Professional Documents
Culture Documents
DOI 10.1007/s00158-013-0922-9
RESEARCH PAPER
Received: 15 October 2012 / Revised: 26 February 2013 / Accepted: 2 March 2013 / Published online: 12 June 2013
© Springer-Verlag Berlin Heidelberg 2013
the response of a cylinder to various forms of imperfec- Furthermore, the single dimple imperfection was consid-
tions was not included in traditional sources of KDFs. The ered as a realistic and worst imperfection of thin-walled
excessive conservatism usually corresponds to an increase cylindrical shells.
of structural weight, which is particularly important in Additionally, aiming to obtain improved KDFs for cylin-
the development of large-diameter launch vehicles. With drical shells, considerable efforts were put forth to deter-
the advent of China’s newly developed and future heavy- mine the lower bound of the buckling load based on
lift launch vehicles, the diameter of core stage increases numerical analyses, see Arbocz and Williams (1977), Deml
from 3.35 m to 5 m, then to 9 m, and the study of R/t and Wunderlich (1997), Hilburger and Starnes (2004),
becomes an important issue, see Wang et al. (2012). Obvi- Arbocz and Hol (2005), Biagi and Medico (2008), Prabu
ously, a launch vehicle is likely to be overweight eventually et al. (2010), Degenhardt et al. (2010), and Lindgaard
with an unchanged R/t. However, with regard to a largely et al. (2010). In general, these approaches can mainly be
increased R/t, the effect of initial imperfections on the load- categorized as two approaches. The first approach is to
carrying capacity of a cylindrical shell may become more take the measured initial geometrical imperfections into
significant, which results in an extremely high imperfection account, see Arbocz and Williams (1977), Hilburger et al.
sensitivity. (2004), and Hühne et al. (2008), and high-fidelity nonlin-
In the past, the effects of several types of imperfec- ear finite element analyses that include the effects of a
tions on the load-carrying capacity of thin-walled struc- manufacturing-process-specific geometric imperfection sig-
tures were studied in detail, such as Croll and Batista nature were developed and validated by Hilburger et al.
(1981), Croll and Ellinas (1983), Croll (1995), Bushnell (2004). Besides, the effect of random imperfections is
(1981), Bushnell and Bushnell (1996), Eurocode 3 (1999), another subject of active research, including the variability
Mandal and Calladine (2000), Elishakoff (1983, 2000, of thickness, material properties, boundary conditions, load-
2012), Elishakoff et al. (2001), Teng and Song (2001), Hunt ings, etc. (Elishakoff and Arbocz 1982; Elishakoff 1983,
et al. (2003), Sosa et al. (2006), Huhne et al. (2008), Hrinda 2000, 2012; Elishakoff et al. 2001; Koiter et al. 1994;
(2010), Haynie and Hilburger (2010). Most of these studies Papadopoulos and Papadrakakis 2005). Also, stochastic
were performed using numerical methods, due to the fact FEA of shells was presented using the spectral represen-
that the modern CAE technology has enabled the buckling tation method for the description of the random fields
analysis of cylindrical shells accounting for both geometric (Papadopoulos and Papadrakakis 2004). Further, the effects
and material nonlinearities together with initial imperfec- of material and thickness spatial variation on the buck-
tions. Among various types of imperfections, the eigen- ling load of shells with random geometric imperfections
mode shape imperfection is the most commonly adopted were investigated (Papadopoulos et al. 2009). Based on
one at the design stage, see Teng and Song (2001). The the various data of measured imperfections, probabilistic
European standard for steel shell structures recommends approaches where imperfections were introduced as ran-
that the imperfection should be specified in the form of dom variables with a certain distribution were utilized to
eigen-mode shapes, with its amplitude linked to fabrication derive the improved KDFs for cylindrical shells by Biagi
quality, unless a different unfavourable pattern is justified, and Medico (2008). However, each manufacturing pro-
see Eurocode 3 (1999). Besides, a significant contribution to cess causes a specific imperfection pattern, thus extremely
the area of buckling and post-buckling analyses for imper- large number of experiment data is needed, which is dif-
fect stiffened shells was the development of PANDA2 by ficult to collect in practice. The other approach is to find
Bushnell and Bushnell (1996), taking into account a vari- the “worst” geometrical imperfections, which yields the
ety of imperfections, such as global, local, inter-ring, and lowest performance and thereby a lower bound for the load-
out-of-roundness imperfections. Additionally, Hühne et al. carrying capacity of structures. In recent years, the imper-
(2008) developed a new approach to create a local dimple- fections were represented by a linear combination of base
shape imperfection, which is typically similar to the type shapes which were constructed from a number of eigen-
of dimple that forms in a compressed shell at the onset modes. Then the determination of the “worst” imperfection
of buckling as observed in tests. Further, three different for a given structure was formulated as an optimization
imperfection types were compared, including an imperfec- problem by varying the imperfection amplitudes of each
tion with eigen-mode shape, an imperfection caused by base shape, see Lindgaard et al. (2010). The concept of
a lateral perturbation load, and a single dimple imperfec- this work is advanced and universal for the imperfection-
tion by Haynie and Hilburger (2010). Results indicate that sensitive structures. Although the obtained worst imperfec-
the lateral perturbation technique and the single dimple tions may lead to extremely lower performances of struc-
imperfection produce response characteristics that are tures (Hühne et al. 2008), fortunately, no matter how to
physically meaningful, which were then validated by choose the base imperfection shape it will result in a more
experimental results, see Haynie and Hilburger (2012). robust design.
Determination of realistic worst imperfection for cylindrical shells using surrogate model 779
In practice, it should be noted that KDFs are extremely bending strain energy to remain in equilibrium. Several
related to the structural parameters, material systems, man- incremental-iterative solution approaches have been exten-
ufacturing technology, etc. This may be the reason that sively studied and reported in the existing publications,
engineering practice tends to rely on experiments to deter- and the modified Riks method is an algorithm that enables
mine the KDFs. However, it is economically unsustain- effective solution of such cases, see Crisfield (1981) and
able and impossible to obtain KDFs entirely dependent Powell and Simons (1981).
on experiments, especially for large-scale structures. Since The essence of the modified Riks method is to follow
the main motivation of this study is to assist and guide the static equilibrium path of a structure in a space defined
the experiments and thus provide a reference for improv- by the nodal variables and the loading parameter. Both geo-
ing KDFs, dimple imperfections are selected as the subject metrical and material nonlinearities are considered using the
of this study, based on the consideration that this type Newton method as a basic algorithm and gradually increas-
of imperfections is probably produced in the processes of ing the applied loads, thus the effect of local instabilities
manufacturing, transporting and installation of structures. on the global buckling of structures can be accounted for.
Specifically, the reduction of the buckling load of cylindri- Additionally, in this approach, it is essential to control the
cal shells based on a finite number of dimple imperfections increment step, which is limited by moving a given distance
is investigated in this study. The cylindrical shells with along the tangent line to the current solution point and then
too large amplitude of imperfections or too many dimple searching for equilibrium in the plane that passes through
imperfections are not the scope of concerns in this paper, the point thus obtained and that is orthogonal to the same
because these shells would be considered as substandard tangent line (Crisfield 1981; Powell and Simons 1981).
products in launch and space vehicles and other industrial Besides, dynamic explicit analysis can provide an accu-
applications. rate and robust prediction for the buckling and post-buckling
In this study, the influence of a single dimple imper- behavior of thin-walled structures, which is proved to be in
fection on the load-carrying capacity of cylindrical shells fairly good agreement with experimental result, when using
is firstly investigated by varying the amplitudes, directions the explicit central-difference time integration rule (Lanzi
and positions of imperfection. Then a single δ function 2004; Hao et al. 2012). However, this type of analysis may
is introduced to describe the profile of the dimple imper- be extremely time-consuming because of the large num-
fection in a more general form, which enables to analyze ber of small time increments determined by the minimum
conveniently the effect of the imperfect zone area on the element size.
load-carrying capacity of cylindrical shells. Thereafter, a In a buckling analysis, an axial displacement loading is
surrogate-based optimization framework of determining the applied as axial shortening at one end of the structure grad-
worst realistic imperfection is proposed to investigate the ually, and the sum of reaction forces of the other end is
reduction of the buckling load of cylindrical shells based calculated as an axial load at each time increment. The
on a finite number of dimple imperfections, where the buckling load is specified by the peak point of the load-
variables include the amplitude of the perturbation load shortening curve, which limits the load-carrying capacity,
and the position numbers of the dimple imperfections. The as shown in Fig. 1. The evolution of buckling phenomena,
possible dimple imperfections may be introduced during in terms of deformation, stress and strain, into the post-
the service of cylindrical shells. Finally the effectiveness buckling field until structural collapse resulted from the
of the present method is demonstrated by an example, global instability can be investigated.
and the proposed framework is proved to be more effi-
cient to find the realistic lower bound of buckling load
for cylindrical shells compared with direct optimization
using MIGA.
2 Method of analysis
n
where
R = [R(x1 ), R(x2 ), . . ., R(xn )]T , ij = ϕ
R(x) = λj ϕ(r) (1) xi − xj (i = 1, 2, . . ., n, j = 1, 2, . . .n), and λ = [λ1 ,
j =1 λ2 , . . ., λn]T . Similar to the polynomial regression method,
the coefficients λ can be obtained by solving (4).
Surrogate-based optimization consists of inner optimiza-
r = x − xj (2)
tion and outer update, as shown in Fig. 2. Actually, the inner
where λj is the weight coefficient, x is the vector of design optimization is entirely based on the surrogate model and
variables, xj is the vector of design variables at the jth needs negligible computational cost. The optimization is
sampling point, r is the Euclidean distance between the considered to be converged if the relative error between the
predicted point and the jth sampling point, φ is the basis results obtained from the surrogate model and the ones from
function, and n is the number of sampling points. the exact FEA is less than 0.1 %, otherwise the surrogate
A typical basis function for RBF model is the multi- model is updated by the exact FEA and another inner opti-
quadric (MQ) function, and it can be stated as mization search is performed based on the new surrogate
model. Updating surrogate model also requires negligible
ϕ(r) = r 2 + c2 (3) computational cost, because it only involves refitting the
model with both old and new sampling data.
where c is a given constant ranging from 0 to 1. Replacing A surrogate model needs to be validated with error
x and R(x) in (1) with the n vectors of design variables and analysis before it can be used with full confidence. Thus,
Fig. 2 Framework of
surrogate-based optimization
Determination of realistic worst imperfection for cylindrical shells using surrogate model 781
where yi is the response value at the ith test point from the
exact function, ỹi is the prediction value from the surrogate
model, n is the number of sampling points. The smaller val-
ues of three error metrics (closer to 0), the higher quality of
the surrogate model.
n Then, numerical methods including the modified Riks
1
n |yi − ỹi | method and dynamic explicit method are carried out
i=1
%AvgErr = 100 (6) respectively by Abaqus (2008) accounting for geometric
n
nonlinearities. Four-node shell element with reduced inte-
1
n yi
i=1 gration (S4R) is utilized to discretize the model, and the
⎡ ⎤ element nodes include three translational degrees of free-
dom and three rotational degrees of freedom. A cylindrical
⎢ |yi − ỹi | ⎥
%MaxErr = Max ⎢
⎣100 n
⎥
⎦ (7) coordinate system is defined with its origin at the center
1
yi of the√fixed end. The element size is chosen in the order
n
i=1 of 0.5 Rt for two methods as proposed by Meyer-Piening
Fig. 5 Schematic of
perturbation loads with three
directions. a case 1: along the r
axis in the zr plane; b case 2: at
a 45-degree angle to the r axis in
the zr plane; c at a 45-degree
angle to the r axis in the rθ plane
782 B. Wang et al.
et al. (2001). To balance the computational cost and accu- analysis under a given perturbation load. Then, the defor-
racy, a loading time of 200 ms is chosen eventually for mation of the cylindrical shell is introduced to the perfect
the dynamic explicit analysis. The buckling loads of the geometry by modifying the nodal coordinates. Finally, an
perfect shell obtained by the dynamic explicit method and
the modified Riks method are 66.1 kN and 69.1 kN respec-
tively. The load vs. end-shortening curves obtained by two
methods are shown in Fig. 4, which agree reasonably well
with the increase of the end-shortening before reaching the
point of buckling load. However, the computational cost
of such an explicit analysis is about three times of the one
of the modified Riks method. Consequently, although the
dynamic explicit analysis is able to obtain the buckling and
post-buckling deformation shapes of structure, the modified
Riks method is adopted for the buckling analyses and then
optimizations in this study for its reasonable accuracy and
high efficiency.
Fig. 6 Imperfection sensitivity curves for perturbation loads with Fig. 7 The distribution of five dimple positions along the axial
three directions direction (Z1–Z5)
Determination of realistic worst imperfection for cylindrical shells using surrogate model 783
Fig. 8 Imperfection sensitivity curves of cylindrical shells with vary- Fig. 10 Imperfection sensitivity curves obtained by the dynamic
ing dimple positions (Z1–Z5) explicit method and the modified Riks method for Z1
784 B. Wang et al.
20 N, two local minimums of buckling loads (38.1 kN for shown in Fig. 9. The maximal amplitude of the deviation
Z1 and 38.9 kN for Z5) are reached along the imperfection from the perfect geometry caused by the perturbation load
sensitivity curves respectively. For N > 30 N, the buckling is 0.3 mm when N = 10 N (1.5 mm when N = 30 N,
loads almost do not decrease with the further increase in the 3.7 mm when N = 60 N, 5.2 mm when N = 90 N).
perturbation loads for each dimple location. Compared with the radius of 250 mm, the shell length
The relationship between the perturbation load N and of 510 mm and the thickness of 0.5 mm, these imper-
the amplitude of dimple imperfection w can be obtained, as fections when N < 30 N are relatively small and not
Table 1 Deformation shapes of the cylindrical shell by the dynamic explicit method with various perturbation loads and end-shortenings for Z1
N = 7.5 N N = 20 N N = 40 N
Ue =
0.332
mm
Ue =
0.357
mm
Ue =
0.382
mm
Ue =
0.407
mm
Ue
=0.433
mm
Ue =
0.458
mm
Ue =
0.483
mm
Determination of realistic worst imperfection for cylindrical shells using surrogate model 785
visible to be detected, thus considered to be realistic in Pcr = 37.8 kN. After reaching this minimum, the imperfec-
this study. tion sensitivity curve re-rises in a small amplitude and then
To further investigate the cause of the local minimum of approximately converges. When N = 40 N, the trend of the
curves in Fig. 8, a series of dynamic explicit analyses are deformation shape is similar to the one of N = 20 N. The
performed for the load position Z1. The comparison of the difference is that the local buckling is relatively unstable and
imperfection sensitivity curves obtained by two methods is spreads out wavelike at Ue = 0.382 mm for N = 20 N,
shown in Fig. 10, and the tendency of two curves is quite while the shell is still able to carry further load at Ue =
similar. Additionally, since the dynamic explicit analysis is 0.382 mm, and the local buckling evolves into a global fail-
able to obtain the post-buckling response of the structure, ure rapidly at Ue = 0.407 mm for N = 40 N. In essence,
the deformation shapes of three cylindrical shells which this difference may exist in the load path of such a redun-
stand for three key points along the imperfection sensitivity dant structure under axial compression. A larger amplitude
curve are plotted in detail, as shown in Table 1. Specifi- of dimple imperfection usually results in a lower stiffness
cally, at the onset of the imperfection sensitivity curve, e.g. and therefore a smaller compressive load for the region
N = 7.5 N, the dimple shape does not result in a local around the dimple imperfection, while higher load may be
buckling with the increase of the end-shortening Ue , and distributed to the region far away from the dimple imperfec-
the global buckling occurs firstly at Ue = 0.458 mm cor- tion. Therefore the total load-carrying capacity may instead
responding to a buckling load of Pcr = 46.4 kN. When be higher in this case compared with the one of a smaller
N = 20 N, the imperfection sensitivity curve reaches a amplitude of dimple imperfection. The lower bound of Pcr
minimum. For this case, the local buckling occurs at a low given by these curves using the dynamic method and the
level of the end-shortening. With the increase of the end- modified Riks method are 37.8 kN and 36.2 kN respectively.
shortening, the local buckling evolves into a global buckling In the previous works, a relatively large perturbation
at Ue = 0.382 mm corresponding to a buckling load of load was chosen to determine the lower bound of the buck-
ling load. However, the above results indicate that a large
perturbation load may not lead to the lower bound of the
buckling load.
Fig. 12 Imperfection sensitivity curves of cylindrical shells with Fig. 13 Schematic of the shell elements in the interferential zone for
varying diameters and amplitudes of imperfections the cases of λ = 29, 30, 31 mm
786 B. Wang et al.
distance between the node in the imperfect zone and the cen-
ter of the imperfection. The profile together with the FEM
model of such an imperfection is shown in Fig. 11.
In this study, three typical amplitudes of the dimple
imperfection are investigated, t, 3t and 5t respectively. The
center of the dimple imperfection is assumed to be located
at the mid-length of the shell along the axial direction. Then
the imperfection sensitivities of cylindrical shells with vary-
ing λ are analyzed, as shown in Fig. 12. Similar to the
Fig. 14 Schematic of the interferential zones of cylindrical shells dimple imperfection caused by the perturbation load, the
buckling loads drop sharply with the increase of λ on the
onset of three curves, then the buckling loads converge to
3.4 Response of cylindrical shell with a single delta almost the same value when the diameter λ > 30 mm.
function imperfection Finally, the lower bound of the buckling loads given by these
curves is 40.7 kN, which is close to the one given by the
Considerable efforts have been put forth to determine ana- curve of Z3 in Fig. 8 (Pcr = 41.0 kN).
lytically the effects of geometrical imperfections which are It should be noted that the curve of δ0 = t in Fig. 12
described by mathematical models on the buckling response shows a local minimum at λ = 30 mm. To investigate the
of structures, ranging from the stability problem of columns cause of this minimum, the buckling analyses of the cylin-
to the buckling of stiffened panels (Timoshenko and Gere drical shells with three dimple imperfections (λ = 29, 30,
1961; Calladine 1983; Thompson and Hunt 1984; Croll 31 mm) are performed, and the buckling loads are 41.0,
1995). Thus it is a natural way to develop a mathematical 40.7, 41.8 kN respectively. Finite elements are utilized to
model in order to describe the dimple imperfections in a discretize the cylindrical shell model, and the dimple imper-
more general form. Based on the function to describe the fections are introduced by modifying the nodal coordinates
weld depression in the study of Rotter and Teng (1989), of the perfect geometry, as shown in√ Fig. 13. The ele-
a modified δ function can be expressed as (9) and (10) to ment size is chosen in the order of 0.5 Rt as proposed by
describe the circular dimple imperfection, which is assumed Meyer-Piening et al. (2001). For the case of λ = 30 mm,
to be in the shape of a circle, indicated by the round gray approximately 30 elements fall into the imperfect zone. For
area in Fig. 11. the case of λ = 31 mm, several elements near the bound-
πs ary fall into the imperfect zone. Otherwise, for the case
δa (s) = δ0 e−πs/λ cos (9)
λ of λ = 29 mm, several elements move out of the imper-
fect zone. This may cause numerical noises to the curve in
Fig. 12. Though element refinement would be conductive
s= (Rθ − Rθ0 )2 + (z − z0 )2 (10)
to the solution of this problem, however, the computa-
where λ and δ0 are the diameter and amplitude of the dim- tional burden may cause serious concerns, especially for
ple imperfection, θ0 and z0 are the circumferential and axial the surrogate-based optimization. Fortunately, the buckling
coordinates of the centre of the dimple imperfection, s is the loads with three imperfections (λ = 29, 30, 31 mm) are
Fig. 16 Schematic of the combination of load positions together with Fig. 18 Effect of the distance l on the buckling load of the cylindrical
Sa and Sc shell
788 B. Wang et al.
l = 0 mm l = 20 mm l = 35 mm l = 40 mm l = 45 mm l = 50 mm l = 60 mm l = 80 mm
√
maximum perturbation load Nmax in the proposed method Sc = 3l/2 (12)
are determined according to the current fabrication quality
and detection tolerance. Meanwhile, the range of the per-
turbation load N, which is a variable of the surrogate-based L 2L
na = −1= −1 (13)
optimization, can be specified as [0, Nmax ]. In Step 2, a Sa 3l
combined dimple shape shown in Fig. 16 is introduced to
perform the imperfection sensitivity analysis, in which the 2πR 4πR
considered maximum imperfection amplitude and the cor- nc = = √ (14)
Sc 3l
responding perturbation load Nmax are adopted. By varying
where Sa and Sc are the distances of two adjacent load
the distance l (the distance between the circumcenter and
positions in the axial and circumferential directions respec-
each triangle vertex in Fig. 16) from zero, the curve rep-
tively, na and nc are the numbers of possible load positions
resenting the relationship between the buckling load and
in the axial and circumferential directions respectively. It
l can be plotted. The results in Section 3.2 indicate that
is obvious that the smaller l means more possible load
the buckling load almost does not decrease with the further
positions, higher calculation accuracy, and of course larger
increase of the perturbation load after the turning point of
design space. However, the requirement of more samples
the imperfection sensitivity curve (N > 30 N) (N = 30 N,
in the surrogate-based optimization due to the larger design
Pcr = 41.5 kN; N = 90 N, Pcr = 41.0 kN). Initially, the
space may cause serious concerns. Thus l is suggested to
buckling load of cylindrical shell with the combined dim-
take a value of le herein, and such a combined pattern is
ple imperfection would be close to the one obtained by the
approximately considered to represent a rational distance
single dimple imperfection, since three load positions are
between each load position, aiming to find the worst dimple
overlapped when l = 0 mm. Then, the buckling load would
imperfection based on three dimple imperfections.
vary with the increase of the distance l, and the distance l
Once Na and Nc are determined, each position where the
corresponding to the first minimum buckling load is referred
dimple imperfection may be produced is assigned with a
to as effective distance le . Such a combined pattern with
position number, as shown in Fig. 17. The position number
the effective distance is approximately expected to represent
a rational distance between each loads position, assuming
that the detrimental interaction of adjacent load positions
has been enveloped. Based on the effective distance le , the
spacing of load positions along the axial and circumferen-
tial directions Sa and Sc can be determined by (11) and (12).
Further, the numbers of possible load positions in the axial
and circumferential directions Na and Nc can be determined
by (13) and (14). Note that these load positions are assumed
to be evenly distributed in the shell.
8.4 6.0 24.8 Fig. 19 Iterations of outer updates in the surrogate-based optimization
by the proposed method
Determination of realistic worst imperfection for cylindrical shells using surrogate model 789
fections are 158, 180 and 271, while the perturbation load is 8007 is relatively conservative, and the proposed method
18.1 N. The imperfect geometry of the optimum design in has the possibility of reducing the conservatism of KDF.
an expanded view obtained by MIGA is shown in Fig. 23. In practice, industry practice tends to rely on experi-
Although the position numbers are not exactly the same ments to determine the KDFs. However, it is economically
as the ones obtained by the proposed method, the two unsustainable and impossible to obtain KDFs entirely
worst imperfection cases have the similar relative distances dependent on experiments, especially for large-scale struc-
between each load position. Because the cylindrical shell is tures. Numerical methods as well as virtual experiments are
cyclic symmetry along the circumferential direction, only promising to assist and guide the experiments, then further
the relative positions of each load have the influence on the provide a reference for the improved KDFs. For the pur-
load-carrying capacity of the cylindrical shell. pose of comparison, the imperfections of different order
From the point-of-view of computational cost, the pro- eigen-mode shapes which are generally utilized to assess the
posed method has a significant advantage, as the total runs imperfection sensitivity of thin-walled structures are also
of the exact FEA is 113 and the total computational time is considered, as shown in Fig. 24. The lower bound of the
about 103,000 s, while MIGA needs 600 runs of the exact buckling load can be obtained by varying the maximum
FEA for a quasi-optimum design and 1,000 runs for the amplitude of imperfection from 0 to 0.5 mm (one skin thick-
optimum design, and accordingly the computational time is ness) for each mode order, and then the KDFs can be derived
about 900,000 s, thus 797,000 s savings are achieved by the subsequently, as listed in Table 4. It is worth noting that the
proposed method. KDF obtained by the first order eigen-mode shape is even
lower than the one obtained by NASA SP-8007, however,
KDF = 1 − 0.901(1 − e−φ ) (18)
the first order eigen-mode shape is also not the worst one
among the six selected ones, since the KDF obtained by the
φ= R/t/16 (19)
10th order eigen-mode shape is 0.28.
Also, KDF obtained by NASA SP-8007 using (18) and (19) Furthermore, another two cylindrical shells with shell
is 0.32, as listed in Table 4. In the proposed study, KDF is length L = 1, 020 and 255 mm are established. Aside
defined as from the parameter L, all the other parameters are identical
with the previous shell, including the boundary condition
KDF = Pimp /Pcr (20) and loading pattern. Comparison of KDFs obtained by the
imperfections of different order eigen-mode shapes for three
where Pimp is the lower bound of the buckling loads of shells can be observed from Fig. 25. As can be seen, the
imperfect cylindrical shells. It is evident that NASA SP- first order eigen-mode shape is not the worst one for each
Arbocz J, Hol JM (2005) On a verified high-fidelity anal- In: 51st AIAA/ASME/ASCE/AHS/ASC Structures, Structural
ysis for axially compressed orthotropic shells. In: 46th Dynamics and Materials Conference, Orlando, FL, [AIAA-2010-
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics 2532]
and Materials Conference, Austin, TX, [AIAA-2005-2302] Haynie WT, Hilburger MW (2012) Validation of lower-bound
Arbocz J, Starnes JH (2002) Future directions and challenges in shell estimates for compression-loaded cylindrical shells. 53rd
stability analysis. Thin-Walled Struct 40(9):729–754 AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics
Arbocz J, Williams JG (1977) Imperfection surveys on a 10-ft- and Materials Conference, Honolulu, [AIAA-2012-1689]
diameter shell structure. AIAA J 15(7):949–952 Hilburger MW, Starnes JH (2004) Effects of imperfections of
Biagi M, Medico FD (2008) Reliability-based knockdown factors the buckling response of composite shells. Thin-Walled Struct
for composite cylindrical shells under axial compression. Thin- 42(3):369–397
Walled Struct 46(12):1351–1358 Hilburger MW, Nemeth MP, Starnes JH (2004). Shell buckling
Bushnell D (1981) Buckling of shells—pitfall for designers. AIAA J design criteria based on manufacturing imperfection signatures.
19(9):1183–1226 [NASA/TM-2004-212659]
Bushnell D, Bushnell WD (1996) Approximate method for the opti- Hrinda GA (2010) Effects of shell-buckling knockdown factors in
mum design of ring and stringer stiffened cylindrical panels and large cylindrical shells. In: 5th International Conference on High
shells with local, inter-ring, and general buckling modal imperfec- Performance Structures and Materials, Tallinn, Estonia
tions. Comput Struct 59(3):489–527 Hühne C, Rolfes R, Breitbach E, Teßmer J (2008) Robust design of
Calladine CR (1983) Theory of shell structures. Cambridge University composite cylindrical shells under axial compression-simulation
Press, Cambridge and validation. Thin-Walled Struct 46(7):947–962
Calladine CR (1995) Understanding imperfection-sensitivity in the Hunt GW, Lord GJ, Peletier MA (2003) Cylindrical shell buckling:
buckling of thin-walled shells. Thin-Walled Struct 23(1–4):215– a characterization of localization and periodicity. Discret Contin
235 Dyn Syst 3(4):505–518
Crisfield MA (1981) A fast incremental/iteration solution procedure Jin R, Chen W, Simpson TW (2001) Comparative studies of meta-
that handles ‘snap-through’. Comput Struct 13(1–3):55–62 modelling techniques under multiple modelling criteria. Struct
Croll JGA (1995) Towards a rationally based elastic–plastic shell Multidisc Optim 23(1):1–13
buckling design methodology. Thin-Walled Struct 23(1–4): Koiter WT, Elishakoff I, Li YW, Starnes JH (1994) Buckling of an axi-
67–84 ally compressed cylindrical shell of variable thickness. Int J Solids
Croll JGA, Batista RC (1981) Explicit lower bounds for buckling of Struct 31(6):797–805
axially loaded cylinders. Int J Mech Sci 23(6):331–343 Lanzi L (2004) Composite stiffened panels in post buckling: exper-
Croll JGA, Ellinas CP (1983) Reduced stiffness axial load buckling of iments and dynamic explicit analyses with LS-DYNA. In: 30th
cylinders. Int J Solids Struct 19(5):461–477 European rotorcraft forum, Marseilles, France
Degenhardt R, Kling A, Bethge A, Orf J, Karger L, Zimmermann Lindgaard E, Lund E (2011) Optimization formulations for the max-
R, Rohwer K, Calvi A (2010) Investigations on imperfection imum nonlinear buckling load of composite structures. Struct
sensitivity and deduction of improved knock-down factors for Multidisc Optim 43(5):631–646
unstiffened CFRP cylindrical shells. Compos Struct 92(8):1939– Lindgaard E, Lund E, Rasmussen K (2010) Nonlinear buckling
1946 optimization of composite structures considering “worst” shape
Deml M, Wunderlich W (1997) Direct evaluation of the ‘worst’ imper- imperfections. Int J Solids Struct 47(22–23):3186–3202
fection shape in shell buckling. Comput Methods Appl Mech Eng Mandal P, Calladine CR (2000) Buckling of thin cylindrical
149(1–4):201–222 shells under axial compression. Int J Solids Struct 37(33):
Elishakoff I (1983) How to introduce the imperfection-sensitivity 4509–4525
concept into design. In: Thompson JMT, Hunt GW (eds) Col- Meyer-Piening HR, Farshad M, Geier B, Zimmermann R (2001) Buck-
lapse: the buckling of structures in theory and practice. Cambridge ling loads of CFRP composite cylinders under combined axial and
University Press, Cambridge torsion loading—experiments and computations. Compos Struct
Elishakoff I (2000) Uncertain buckling: its past, present and future. Int 53(4):427–436
J Solids Struct 37(46–47):6869–6889 Papadopoulos V, Papadrakakis M (2004) Finite element analysis of
Elishakoff I (2012) Probabilistic resolution of the twentieth century cylindrical panels with random initial imperfections. J Eng Mech
conundrum in elastic stability. Thin-Walled Struct 59:35–57 130(8):867–876
Elishakoff I, Arbocz J (1982) Reliability of axially compressed cylin- Papadopoulos V, Papadrakakis M (2005) The effect of material and
drical shells with random axisymmetric imperfections. Int J Solids thickness variability on the buckling load of shells with random
Struct 18(7):563–585 initial imperfections. Comput Methods Appl Mech Eng 194(12–
Elishakoff I, Li YW, Starnes JH Jr (2001) Non-classical problems 16):1405–1426
in the theory of elastic stability. Cambridge University Press, Papadopoulos V, Stefanou G, Papadrakakis M (2009) Buckling anal-
Cambridge ysis of imperfect shells with stochastic non-Gaussian material and
Eurocode 3 (1999) Design of steel structures, Part 1.6: General thickness properties. Int J Solids Struct 46(14–15):2800–2808
rules-supplementary rules for the strength and stability of shell Powell G, Simons J (1981) Improved iterative strategy for nonlinear
structures. CEN, Brussels structures. Int J Numer Methods Eng 17(10):1455–1467
Gutmann HM (2001) A radial basis function method for global Prabu B, Raviprakash AV, Venkatraman A (2010) Parametric study
optimization. J Global Optim 19(3):201–227 on buckling behaviour of dented short carbon steel cylindrical
Hao P, Wang B, Li G (2012) Surrogate-based optimum design shell subjected to uniform axial compression. Thin-Walled Struct
for stiffened shells with adaptive sampling. AIAA J 50(11):2389– 48(8):639–649
2407 Queipo NV, Haftka RT, Shyy W, Geol T, Vaidyanathan R, Tucker PK
Hardy RL (1971) Multiquadric equations of topography and other (2005) Surrogate-based analysis and optimization. Prog Aerosp
irregular surfaces. J Geophys Res 76(8):1905–1915 Sci 41(1):1–28
Haynie WT, Hilburger MW (2010) Comparison of methods to predict Rotter J, Teng J (1989) Elastic stability of cylindrical shells with weld
lower bound buckling loads of cylinders under axial compression. depressions. J Struct Eng 115(5):1244–1263
794 B. Wang et al.
Sosa EM, Godoy LA, Croll JGA (2006) Computation of lower-bound Venter G, Haftka RT (1997) Response surface approximations for
elastic buckling loads using general-purpose finite element codes. fatigue life prediction. In: 38th AIAA/ASME/AHS/ASC Struc-
Comput Struct 84(29–30):1934–1945 tures, Structural Dynamic, and Materials Conf., Kissimmee, FL,
Teng JG, Song CY (2001) Numerical models for nonlinear analysis [AIAA-1997-1331]
of elastic shells with eigenmode-affine imperfections. Int J Solids Wang B, Hao P, Du KF, Li G (2011) Knockdown factor based on
Struct 38(18):3263–3280 imperfection sensitivity analysis for stiffened shells. Int J Aerosp
Thompson JMT, Hunt GW (1984) Elastic instability phenomena. Light Struct 1(2):315–333
Wiley, Chicester Wang B, Hao P, Li G, Du KF (2012) Surrogate-based optimization of
Timoshenko SP, Gere JM (1961) Theory of elastic stability. McGraw- stiffened shells including load-carrying capacity and imperfection
Hill, New York sensitivity. Thin-Walled Struct (submitted)