monte Carlo simulation and colocation method are
kind of results is coincident with the meaning of the probabilistic given in Table 9.
theory and interval mathematics. It can be seen from the results that the proposed method can
accurately calculate the interval of buckling eigenvalue of ring
5.3. The uncertain eigenvalue buckling analysis of ring beam beam structure. The width of buckling eigenvalue increases with
the increase of uncertain parameter. The accuracy of the proposed
In this example, we calculated the interval of buckling value method decreases with the increase of uncertain parameter.
of a ring beam using Euler beam element using finite element
method. The parameters of the ring beam are given in Table 7. 5.3.1. The coincidence verification with probabilistic method
The ring beam is divided into 160 elements in the analysis. The In this example, we also calculate the buckling eigenvalue in-
elastic modulus and passion ratio are assumed to be unknown- terval with both interval perturbation method and probabilistic
but-bounded parameter whose interval is method, the results are plotted in Fig. 9.
7
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
Acknowledgements
Appendix
Fig. 9. Interval bounds calculated by interval method and probabilistic method. 1. The solution of (qi ) I using perturbation method.
Based on perturbation theory, the eigenvalue buckling function
can be expressed as
From this numerical example, it is seen that the present inter-
val analysis method yields larger bounds, which means that the K e0 + ε K e0 − q0i + ε q0i K g ϕ 0i + εϕ 0i = 0 (42)
interval perturbation method contains the interval calculated by
probabilistic. which can be expanded as:
6. Conclusions
K e0 ϕ 0i + ε K e0 ϕ 0i − q0i K g ϕ 0i − ε q0i K g ϕ 0i + K e0 ε ϕ 0i
Describing unknown-but-bounded parameters as interval vari- + ε 2 K e0 ϕ 0i − q0i K g ε ϕ 0i 0i − ε 2 q0i K g ϕ 0i = 0
ables, the paper proposed a new model to predict the upper bound
and lower bound of structural buckling load with combination of (43)
8
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
Ignoring the second and higher rank terms in Eq. (43), we can get [16] B.E. Jun Zhang, Effects of uncertain material properties on structural stabil-
9
Z. Qiu and X. Li Aerospace Science and Technology 113 (2021) 106634
[42] G. Goos, J. Hartmanis, K. Nickel, Interval mathematics, Math. Comput. 30 (1976) [44] C. Jiankang, W. Rupeng, Perturbation solution for thermal expansion buck-
894, https://doi.org/10.2307/2005410. ling of no expansion joint slope pavement plate, Appl. Math. Mech. 3 (1993)
[43] E.R. Ziegel, Probability and statistics for engineering and the sciences, Techno- 277–284, https://doi.org/10.1007/BF02451412.
metrics 46 (4) (2004) 497–498, https://doi.org/10.1198/tech.2004.s245.
10