Compression Buckling of Z-Stiffened Aluminum Panels, with ...
AIAA-2002-1513 BUCKLING LOAD CALCULATIONS OF THE ...
-
Upload
khangminh22 -
Category
Documents
-
view
0 -
download
0
Transcript of AIAA-2002-1513 BUCKLING LOAD CALCULATIONS OF THE ...
AIAA-2002-1513
BUCKLING LOAD CALCULATIONS OF THE ISOTROPIC SHELL A-8
USING A HIGH-FIDELITY HIERARCHICAL APPROACH
Johann Arbocz*
Delft University of Technology, The Netherlandsand
James H. Statues**
NASA Langley Research Center, Hampton, VA 23681-001
ABSTRACT
As a step towards developing a new design
philosophy, one that moves away from thetraditional empirical approach used today in
design towards a science-based designtechnology approach, a test series of 7 iso-tropic shells carried out by Arbocz andBabcock [1] at Caltech is used. It is shown howthe hierarchical approach to buckling load
calculations proposed by Arbocz et al [2] canbe used to perform an approach often called"high fidelity analysis", where the uncertaintiesinvolved in a design are simulated by refinedand accurate numerical methods. The Delft
Interactive Shell DEsign COde (short,
DISDECO} is employed for this hierarchical
analysis to provide an accurate prediction ofthe critical buckling load of the given shellstructure. This value is used later as areference to establish the accuracy of the
Level-3 buckling load predictions. As a finalstep in the hierarchical analysis approach, thecritical buckling load and the estimatedimperfection sensitivity of the shell are verified
by conducting an analysis using a sufficientlyrefined finite element model with one of the
current generation two-dimensional shellanalysis codes with the advanced capabilitiesneeded to represent both geometric andmaterial nonlinearities.
t_
Professor, Faculty of AerospaceEngineering, Fellow AIAASenior Engineer, Structures and MaterialsCompetency, Fellow AIAACopyright O 2002 by J. Arbocz andJ.H. StarnesPublished by AIAA with permission
INTRODUCTION
It is generally agreed that, in order to makethe development of the Advanced SpaceTransportation System a success and toachieve the very ambitious performance goals(like every generation of vehicles 10x safer and10x cheaper than the previous one), one mustmake full and efficient use of the technical
expertise accumulated in the past 50 years orso, and combine it with the tremendouscomputational power now available. It isobvious that with the strict weight constraintsused in space applications these performancegoals can only be achieved with an approachoften called "high fidelity analysis', where theuncertainties involved in a design are
simulated by refined and accurate numericalmodels. In the end the use of =high fidelity _numerical simulation will also lead to overallcost reduction, since the analysis and design
phase will be completed faster and onty thereliability of the final configuration needs to beverified by structural testing.
The light-weight shell structures used inaerospace applications are often bucklingcritical. The buckling load calculations are
usually carried out by one of the manycurrently available finite element basedcomputer codes [e.g., 3,4]. In order to reducecomputer execution time, buckling analysesare often done using only the small
displacement stiffness matrix K o . This
approach is used, despite the fact that the"initial stability problem" so formulated can onlygive physically meaningful answers if theelastic solutions based on K o (at least
approximately) are identically equal to zero [5].When the qualitative nature of the expected
behavior is completely unknown, the stability ofthe structure must be investigated using the full
tangent stiffness matrix KT in order to
guarantee accurate and reliable buckling loadand buckling mode predictions. In order todiscover the load level at which K T ceases to
be positive definite (that is, the load level whenCopyright C, 2002 by Johann Arbocz. Published by the &rnerican Institute of Aeronautics and Astronautics, Inc., with permission.
2396
buckling occurs), a step-by-step analysisprocedure is needed.
In addition, it is imperative (though oftencompletely neglected), that at the beginning ofany stability investigation, the accuracy of thediscrete model used should be checked
against available analytical or semi-analyticalresults. This step is part of a mandatory studyneeded in order to establish the dependence ofthe buckling load predictions on the meshdistribution used. Furthermore, as has beenpointed out in the past by Byskov [6], if onecarries out imperfection sensitivityinvestigations, which involve an extension ofthe solution into the postbuckling responseregion, further mesh refinement may beneeded since the wavelength of the dominantlarge deformation pattern may often decreasesignificantly.
Finally, whenever one is engaged in shellstability analysis it is especially important thatone is aware of the possible detrimental effectsof a whole series of factors, that have beeninvestigated extensively in the late 1960s andthe early 1970s. Thus for an accurate andreliable prediction of the critical buckling loadof a real structure, one must account not onlyfor the influence of initial imperfections [e.g.,7,8] and of the boundary conditions [e.g. 9], butone must also consider the effects of stiffener
and load eccentricity [e.g., 10] and theprebuckling deformations caused by the edgerestraints [e.g. 11,12].
A test series of 7 isotropic shells carriedout by Arbocz and Babcock [1] at Caltech isused to illustrate how such a hierarchical
approach to buckling load calculations can becarried out. The platform for the multi-levelcomputations, needed for an accurate
prediction of the critical buckling loads and areliable estimation of their imperfectionsensitivity, is provided by DISDECO [13]. Withthis open ended, hierarchical, interactivecomputer code the user can access from his
workstation a succession of programs ofincreasing complexity.
SOLUTION OF THE BUCKLING PROBLEM
In the following it will be shown that with thehelp of DISDECO, the Delft Interactive ShellDEsign COde, the shell designer can study thebuckling behavior of a specified shell, calculateits critical buckling load quite accurately andmake a reliable prediction of the expecteddegree of imperfection sensitivity of the criticalbuckling load. The proposed procedureconsists of a hierarchical approach, where theanalyst proceeds step-by-step from the simpler(Level-I) methods used by the early
investigators to the more sophisticatedanalytical and numerical (Level-2 and Level-3)methods used presently.
Level-1 Perfect Shell Buckling Analysis
The geometric and material properties ofthe isotropic shell A-8 of Ref. [1] are listed inTable 1.
Table 1.
Geometric properties of Caitech isotropic shell A-8
ttotal(= h) =0.00464 in (=0.117856 mm )
L =8.0 in (= 203.2 mm )
R =4.0 in (= 101.6 mrn )
E = 15.2x106 psi (= 1.0480x105 N/mrn 2 )
v =0.3
D
Assuming a perfect shell(W=0) and the
following membrane prebuckling state
t
W (°) = hWv = hA12 Xc
F(O)= Eh 2 1;Ly 2cR 2
where
(1)
o _ Nx . Eh
O'cg Nc_ ' ¢rct _', Ncg = Gcgh
and c = 43(1 -v 2)
then the nonlinear equations governing theprebuckling state are identically satisfied andthe linearlzed stability equations reduce to aset of equations with constant coefficients. Ithas been shown in Ref. [14] that by assumingan asymmetric bifurcation mode of the form
W(1) x y= hsinm_Ecosn _
where
(2)
m = k = number of axial haft wavesn = e = number of circumferential full waves
one can reduce the solution of the linearizedstability equations to an algebraic eigenvalue
problem. Notice that the eigenvalue Zmn
depends on the wave numbers m and n. The
critical load parameter Zc is the lowest of all
possible eigenvalues. Thus finding ;Lc
involves a search over the integer valued wavenumbers m and n. Using the Level-1
2397
computational module AXBIF [14] a searchover integer valued axial half-wave numbers myielded the lowest eigenvalues listed in Table 2for the specified circumferential wave numbersn.
Table 2.
Buckling loads of the Caltech isotropic shell A-8Buckling load map for the perfect shellusing AXBIF [14] (Nc£ = - 49.51517 Ib/in)
n = 4 _m = 1.000 (m = 34)
n = 5 Zm = 1.000 (m = 34)
n = 6 _m = 1.000 (m = 34)
n = 7 _.m = 1.000 (m = 33)
n = 8 ;Lm = 1.000 (m = 33)
n : 9 zm: 1.ooo (m = I)
_.cm = 1.000 (m = 33)
n = 10 X m : 1.000 (rn = 33)
n = 11 _m = 1.000 (m = 33)
n = 12 Xcm = 1.000 (m = 32)
n : 13 ;Lcm = 1.008 (rn = 2)
;Lcm = 1.000 (m = 32)
n = 14 _m : 1.000 (m = 32)
n = 15 Z.crn = 1.001 (m : 3)
_n = 1.000 (m = 31 )
As is known the lowest eigenvalue is arepeated eigenvalue with a high degree ofmultiplicity. The family of modes belonging to
the lowest eigenvalue ;Lc = 1.000 lie on the so-
called Koiter circle [7].To facilitate the interpretation of the
numerical results obtained, DISDECO providesthe user with various graphical interfaces.
Thus the results of the search for the critical
(lowest) buckling load ;Lc can be displayed in
a contour map as shown in Fig. 1. In order toprovide a quick overview of the distribution ofeigenvalues, the values displayed in thecontour plot are normalized Thus in Fig. 1 thefollowing normalized eigenvalues are plotted
_.m : Nm-49.51517
Notice that the critical buckling load can becalculated using a simple multiplication
= =1.000 (-49.51517) = -49.51517 Ib/in
Level-2 Perfect Shell Buckling Analysls
To investigate the effects of edge constraintand of different boundary conditions on the
critical buckling load of the perfect shell
(0=0) one has to switch to the Level-2
computational module ANILISA [15]. In thismodule the axisymmetric prebuckling state isrepresented by
W (°) = hW v + hw o (x)
F(O) Eh 2. 14 2+R2fo(X)]
(3)
It has been shown in Ref. [15] that with theseassumptions the prebuckling problem isreduced to the solution of a single fourth order
ordinary differential equation with constantcoefficients, which always admits exponentialsolutions. Closed form solutions for simply
supported and clamped boundary conditionshave been published in the literature [16].
For isotropic shells the linearized stability
equations admit separable solutions of theform
W (1) = hWl(X)cosne
F(1) = E Rh2 fl(x)cosneC
where e = yR
(4)
Using a generalization of Stodola's method[17] first published by Cohen [18] the resultingnonlinear eigenvalue problem is reduced to asequence of linearized eigenvalue problems.The resulting ordinary differential equations aresolved numerically by a technique known as"parallel shooting over N-intervals" [19]. Noticethat by this approach the effect of edgerestraint and the specific boundary conditionsare satisfied rigorously. To find the critical load
parameter Zc an n-search must be carried
out, whereby one must be careful to find not alocal minimum but the absolute minimum. Ascan be seen from the results presented inTable 3 the n-search using membrane
prebuckling and a rigorous satisfaction of SS-3(N x = v = w = M x = 0) boundary conditions for
the stability problem yields indeed the repeated
eigenvalues Zcm -1.0 located on the "Koiter"
2398
circle. The slight variations in the eigenvaluesare due to the fact that both the axial half-wavenumbers m and the circumferential full wave
numbers n are integers.
Table 3
Buckling loads of the Caltech isotropic shell A-8
(Nc£ =-49.51517 Ib/in)
Buckling load map for the perfect shell using
ANILISA [15] (B.C. N x = v = w = M x = 0)
Prebuckling: Membrane Nonlinear
n =8 _Lm= 1.10977(m=1) _,n1=0.951155
zm = 1.00001(m=33) _nl= 0.966181*
n = 9 ;Lm = 1.00004(m=1) _nl= 0.934990
_._n = 1.00000(m=33) _Lcnl=0.957035*
n = 10 zm = 1.08160(m=1) _Lnl= 0.940209
= 1.00011(m=33) _,cnl= 0.946124"
n = 11 _ = 1.12539(m--2) _,_1= 0.931492*
_m = 1.00035(m=32) _nl= 0.939878
n = 12 _m= 1.01481(m=2) zcnl=0.915200 *
_,crn = 1.00004(m=32) _,nl= 0.931464
n = 13 _,m = 1.00836(m=2) _L_I= 0.908528*
Z,m = 1.00004(m=32) _.cnl= 0.918482
n = 14 _,cm= 1.04064(m=3) _,_1= 0.903256
_,cm= 1.00030(m=31) _,nl= 0.906963"
n = 15 Z,m = 1.00060(m=3) zcnl= 0.891368
;Lm = 1.00000(m=31) _nl= 0.898837*
n = 16 ;Lm = 1.02041(m=3) _,cnl= 0.885185
_,crn= 1.00023(m=31) ;Lnl= 0.886606"
n = 17 _,_-- 1.00086(m=4) _nl= 0.875881*
xm = 1.00001(m=30) _,nl= 0.878312
n = 16 _m = 1.01300(m=4) _,nl= 0.868406*
= 1.00027(m=30) _Lnl= 0.869200
n = 19 _,cm = 1.00014(m=5) z,nl= 0.861087
_m = 1.00001(m=29) _,cnl= 0.861626*
n = 20 ;Lm = 1.00164(m=6) xnl= 0.854891
Z,m = 1.00006(m=28) _,cnl= 0.855033*
Table 3 - continuation
Prebuckllng: Membrane
n = 21 )_m = 1.00378(m=7)
n = 22
n=23
n=24
n = 25
n =26
n=27
Nonlinear
2_nl = 0.849839*
;Lm = 1.00023(m=27) _,cnl= 0.849927
= 1.00183(m=7) _.cnl-- 0.846336"
2,,m= 1.00014(m=27) 2_,1= 0.846344
_m = 1.00114(m=8) _,_1= 0.844502
_L_ = 1.00012(m:26) _nl= 0.844511"
Z,_ = 1.00071(m=10) Z_I= 0.844480"
_.cm = 1.00011(m=24) _1= 0.844481
Z,m = 1.00000(m=11) 2,,nl= 0.846279*
_,crn = 1.00000(m=23) z,nl= 0.846280
;Lm= 1.00002(m=13) _,1=0.849833"
= 1.00001(m=21) _nl= 0.849833
2.,m = 1.00027(m=17) 2,,_1= 0.854990"
_,m = 1,00032(m=18) _,_I= 0.854990
* anti-sym at x = 1_/2.
The most accurate Level-2 solutions are
obtained when one employs a rigorous
nonlinear prebuckling analysis. As can be seenfrom the results listed in Table 3, for this
particular shell the critical buckling loads with
nonlinear prebuckling are always lower thanthe corresponding results obtained using a
membrane prebuckling analysis. Notice that
the critical load N c can be calculated easily bymultiplying the lowest eigenvalue ;Lc by thenormalizing factor Nc£ = -49.51517 Ib/in
yielding
Nc = 2,,cNc_ = -41.815 Ib/in (n = 24)
In fig. 2 the critical buckling modes usingmembrane and rigorous nonlinear prebucklingare depicted. Notice that the solutions withnonlinear prebuckling differ significantly fromthe ones obtained using membraneprebuckling, especially at n = 24 where oneobserves a typical edge buckling typebehavior.
Level-3 Perfect Shell Buckllng Analysis
To verify the earlier predictions the finitedifference version [20] of the well known shell
2399
analysis code STAGS [21] will be used. Inorder to be able to represent the measuredinitial imperfections accurately, the whole shellwill be modeled.
Initially a convergence study must becarried out in order to establish the mesh sizeneeded for accurate modeling of the bucklingbehavior of the shell in question. For this
purpose the asymmetric bifurcation from anonlinear prebuckling path option was used,whereby the earlier results obtained with theLevel-2 module ANILISA listed in Table 3serve as a reference.
In the convergence study, at first, for a fixednumber of mesh points in the axial direction(NR = 161) the number of mesh points in thecircumferential direction (NC) was increaseduntil the bifurcation load approached a
horizontal tangent. As can be seen from Fig. 3the results converge to a limiting value frombelow at about NC = 261. Next, for a fixed
number of mesh points in the circumferentialdirection (NC = 161) the number of rows (NR)was varied. This time convergence is fromabove and as can be seen from Fig. 3 the
horizontal tangent is reached at about NR =261.
To illustrate the difference between usingcoarser meshes to speed up the computationsand the finer meshes which produce (nearly)
converged solutions, initially the first threeeigenvalues and eigenvectors are computed.
Using a mesh of 161 rows and 181 columns(a model with 89488 D.O.F.'s and a maximumsemi-bandwidth of 713) and SS-3 boundary
conditions (N x = -N o, v = W = Mx = 0) the
following 3 lowest eigenvalues were obtained:
_,(c1) = 0.833361 (n = 26) --->
N(c1) = Z(cl)Nc£ = -41.264 Ib/in
Z(2) = 0.833374 (n = 26) -->
N (2) = _,(2)Nct =- 41,265 Ib/in
_(3) = 0.834238 (n = 25) --->
N (3) = _,(3)Nct= -41.307 Ib/in
Z(c1) = 0.844993 (n = 24) -->
N(c1 : =-41.840Ib n
;L(2) = 0.845006 (n = 24) --_
N(c2) = L(c2)Nc_, =- 41.84lib/in
_.(3) = 0.845100 (n = 25)
N(3) = Z(c3)Nct = -41.845 Ib/in
Details of the critical mode are displayed in
Figures 4-5 for the coarser 161x181 model andin Figures 6-7 for the finer 261-261 model.Notice that using the more refined 261x261model the sequence of the 3 lowest bucklingloads and the corresponding buckling modes
agree closely with the predictions obtained withthe Level-2 module ANILISA (see also Table 3)
IMPERFECTION SENSITIVITY STUDY
That initial imperfections may decrease theload carrying capacity of thin-walled shellstructures is by now widely known and
accepted. However, in order to calculate theeffect of initial imperfections one must knowtheir shape and amplitude, an information that
is rarely available.
In the absence of initial imperfection
measurements, as a first step one mustestablish whether a given shell-loading
combination is imperfection sensitive, and ifthe answer is positive to estimate how
damaging certain characteristic imperfection
shapes are.
Single Axisymmetric Imperfection
Based on Koiter's pioneering work on the
effect of initial imperfections [7,22] the simplestimperfection model consists of a singleaxisymmetric imperfection
_' = h_l cosi_x (5)L
Next these computations were repeated usinga mesh of 261 rows and 261 columns (a modelwith 207508 D.O.F.'s and a maximum semi-bandwidth of 1037) and the same SS-3
boundary conditions yielding the followingresults
where i is an integer denoting the number of
half-waves in the axial direction and _1 is the
amplitude of the axisymmetric imperfectionnormalized by the shell waU-thickness h.
If one assumes that both the axial load and the
boundary conditions are independent of thecircumferential coordinate, then the prebuck-
2400
ling solution will also be axisymmetric, a factthat simplifies the solution considerably.
Level-1 Analysis of AxisymmetricImperfection
Neglecting the effect of the prebucklingboundary conditions the nonlinear equationsgoverning the prebuckling state admit thefollowing axisymmetric solutions
W (°) =hW v +hwo(x)
F(o)= Eh 2 l_,y2+fo(x )cR 2
where
(6)
w , Xwo(x) = hE1cos i_
fo (x) =Z,Ci - _,
(1+ B210r'2) Eh 3 - x
2C_2_2 C F_lCOSi_L
(7)
1 -- (1+ B21°_2)2
Notice that the linearized stability equationsbecome now a set of equations with variablecoefficients. The reduced wave number oci
and the normalized stiffness coefficients A22,
B21 and 5_ 1 are all listed in Ref. [14].
It has been shown in Ref. [14] that byassuming an asymmetric bifurcation mode ofthe form
W (1)= hslnm=LcosnY (8)
whe re
m = k = number of axial half wavesn = e -- number of circumferential full waves
a Galerkin type approximate solution yields for
the eigenvalues (read, buckling loads) _. of theproblem a characteristic equation in the form ofa cubic polynomial
_3 _ (Zmn.c + 2_.c i _ (_1_18i=2m);L2 (9)
+{2Zmn'c + _'c i + (C2 -C1)_lai=2m}Zci _"
^ -2 2-{Zmn'c + C2_18i=2m + ((_3 - C4_im )_1 }Zci = 0
where
2401
5i=2m = 1
=0
if i=2m
otherwise
5im=1 ifi=m
= 0 otherwise
and the constants C1,(_2 .... are listed in
Appendix C of Ref. [14].Here it must be remembered that one will
only get any noticeable degrading influence of
the assumed axisymmetric imperfection if _1
is negative and if the coupling condition i=2mis satisfied. The physical explanation for thiscan be found in Koiter's 1963 paper [22].Furthermore, in order to obtain the smallest
real root of Eq. (9), for a given axisymmetric
imperfection _1 an n-search must be carried
out. It should also be noticed that the terms
involving the Kronecker delta _i=2m are all
linear in _1, and thus they dominate the buck-
ling behavior of the shell with axisymmetricimperfection.
Assuming that the most likely axisymmetdcimperfection of the silver pointed wax mandrelused to electroplate the isotropic shell A-8 isgiven by
= I_ 1 COS2_ L (10)
the Level-1 DISDECO computational moduleAXBIF generated the solid curve shown in Fig.8. Notice that the curve is normalized by
Nct =-49.515171b/in, the classical buckling
load of an isotropic shell, obtained around1910 by Lorenz [23], Timoshenko [24] andSouthwell [25]. Notice also that an initialimperfection amplitude equal to the wall
thickness of the shell (_1 = -1.0) generates a
"knockdown factor" of Zc = 0.706, resulting in
the following buckling load
N c = ;LcNct =
0.706(-49.51517) = -34.958 Ib/in
However, if one assumes an imperfection inthe form of the classic axisymmetric bucklingmode of Koiter [7]
_/= h_l cosict_ L
where
L 2/ C-. 2)
then for the isotropic shell A-8 ict - 34.0, and
the Level-1 DISDECO computational moduleAXBIF found that the minimum buckling load
1occurred when m = -_ic,e for different values of
n (which depended on the value of _1 ). These
results are shown in Fig. 9.For this case an initial imperfection amplitudeequal to the wall thickness of the shell
(_1 =-1.0) generates a "knockdown factor" of
;Lc = 0.196 resulting in the following rather low
buckling load
Nc= =0.196(-49.51517) = -9.705 Ib/in
The last two buckling loads clearlydemonstrate the importance of the shape ofthe initial imperfections involved and the role ofthe corresponding eigenvalues. Using the long
wave axisymmetric imperfection (i=2) the
corresponding normalized axisymmetricbuckling load (eigenvalue)of the perfect shell is
;Lci =144.321 (a very high value) yielding a
low imperfection sensitivity. For the short waveimperfection (i = 34) the corresponding
normalized axisymmetric buckling toad
(eigenvalue) of the perfect shell is
;Lci =1.000. Thus it lies on Koiter's circle
yielding a high imperfection sensitivity.
Level-2 Analysla of AxlsymmetricImperfection
Since the external loading, the boundaryconditions and the assumed initial imperfection
are axisymmetric, therefore the prebucklingsolution will also be axisymmetric. It has beenshown in Ref. [26] that by assuming
W (°) = hWv + hwo(X)
F(o) Eh 2. 1_ 2+R2fo(x)}
(11)
the solution of the nonlinear partial differential
equations governing the prebuckling state canbe reduced to the solution of a single fourthorder ordinary differential equation withconstant coefficients, which can be solved
routinely.For isotropic shells the resulting linearized
stability equations admit separable solutions ofthe form
W (1) = h Wl(X)cosne
F(I)= ERh2 f1(x)cosnOc
Ywhere e =--,
R
(12)
Solution proceeds as outlined on Ref. [26].Using an updated version of the Level-2computational module MANILISA [27] and SS-3 (N x -- v = w = M x = 0) boundary conditions
one obtains the results presented in Table 4.Notice that a rigorous nonlinear prebuckling
analysis was used and an n-search wascarried out for each specified axisymmetric
imperfection amplitude _1"
The values of Table 4 are plotted as the
dashed curve in Fig. 9. A comparison of theresults obtained via the Level-1 module AXBIF
(solid curve) and the Level-2 moduleMANILISA (dashed curve) shows that also inthe case of axisymmetric imperfections arigorous pre-buckling analysis should be used,Especially for very small initial imperfection
(r_ll < 0.1) the Level-1 predictionsamplitudes/ w
are inaccurate and overestimate the critical
buckling load. Notice further that both curveshave been normalized by classical buckling
load Nct = -49.51517 Ib/in. This way the effect
of using a rigorous prebuckling analysisbecomes easily discernible.
Table 4Buckling loads of the Caltech isotropic shell A-8
(Nc£ = -49.51517 Ib/in)
Short wave axisyrnmetric imperfection
4 x_//h =_1cos3 lz_ using MANILISA [27]
(B.C.: N x = v = w =M x = 0)
0.
- 0.01
- 0.02
- 0.03
- 0.04
- 0.05
- 0.06
- 0.07
- 0.08
- 0.09
0.844480 (n=24)
0.847663 (n=21)
0.786079 (n=21)
0.740449 (n=21)
0.703421 (n=21)
0.671958 (n=21)
0.644481 (n=21)
0.620047 (n=21)
0.598039 (n=21)
0.578030 (n=21)
_1 _nl
- 0.1 0.549168 (n=17)
- 0.2 0.420091 (n=17)
- 0.3 0.354645 (n=19)
- 0.4 0.305484 (n=18)
- 0.5 0.271507 (n=17)
- 0.6 0.247002 (n=16)
- 0.7 0.229213 (n=15)
- 0.8 0.214064 (n=20)
- 0.9 0.200103 (n=19)
- 1.0 0.189011 (n=18)
2402
Level-3 Analysis of AxlsymmetrlcImperfection
Recalling that since both the axial load andthe boundary conditions are independent of thecircumferential coordinate, therefore theprebuckling solution will also be axisymmetric,one can use once again the asymmetricbifurcation from a nonlinear prebuckling pathoption. By modeling the full shell the code canchoose itself the critical number of full waves inthe circumferential direction. No n-search must
be carried out. Using a uniformly spaced meshof 161 rows and 181 columns and the user
written subroutine option WlMP to introducethe following axisymmetric imperfection
= 1.0hcos 2_ xL
(remember STAGS defines W positiveoutward) the following critical bifurcation loadwas found
Nc = -32.582 Ib/in
As can be seen from Fig. 10 the criticalbuckling mode has n = 9 full waves in the
circumferential direction and a single half wavein the axial direction. The nondimensional
bifurcation load of the shell with axisymmetric
imperfection is for _1 = 1.0
Lc _ Nc _ -32.582 _0.65802Nc£ -49.51517
Notice also that the Level-2 MANILISA
prediction (_,nl = 0.683, n = 9) agrees closely
with the Level-3 STAGS-A prediction
(_.nl =0.658, n = 9). The slight difference is
partly due to the fact that MANILISA uses the
Donnell type nonlinear shell equations,
whereas STAGS-A employs the higher order
Marlowe-FI0gge equations.
Single Asymmetric Imperfectlon
The effect of a single asymmetric initialimperfection can be investigated either bysolving the full nonlinear response problem orby employing the well known Lyapunov-Schmidt-Koiter [7] reduction technique. Wheninvestigating the degrading effect of a singlemode asymmetric imperfection
• x yW = t_2 slnmnEcosn _- (13)
where m and n are integers denoting thenumber of axial half-waves and the number of
2403
circumferential full waves, respectively,instability occurs at the limit point of theprebuckling state in the generalized load-deformation space. Assuming that theeigenvalue problem for the critical (lowest)buckling load Ac will yield a unique asym-
metric buckling mode W (1), then for an
imperfect shell (_2 #0) the shape of the
generalized load-deflection curve in the vicinityof the bifurcation point A = Ac is given by the
following asymptotic expansion
(A - A c )_ = AcaP_2 + Acb_ 3 + ...
- -(^ - ^c + o( 2)(14)
Expressions for the postbuckling coefficients=a" and =b" and the imperfection forms factors"(x" and "13" are derived in References [28,29].If the limit point is close to the bifurcation point,
then the maximum load A s that the structurecan carry prior to buckling can be evaluatedfrom Eq. (14) by maximizing A with respect toF_. For cases where the first postbucklingcoefficient "a" is zero, this analysis yields themodified Koiter formula [29]
(1-Ps) 3'2 = 3jE Yb[1-- (1-p_ )l 21(15)
where Ps = As/Ac.
Notice that, if the second postbucklingcoefficient "b" is positive, Eq. (15) has no realsolutions. Thus the buckling load of thespecified shell-loading combination is not
sensitive to small asymmetric initialimperfections of the shape given by Eq. (13).If, however, the second postbucklingcoefficient "b" is negative, the equilibrium loadA decreases following buckling and thebuckling load of the real structure As issensitive to the asymmetric initial imperfectionspecified by Eq. (13).
Level-1 Analysis of AsymmetricImperfection
For the isotropic shell under investigation,
as can be seen from the partial results listed inTable 2, there are many identical eigenvalues.
Hence, strictly speaking, the proposed form of
the perturbation expansion given by Eqs. (14)is not applicable, since the nonlinear
interaction between the many nearlysimultaneous eigenmodes is not accounted for.
Thus the following results, where one
considers the eigenfunctions corresponding tocertain critical eigenvalues chosen one at the
time,canat bestgivean indicationas to theseverity of the expected imperfectionsensitivity.
Initially, we assume a long waveasymmetricimperfectionaffineto oneof thecriticalbucklingmodesof theperfectCaltechisotropicshellA-8,computedby the Level-1computationalmoduleAXBIF(seealso Table
3)
= hE 2 sinTtXc°s9YL R
(16)
Next using the Level-1 computational moduleBFACT to carry out the initial postbucklinganalysis yields the following results
Kc = Km = 1.00004 (m = 1, n = 9)
b = - 0.010636 a = [_= 1.0
Substituting these values into Eq. (15), onecan plot the degrading effect of an asymmetricimperfection of the shape given by Eq. (16) as
a function of its amplitude _2. As can be seen
from Fig. 11 an initial imperfection amplitudeequal to the wall thickness of the shell
(_2 = 1.0) generates a "knockdown factor" of
Xs =0.679, resulting in the following low
buckling load
Ns = XsNct =
0.679 (-49.51517) = -33.621 Ib/in
Notice that the imperfection form factors
"a" and "13" are identical equal to 1.0 because
BFACT uses membrane prebuckling to
calculate the necessary first and second orderfields and the assumed asymmetric
imperfection shape of Eq. (16) is affine to thebuckling mode. If, however, one assumes thefollowing short-wave asymmetric imperfection
V y_/= hT.,2sin 17_" cos 27L R
which lies also on the Koiter circle, then the
initial postbuckling analysis, done once againby the Level-1 computational module BFACT,yields the following results
Xc= X_ = 1.00025(m =17,n = 27)
b = -220.84 a = 13= 1.0
Using these values in Eq. (15) one obtains Fig.12. As can be seen, an initial imperfection
amplitude equal to the shell wall thickness
(T..,2= 1.0) this time generates a =knockdown
factor" of Ks =0.0249, resulting in an
unrealistically low buckling load of
Ns = KsNct =
0.0249(-49.51517) = -1.233 Ib/in
Here one obviously cannot rely on the
prediction of the Level-1 computational moduleBFACT, which uses membrane prebuckling
analysis. In such a situation, as has beenshown by Hutchinson and Frauenthal [30], onemust repeat the postbuckling analysis using arigorous prebuckling solution which includesthe nonlinear effects due to the edge restraints.
Level-2 Analysis of AsymmetricImperfection
To investigate the effects of edge-constraintand/or different boundary conditions on the
imperfection sensitivity of the critical bucklingload of the Caltech isotropic shell A-8 one hasto switch to the Level-2 module ANILISA [15]
and run its postbuckling analysis option. In thismodule, as described eartier, the axisymmetric
prebuckling state is represented by Eqs. (3),the buckling modes by Eqs. (4) and the
postbuckling state by
W (2) = h[wcx(x)+ wf_(x)cosne]
ERh 2F(2) = _ [fez(x) + f13( x ) cos 2n0]
c
(18)
where 8 = y / R. Details of the computational
procedures used are reported in Refs. [15,26].Initially, let us assume that the specified
asymmetric imperfection is given by Eq. (16)
_ = h_2 sin=Xcos9eL
where (_= y / R . Running ANILISA with rigo-
rous prebuckling and SS-3 boundaryconditions (Nx = -No, v = w = Mx = 0) yields
the following results
Xc = Knl = 0.934990(n = 9)
b =-0.016202; _= 0.20830; 6 =-0.38232
Using Eq. (15) to plot the degrading effectof the asymmetric imperfection specified by
Eq. (16) as a function of its amplitude _2 one
2404
obtains the results displayed in Fig. 13 as asolid line.
Here one must remember that Koiter's
Sensitivity Theory is asymptotically exact, thatis, it yields accurate predictions for sufficientlysmall imperfections, whereby what issufficiently small may vary from case to case.Also, Eq. (15) was obtained by using theperturbation expansion given by Eq. (14),
where terms of order (_) are neglected. As
can be seen from the dotted curve plotted inFig. 13, by using more advancedcomputational modules such as COLLAPSE[31], where a full nonlinear solution is used and
terms up to and including order (_2) are
kept, one obtains slightly lower predictions.
Notice that up to about _2 =0.1 the
asymptotic predictions from ANILISA and thenonlinear results of COLLAPSE agree veryclosely. Thus one can say that in this case therange of validity of the asymptotic solution is
0>_2 _0.1.
For an initial imperfection amplitude equal
to the wall thickness of the shell (T_.,2= 1.0) the
asymptotic solution yields a =knockdown factor"
of Zs =0.806, which corresponds to a
buckling load of
N s = ZsNct =
0.806(-49.51517) = -39.909 Ib/in
For the same initial imperfection amplitude
of _2 =1.0 the nonlinear solution using
COLLAPSE yields a =knockdown factor" of
Xs = 0.68, which corresponds to a bucklingload of
Ns = ;LsNct =
0.68(-49.51517) = -33.670 Ib/in
Let us next consider the short wave
asymmetric imperfection given by
_/= hT.._2sin 17_Xcos 240 (19)L
where e= y/R. The rigorous asymptotic
solution obtained by running ANILISA with SS-3 boundary conditions (N x = -N o,
v=w=M x =0) yields now the followingresults
Zc = _nl = 0.844480 (n = 24)
b = --0.50329; a = 0.20113; _ = -0.012022
Using these values with Eq. (15) yields thesolid curve plotted in Fig. 14. Rerunning thesame case with the nonlinear module
COLLAPSE yields the dotted curve plotted inFig. 14.
Notice that in this case the asymptoticpredictions from ANILISA and the nonlinearresults from COLLAPSE agree closely up to
about _ = 0.03. Thus the range of validity of
the asymptotic solution is only 0 > _ ___0.03.
For an initial imperfection amplitude of
T__=1.0 the asymptotic solution yields a
"knockdown factor" of _s = 0.400, which
corresponds to a buckling load of
Ns = ZsNcp =
0.400(-49.51517) = -19.806 Ib/in
For the same initial imperfection amplitude of
F._ =1.0 the nonlinear solution yields a
=knockdown factor" of Xs =0.30, which
corresponds to a buckling load of
N s = ;LsNcf =
0.30(-49.51517) = -14.855 Ib/in
For an initial imperfection amplitude equal
to the wall thickness of the shell (T.,2= 1.0) the
asymptotic solution yields a =knockdown factor"
of _s =0.806, which corresponds to a
buckling load of
Ns = ZsNct =
0.806(-49.51517) = -39.909 Ib/in
MEASURED INITIAL IMPERFECTIONS
In order to apply the theory of imperfectionsensitivity with confidence, one must know thetype of imperfections that occur in practice. In1969 Arbocz and Babcock [1] published theresults of buckling experiments where, for thefirst time, the actual initial imperfections andthe prebuckling growth of the midsurface ofelectroplated isotropic shells were carefullymeasured and recorded by means of anautomated scanning mechanism.
Mideurface Initial Imperfections
As can be seen from Fig. 15, the measuredinitial midsurface imperfections of shell A-8 [1]show a rather general distribution dominated
24O5
by an n=2 mode. One can use the followingdouble Fourier series
_/ 7 _ . x 18-= EW,ocO,,,c+E. o,CO ,-hi=1 I=1
(20)
19 99
_=1 k,_=l
115x y
+k,t=l
to represent the measured initial imperfectionsaccurately, where Fourier coefficients withabsolute values less than 0.001 are neglected.
A convenient measure of the size of the
initial imperfections is their root-mean-square(RMS) value. By definition
2_R L
A2 s = 12 R---CJ" (x'y12dxdy 121 o o
Thus
h =_.z_ ioI k,_
2 2 (22)= Aax i + Aasy
For the A-8 shell the RMS values are
Z_axi = 0.5077 ; Aasy = 0.6882
Using these measured RMS values with thepreviously discussed initial imperfectionmodels yields the following buckling loadpredictions:
For the long wave
_/= h_l cos2_ Laxisymmetric imperfection
if _1 = Aaxi = --0.5077 then from
MANILISA [27] Xnl = 0.824 (n = 9)
pcnl = 2_R2_nlNct = -1025.428 Ibs
For the short wave
_./= h_l cos 34_ Laxisymmetric imperfection
if _1 = Aaxi = -0.5077 then from
MANILISA [27] _1 = 0.266 (n = 27)
pnl= 27¢R;_lNct = -331.024 Ibs
Turning now to the long wave asymmetric
imperfection _/= hT--.2sin_Xcos9 yL R
if _ = Aasy = 0.6882 then from
COLLAPSE [31] X_ I = 0.740 (n = 9)
pnl = 2_RZ_.INcl = _920.894 Ibs
For the short wave asymmetricx y
imperfection _/= h_2 sin17_cos24_
if _ = Aasy = 0.6882 then from
COLLAPSE [31] _1 = 0.324 (n = 24)
p;'= =403202bs
Comparing these buckling load predictions withthe experimental buckling load of
Pexp = -782"71bs it is clear that the long wave
imperfection models yield an upper bound,whereas the short wave imperfection models
yield a lower bound.For a more accurate estimate of the
buckling load one must use the measuredinitial imperfections in codes like STAGS
[20,21] to carry out 2-dimensional nonlinearcollapse analysis. Employing a user wdttensubroutine WlMP to input the double Fourierseries of Eq. (20), the 161x181 STAGS modelyielded with SS-3 boundary condition
(N x = -N o, v = w = M x = 0) a collapse load of
Ps =-900-1031bs' whereas the same model
with C-4 boundary condition (u = uo ,
v=w=w, x=0) yielded a collapse load of
Ps = -976.4671bs" Both of these values are
significantly higher than the experimental
buckling load of Pexp = -782-71bs •
Boundary Initial Imperfections
In References [32,33] it was shown that
boundary imperfections can have a significantdegrading effect on the buckling loads ofaxially compressed cylindrical shells. Theflatness of the end ring, attached to the load-cell of the Caltech test set-up shown in Fig. 16,
was measured about 20 years after it was
2406
used to test shell A-8. As can be seen in Fig.17 there is a very large amplitude waviness ofthe end support, with maximum deviations ofabout + 3 wall thicknesses.
The measured boundary imperfections aredecomposed in a 1-dimensional Fourier series
Uo = ub(Y) = (23)
_bhflao + n___l(ancosnY + bn sinnY) t
This series then is used in STAGS [20] tomodel the effect of boundary initialimperfections using a modified C-4 boundarycondition
U=Ub(Y) , v=w=W,x=0 (24)
and taking advantage of the dual loadingsystems provided by the program.
Because the grooves of the end rings werefilled with liquid Cerrolow at the time shell A-8was installed in the test set-up, it is to beexpected that after cooling off the hardenedliquid metal has filled-in all the gaps. The effectof this unknown end support was modeled byvarying the amplitude of the boundary
imperfection T_ between 0 and 0.1. As can be
seen from the results displayed in Fig. 18,including both the measured initial midsurfaceimperfections of Eq. (20) and the measuredboundary imperfections of Eq. (23) with an
amplitude of T.,b-0.035, the calculated
collapse load agrees well with the experimentalbuckling load.
DISCUSSION OF THE RESULTS
When comparing and analyzing the resultsobtained sofar it is important to keep in mindthat all Level-1 and Level-2 solutions are
based on approximate representations of theunknown functions. As pointed out in theprevious sections Level-1 solutions use asingle term double Fourier seriesapproximation to reduce the solution of thestability problem, formulated in terms of partialdifferential equations, to algebraic eigenvalueproblems. The effect of edge restraint isneglected (one uses a membrane prebucklingsolution) and the assumed field functions
satisfy approximately SS-3 (N x = -N o,
v = w = M x = 0) boundary conditions.
Level-2 solutions eliminate the y-dependence by a truncated Fourier
decomposition in the circumferential direction.
The resulting system of nonlinear ordinarydifferential equations are solved numerically,whereby both the specified boundaryconditions and the effect of edge restraint arerigorously satisfied. Thus by this approach theonly approximation is that one represents thevariation of the solution in the circumferential
direction by a single harmonic with n fullwaves, whereby an n-search is used toestablish which wave number is the criticalone. The Level-2 module ANILISA can also be
used to investigate the effect of using differentboundary conditions. In Table 5, the results forfour different boundary conditions are
presented. As expected the fully clamped C4boundary conditions has the highest criticalbuckling load. The increase in load carryingcapacity with respect to the weaker SS3boundary conditions is about the same as foran isotropic shell of similar characteristicdimensions (same L/R and PJt ratios) of Ref.[34].
The Level-3 solutions are based either on a2-dimensional finite difference or finite element
formulation. In both cases, if one uses theappropriate meshes, one can obtain rigoroussolutions where all nonlinear effects are
properly accounted for. The only real problemwith Level-3 type solutions is that for eachproblem one must establish the appropriatemesh size. Coarse meshes yield inaccuratesolutions. What is coarse depends on theparticular problem under investigation. Thus,for a general nonlinear solution a convergencestudy must always be carried out.
Using a hierarchical simulation platform
such as DISDECO (Delft Interactive Shell
DEsign Code), where the analyst has at hisdisposal computational modules of differentlevel of sophistication, such a convergence
study can be carried out relatively quickly andaccurately. In Table 5 a summary of the results
obtained in this study is presented usingnormalized variables.
Table 5
Summary of buckling load calculations of theCaltech isotropic shell A-8
(Nct= -49.51517 Ib/in)
Perfect shell analysis
Level-1 Membrane Prebuckting -Repeated eigenvalues with high multiplicity
Lm = 1.000 ---> Pc = -1244.5 Ibs
2407
Level-2 Nonlinear Prebuckling
SS-3 2_,I = 0.844480(n = 24)
--> Pc = -1050.9 Ibs
SS-4 _1 = 0.866974(n = 25)
"-> Pc = -1078.9 Ibs
C-3 Z,nl = 0.910305(n = 24)
Pc = -1132.8 Ibs
C-4 _1 = 0.926871(n = 25)
--> Pc = -1153.4 Ibs
Level-3 SS-3 B.C.
161x181 Model
Z_ I = 0.833361(n = 26)
-> Pc =-1037'11bs
261 x261 Model
2_,I = 0.844993(n = 24)
--) Pc = -1051.6 Ibs
Imperfect Shell Analysis
Long wave axisymmetric imperfection -
_1 = Aaxi = -0.5077
Level-2 ;Lnl = 0.824(n = 9)
---> Pc = -1025.4 Ibs
Short wave axisymmetric imperfection -
_1 = Aaxi = -0.5077
Level-2 _1 = 0.266(n = 27)
--> Pc = -331.0 Ibs
Long wave asymmetric imperfection -
_2 = Aasy = 0.6882
Level-2 Knl = 0.740(n = 9)
"--> Ps = -920-91bs
Short wave asymmetric imperfection -
_2 = Aasy = 0.6882
Level-2 ;k,nl = 0.324(n -- 24)
---> Ps = -403.21bs
Measured midsurface initial imperfection -see Eq. (20)
Level-3 SS-3 B.C. --> Ps = -900-11bs
Level-3 C-4 B.C. '-> Ps = -976"51bs
Measured midsurface + boundary (T--=b = 0.035)
initial imperfection - see Eqs. (20) and (23)
Level-3 Modified C-4 B.C.
(see Eq. (24)) --_ Ps =-780.01bs
Experimental Buckling Load
--> Pexp = -782.7 Ibs
Notice that by including both the midsurfaceand the boundary initial imperfections oneobtained a very good agreement indeed.
The study of the buckling behavior of theisotropic shell A-8 was undertaken with theobjective of establishing a rational analysisprocedure for incorporation of initialimperfections into the design of buckling criticalstructures. For a shell of revolution under
axially symmetric loading there are thefollowing four items necessary for theapplication of the proposed procedure:
a. Buckling analysis capability for shells ofrevolution under axially symmetric loading.
(A computer code such as SRA [18] orBOSOR [35] will satisfy this requirement.)
b. Imperfection sensitivity analysis capabilityfor shells of revolution under axially
symmetric loading. (A computer code suchas SRA [18] or STASOR [36] wilt satisfy thisrequirement.)
c. Geometrically nonlinear analysis capabilityfor a shell of revolution, which can handle
the asymmetry introduced by the generalasymmetric initial imperfections. (TheSTAGS [20,21] computer code will satisfythis requirement.)
d. A design imperfection.At the present time this is the most elusiveitem.
The imperfection model introduced byImbert [37], following an idea by Donnell andWan [38], is one approach, but has yet to beextended to larger cylindrical shell structuresand other shells of revolution.
coN_q _C.ML gt
By relying on a series of theoretical results ofvarious degree of sophistication published inthe literature, the hierarchical approach used inthis paper has resulted in a series of bucklingload predictions of increasing accuracy. It wasshown that in order to be able to arrive at areliable prediction of the critical buckling load
2408
and to make an estimate of its imperfectionsensitivity which can be used with confidence,
one must proceed step by step from simple tomore complex models and solution proce-dures.
In particular one can state, that in order topredict the critical buckling load accurately andto make a reliable estimate of its imperfectionsensitivity, the nonlinear effects caused by theedge restraint conditions must be included inthe analysis. Any solution procedure whichfails to account for these effects, should be
suspect of having provided incorrect results.The most approximate of the here
described analyses, the Level-1 solutionswhich neglect the effects caused by the edgerestraints, can still be used to great advantageto establish the approximate behavior of a shellsubjected to the specified external loading.However, depending on the value of theprebuckling stiffness, resulting from thedifferent types of wall constructions used, thesolutions may be either conservative ornonconservative.
As can be seen from the results shown in
Table 5, the buckling load of the isotropic shellA-8 is sensitive to all the initial imperfectionshapes investigated. For a more specificprediction of the final collapse load, the finalgoal of a "High Fidelity Buckling LoadAnalysis", one has to carry out a refined Level-3 analysis including measured values of all thesignificant generalized imperfections such asthe traditional shell-wall imperfections,variations in the shell-end or loading surfacegeometry and especially for composite shellsvariations in the shell-wall thicknessdistribution. It has been shown in Ref. 33 that
such an approach yields very good agreementbetween the predicted collapse toad and theexperimental buckling load.
ACKNOWLEDGEMENT
The research reported in this paper wassupported in part by NASA Grant NAG 1-2129.This aid is gratefully acknowledged.
1.
2.
F__=E.EBF=B.C_E_
Arbocz, J. and Babcock, C.D. Jr., '_l'he
Effect of General Imperfections on theBuckling of Cylindrical Shells," J. Appl.Mech., Vol. 36, 1969, pp. 28-38.Arbocz, J., Starnes, J.H. and Nemeth,M.P., "A Hierarchical Approach toBuckling Load Calculations," Proceedings40th AIAAJASM F_JASCE/AH S/ASC
Structures, Structural Dynamics andMaterials Conference, April 12-15, 1999,St. Louis, MO, pp. 289-299.
3. Anonymous, NASTRAN, The MacNeal-Schwendler Corporation, Los Angeles,California.
4. Anonymous, ABAQUS, Hibbitt, Karlsson& Sorensen, Inc., Providence, RhodeIsland.
5. Zienkiewicz, O.C., "The Finite ElementMethod," (The third, expanded andrevised edition of The Finite Element
Method in Engineering Science),McGraw-Hill Book Company (UK) Limited,London, 1977, p. 504.
6. Byskov, E., "Smooth PostbucklingStresses by a Modified Finite ElementMethod," DCAMM Report No. 380,Technical University of Denmark, Lyngby,1988.
7. Koiter, W.T., "On the Stability of ElasticEquilibrium," Ph.D. Thesis (in Dutch), TH-Delft, The Netherlands, H.J. Paris,Amsterdam, 1945, English translationNASA TTF-10, 1967, pp. 1-833.
8. Budiansky, B. and Hutchinson, J.W.,"Dynamic Buckling of ImperfectionSensitive Structures," Proceedings 11thIUTAM Congress, Munich, 1964, JuliusSpringer Verlag, Berlin, 1966, pp. 636-651.
9. Hoff, N.J., "Buckling of Thin Shells,"Proceedings of an Aerospace Symposiumof Distinguished Lecturers in Honor ofTheodore von Karm._n on his 80th
Anniversary, Institute of AerospaceSciences, New York, 1961, pp. 1-42.
10. Stuhlman, C.E., De Luzio, A. and Almroth,B., =Influence of Stiffener Eccentricity andEnd Moment on Stability of Cylinders inCompression," AIAA Journal, Vol. 4, No.5, May 1966, pp. 872-877.
11. Fischer, G., "0ber den Einfluss dergelenkingen Lagerung auf die Stabilit_td0nnwandiger Kreiszylinderschalen unterAxiallast und Innendruck," Z.f. Flug-wissenschaften, Vol. 11, 1963, pp. 111-119.
12. Stein, M., The Influence of PrebucklingDeformations and Stresses on the
Buckling of Perfect Cylinders," NASA TR-190, February 1964.
13. Arbocz, J. and Hol, J.M.A.M., =ShellStability Analysis in a Computer-AidedEngineering (CAE) Environment,"Proceedings 34th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamicsand Materials Conference, April 19-22,1993, La Jolla, California, pp. 300-314.
14. Arbocz, J., '_he Effect of Initial
Imperfections on Shell Stability AnUpdated Review," Report LR-695, DelftUniversity of Technology, Faculty ofAerospace Engineering, The Netherlands,September 1992.
2409
15. Arbocz, J. and Hol, J.M.A.M., "Koiter's
Stability Theory in a Computer AidedEngineering (CAE) Environment," Int. J.Solids and Structures, Vol. 26, No. 9/10,
pp. 945-973, 1990.16. Booton, M., "Buckling of Imperfect
Anisotropic Cylinders under CombinedLoading," UTIAS Report No. 203,University of Toronto, 1976.
17. Von K&rm,_n, T. and Blot, M.A.,"Mathematical Methods in Engineering,"McGraw-Hill Book Co., New York, London,
1940, p. 317.18. Cohen, G.A., "Computer Analysis of
Asymmetric Buckling of Ring-stiffenedOrthotropic Shells of Revolution," AIAAJournal, Vol. 6, No. 1, January 1968, pp.141-149.
19. Keller, H., "Numerical Methods of Two-Point Boundary Value Problems," BlaisdellPublishing Co., Waltham, Mass., 1968.
20. Almroth, B.O., Brogan, F.A., Miller, E.,Zele, F. and Peterson, H.T., "CollapseAnalysis for Shells of General Shape; I1.User's Manual for the STAGS-A Computer
Code," Technical Report AFFDL-TR-71-8,Air Force Flight Dynamics Laboratory,Wright-Patterson Air Force Base, Ohio,March 1973.
21. Brogan, F.A., Rankin, C C. and Cabiness,H.D., "STags Access Routines-STARReference Manual," Report LMSCP032595, Version 2.0, Lockheed Palo AltoResearch Laboratory, Palo Alto, California,June 1994.
22. Koiter, W.T., '`The Effect of AxisymmetricImperfections on the Buckling of CylindricalShells under Axial Compression," Koninkl.Ned. Akad. Wetenschap. Proc. B66, 1963,pp. 265-279.
23. Lorenz, R., "AchsensymmetrischeVerzerrungen in d0nnwandigenHohlzylindern," Z..VDI, Vol. 52, 1908, pp.1706-1713.
24. Timoshenko, S., "Einige Stabilit&ts-probleme der Elastizit&tstheorie," Z. Math.Phys., Vol. 58, 1910, pp. 337-385.
25. Southwell, R.V., =On the General Theoryof Elastic Stability," Phil Trans. Roy. Soc.London, Series A, Vol. 213, 1914, pp.187-244.
26. Arbocz, J. and Hol, J.M.A.M., "ANILISA -Computational Module for Koiter'sImperfection Sensitivity Theory,, ReportLR-582, Delft University of Technology,
Faculty of Aerospace Engineering, TheNetherlands, January 1989.
27. Romkes, A., "Stability and ImperfectionSensitivity of Anisotropic Cylindrical Shellsunder General Boundary Conditions,"Memorandum M-809, Delft University of
Technology, Faculty of AerospaceEngineering, The Netherlands, July 1997.
28. Hutchinson, J.W. and Amazigo, J.C.,
"Imperfection Sensitivity of EccentricallyStiffened Cylindrical Shells," AIAAJournal, Vol. 5, No. 3, March 1967, pp.392-401.
29. Cohen, G.A., "Effect of a NonlinearPrebuckling State on the PostbucklingBehavior and Imperfection Sensitivity ofElastic Structures," AIAA Journal, Vol. 6,
No. 8, August 1968, pp. 1616-1619.30. Hutchinson, J.W. and Frauenthal, J.C.,
"Elastic Postbuckling Behaviour ofStiffened and Barreled Cylindrical Shells,"
J. Appl. Mech., Vol. 36, 1969, pp. 784-790.
31. Arbocz, J., de Vries, J. and Hol, J.M.A.M.,
"On the Buckling of Imperfect AnisotropicShells with Elastic Edge Supports underCombined Loading - Part I: Theory andNumerical Analysis", Memorandum M-849,Delft University of Technology, Faculty ofAerospace Engineering, The Netherlands,
May 1997.32. Arbocz, J., The Effect of Imperfect
Boundary Conditions on the CollapseBehavior of Anisotropic Shells," Int. J.Solids Structures, Vol. 27, Numbers 46-47,
2000, pp. 6891-6915.33. Hilburger, M.W. and Stames, J.H. Jr.,
"Effects of Imperfections on the Buckling
Response of Compression-LoadedComposite Shells," Proceedings 41stAIANASME/ASCF-JAHS/ASC Structures,Structural Dynamics and MaterialsConference, 3-6 April 2000, Atlanta,Georgia, (Paper AIAA-2000-1382).
34. Almroth, B.O., "Influence of EdgeConditions on the Stability of AxiallyCompressed Cylindrical Shells", NASACR-161, February 1965.
35. Bushnell, D., "Stress, Stability andVibration of Complex Branched Shells ofRevolution: Analysis and User's Manual forBOSOR-4," NASA CR-2116, 1972.
36. Weustink, A.P.D., =Stability Analysis of
Anisotropic Shells of Revolution underAxisymmetric Load with General BoundaryConditions", Memorandum M-881, DelftUniversity of Technology, Faculty ofAerospace Engineering, The Netherlands,November 2000.
37. Imbert, J., "The Effect of Imperfections onthe Buckling of Cylindrical Shells,"Aeronautical Engineer Thesis, CaliforniaInstitute of Technology, 1971.
38. Donnell, L.H. and Wan, C.C., "Effect of
Imperfections on Buckling of ThinCylinders and Columns under AxialCompression," J. App. Mech., Vol. 17,1950, pp. 73-83.
2410
Buckling toads contour mop for the Coltech isotropic shell A--8
normol[zed buckling Iood lombdoc =_ Uc // --49.51 51 7
o0
EoI
_o°>
0_o°
6
x0
_o
E3 0E 0
0
0o
I ] [
......1.1
1.0
....""1.1'
2
• _............</.-i = 1.0 i i
).00 _.o0 8.00 12.0o 16.00 20.00
number of c[rcumferent[ol full wove5 -- n
contou_
levels1 1 .00
2 1.10
._% 1 .20
4 1.50
5 2.00
Fig. 1 Distribution of buckling loads based on Level-1 membrane prebucklinganalysis - Caltech isotropic shell A-8.
2411
w v +w o
i
a. Prebuckling Shape
-I,O
0.0
t.o
! I IolidW 2 dalhad
b. Buckting Component. Uodes
a. Membrane prebuckling, n=9
;_ = 1.00006
i wv +w O
-,3
-.2 __
-.11.
.Or
"_(max)-L ---="
O. Prebuckling Shape
-t.o
o.o
J i.o
W 1 IOlldW 2 dolhld
_meK)-L
b. Buckling Component Modes
b. Nonlinear prebuckling, n=24
_,nl = 0.844480
w v +w o
-1.o
o.o
317-.2
-.I
.o _=(mox)-L
o. Prebuckling Shape
W 1 IOJld
W2 delhld
t 1.0 ;(mol)-t.
b. Buckling Component Modes
c. Nonlinear prebuckling, n=24
;_nl = 0.844481
Fig. 2 Buckling modes of the axially compressed isotropic shell A-8
SS-3 B.C. -N x = -N o, v = W = M x = 0 : Nct = -49.51517 Ib/in.
2412
-11(Q
(AO
(n.-¢
Q
00
,,3-,j
m
r-e'J
!
0
N"
Zc _] •
_o
E
v
]
7 _
Zfl
II
bo
0.70Normoir|od OxIOI iooa Iornbdoa m Nx / --4Q.51517
0.74 0.78 0,82 0.80
I " i - ! . w - I ' • - I0.80
II
_J
9,.
I
"" II
9 _..I
Ii N,
3f |I =I
I
I
!
I
I
?I
I
I
| _'_1, i , 1
tl
,.. -,",, n
i =", I|
, , - _t, ,
¢
Z;0I
®
Z0
I
ood
t'-t
00
m
V
X
IIV
Oo4
×
®
C°_
"o
ooooo=
Perfect shell A--8 -- 1 61 rows x 181 Columns --
w--field, eige_vatue 1 , Nc=--41 .264 Ib/in,
O00000UOOOOO_ O OOOO@@@OOOOOOOUOOO0000000000@O00@00(_,O00000000000000000000000000000000000000OO0000000000
0 000000000000000000000000000000000(O00000000C,O0000
I?000 O )O 0000 00000000 0000000000GOD000 O0"O ' 00(-6
,/eo.oo 18o.oo aTo.oo
c;rc. coordinate y [0. <= y <-- 360.00]
S$--3 B.C. -- u =0 at told--plane
deltc-- tetc,= _*_ 60 degrees
0.00_60.00
conto_Jr-
levels
1 --0.802 --0.622223 --0.444444 --0.266675 --0.08889
6 0.088897 0.26667
8 0.44,4-449 0.62222
1 0 0.80
a. Contour-map of the critical (lowest) buckling mode
I=grfllct lib4111 A--8 -- 161 r'o,_| x 181 ¢olu_"_ll -- S$ ---_ 8,C. -- _=0 ot mi'4--plar_m
w--f_el¢l, oi_or_vol_ _ , N¢----_1 .264. I1_/i_° CloIta--tetom3150 Clo_roew
g!
30
g o
q
I
To._ .3_33 2 (sol? _.oo • 33_ • gloo'_"
#xisl ilto_.i o r_
/
Fig. 4
b. Axial trace of the critical (lowest) buckling mode
Buckling mode of the axially compressed isotropic shell A-8
SS-3 B.C. N x -- -N o, v -- W = M x = 0; Ncg = -49.51517 Ib/in; STAGS-A results.
2414
Perfect ilhell A--IB -- 161 rows x 181 ¢oi_._mr_ll -- $S -Ii B.¢- -- umO Ot r-r_icl--plOl-le
w--f;oJd. ; m -- .o ge_vc_lLJo 1 . NC 41 2(5.4- Ibf;r% ¢loJto--tetom_O de_lrHII
g
_zj Jl II It II II ]1 II II It Jl II II It II II II II JI [[ II II If II II ill
i_l I/I/I/I/I/I/1t I/I/II I/II I1II tl I/JlI/II I/II I1JlIIIII
o+oo oo.¢m0 t JIO.OO t Iio.oo ll4.o.oo llOO.OO _llo.oo
(=|reumferent;ol Iltotion (degrooll)
c. Circumferential trace of the critical (lowest) buckling mode
Perfect shell A--8 _ 161 fowl x 1 a 1 ¢olumnlll -- SS--3 B.C. -- u--O ot tutti--plane
w--f;eld, i;glr_vol_o 1. N¢_--41 .26_- I_J;r_. _oltc=wtlt_R.RSO degr_ell
:: :: :: :: :: :: : :; :_) : :: i " :: ;: : :0 •
o "_!Ti(_ !! i .i _ _ _! i _ _ !_-_ _ !i i _i _ _! _ _I I_:®. :(_i : :. _i _: i : _ :i :: _: : :o: .,_,.: .: :: _ :: :_ :: :: :i : i
-"r_! i iiio __!ii
o :: :: : _ :: : :: !: : _ :. :_ :! : :: : : :_ : ! :: _) _:
_ .-: _ :" _ :; _# :' _" ._ i _ " _ ,_ i: _: _,_ .-" _ "
c|rc_J_'_ferllnt[ol stotior_ (d
d. Circumferential trace of the critical (lowest) buckling mode
Fig. 5 Buckling mode of the axially compressed isotropic shell A-8
SS-3 B.C. N x = -N o, v = W = M x = 0; Nct = -49.51517 Ib/in; STAGS-A results.
24]5
Perfect shell A--8 -- 261 rows x 261 columns -- SS--3 B.C. -- u--O o± mid--ploNe
w--f;elcI, e;ger_value 1 , Nc----4-1 .83997 Ib/in, delta--tetom360 degrees
6
)00000000000000000}0000000 0000000000000000000VN
X
IV
doL..Jq
,¢X
o
oE
"oL
t00000000000000000000000000000000000000000000000._60. Oo
0.00 gO.O_ 180.00 270.00
cTrc. coord[nc:te y [O. <-- y <-- 360.00]
Cor_toLJr
levels
1 --0.80
2 --0.62222
3 -- O. ,4--4-4-4-4
4 --0.26667
5 --0.08889
6 0.08889
7 0.26667
8 0.44444
90. 6222210 0.80
a. Contour.-map of the critical (lowest) buckling mode
Perfect shill! A--8 -- :Z61 rows x 2.61 colu_nnll _ SS--3 B.C. -- _ I 0 O t rnr_--pl_ne
w--f_llll{:_, O[gCI_V{]ILII! 1. NC----41.8--_g87 Ib/;n, _lalta--tlltom360 de_lrells
g
o.0@ ' 1.0o 2.,)o 3.¢>0 4.00 I1.00 i.o0 .O(: -aMlll mt@tFo_
b. Axial trace of the critical (lowest) buckling mode
Fig. 6 Buckling mode o! the axially compressed isotropic shell A-8
SS-3 B.C. N x = -N o, v = W =M x = 0; Nct = -49.51517 Ib/in; STAGS-A results.
2416
J
i°I/l/l/J/l/tllll/ttl/l/t/l/l/l/l/l/l/l/l/ttl/l/ti,t/.oI/I/Jill/I/I/I/[1I/I/I/I/I1I/i/!/I/I/i/I/!tI/_ VV _V VV _VV
o.oo 6o.o0 i zo.oo I oo.oo =14.o.oo 3oo.oo 34o,_
cr_:_umfof'ont|ol :totlon (clegroeu)
c. Circumferential trace of the critical (lowest) buckling mode
; J ii i i_
I "iii ii
°"Ti i_ i.=i
• l! ::&: :
1
t=ti: : ! _ : =-o : :i
Fig. 7
Perfect IROII A--8 -- 2(51 _OWO x 2(51 col_lr'_mll -- $$--.3 B-C. -- u--O ol; rnic_--plone
® y _ ® 9
i_i..'+i i i ;_fi i i !i._i _iii i!_
' ; ' " : : : ' ; i A ' " '. : ' " ' " ' ,'. ! ' ! - ' : " • i 6
,i, i ._i _ ii iii$ ii ._i ii i_ i! i_, iiCJ ii Jiii i
" i_ '" ,_' ,',i ,: !: ' w : ., " i.: " _ _" _ *
o.oo I_oo t =o.oo t io.oo J_o. oo _oo,oo _eo.oo
(=t'mkJmferemt|OI Itotion (de.Tee=)
d. Circumferential trace of the critical (lowest) buckling mode
Buckling mode of the axially compressed isotropic shell A-8
SS-3 B.C. N x = -N o, v = W = M x = 0; Nc£ = -49.51517 Ib/in; STAGS-A results.
2417
Fig 8 Imperfection sensitivity for long wave axisymmetric imperfection under axial compression
(SS-3 B,C.: N x = -N o, v = W =M x =0; Nct =-49.515171b/in).
Fig 9
oo
I
\
I
EO
"o_oo_o"_¢
_o[
oO _, , = , I , i , , a ,o0o _o=o o.... o.0 -o%
Imperfection sensitivity for short wave axisymmetric imperfection under axial compression
(SS-3 B.C.: N x = -N o, v = W = M x = 0; Nc£ = -49.51517 Ib/in).
2418
ColtQ(::h i=otrop[c shol.I A--8 -- 1 6 1 row8 x 1 8 1
Criticel buckling modo ot NC I -- _2. 582 Ib/_n
o .oo io_oo 1 Io.oo 2_o.oo 31o.oo
circ. coc_rdfnote y ro. _ y _i 360.00]
columns -- SS--3 B.C.
-- Wbor/h= 1.0-coi(2--pi*.x/L)
contour
levels
1 --0.80
2 --0.62222
3 --0.44444
4 --0.26667
5 --0.0888£
6 0.08889
7 0.26667
8 0.44444
9.0.62222
10 0.80
Fig. 10 Contour map of the critical buckling mode for axisymmetric imperfection.
2419
Fig. 11
\
no
_0ocOO
E
I , = , I , = , I , i , I
O.10 o,.4,O O.41o O.IIO
nori.y_oillld |mD(lrflQtle_ a_=Dartudl -- =Sb,=r2
Imperfection sensitivity for long wave asymmetric imperfection under axial compression
using Level-1 routine BFACT (Nct = -49,51517 Ib/in).
Fig. 12 Imperfection sensitivity for short wave asymmetric imperfection under axial compression
using Level-1 routine BFACT (Nc£ = -49.51517 Ib/in).
2420
Fig. 13
r . .,=_lyre_et_a I_l_me"t_¢t;c_n Se_llltlvlt_y ¢=_ blot o_J= =hell A--8 _ SS_*._ El C
r-
;o
I
\
I 0
i:
e_¢
_d
8Ol
Wb<=r/h -- =lbc=r2-1"=l_(¢n.,q/L).=.=,m(e-.y'/R)3
= ANILI_,_=9 =(_P*_* COLL.A, PS¢.n=g
""O- ..... "O--...
• , , i , a , I , , I * , = , , ,
,.0o OJO O,4.o o-ao. 0.80 1 O0
.z d ; I= ,'f tx _lrt wo -- im 2_rmo e _ o eo o_ _m u
Imperfection sensitivity for long wave asymmetric imperfection under axial compression
(SS-3 B.C.: N x = -N o, v = W = M x = 0; Nct = -49.515171b/in).
Fig. 14
"1;8
I
\
,o
¢
:if0,00
""0. ........ _. ....... 0 ....... -E) ......
f , i , J , i , I , * , I , *
R¢•moll|Icl [mlll_fl=t[on omplltu4e -- =|i_Ot2
Imperfection sensitivity for short wave asymmetric imperfection under axial compression
(SS-3 B.C.: N x = -N o, v = W =M x = 0; Nct = -49.51517 Ib/in).
2421
?
Fig. 15
2_
ORCl.I4FERENllAL /U_3LE (ind.]
Measured initial imperfections of Caltech isotropic shell A-8 [1].
g=y/R
Fig. 16
x
T Fully restrained:U =V=W=W,,
11"8751 Brass: v=o.3E=15=106psi
8.5 I__R
Aluminum:
l E • 10=1015psi
V=0.3
3.0,
Steel:E =30=106psi
v=0.3
0
V=W=_
Reference surface
Cerroiow
Oevcon B
Spacer
Devcon B
1IN Steel end plate
Caltech test set-up and corresponding computer model.
2422
Fig. 17
-0.4 ! l
0 00 1110
dmmc=,=m_ p,_cm_, On_=U,,_)
Measured flatness of the Caitech load-cell end-ring.
!
27O 3ffi0
Fig. 18
Cor-f'll:_it'_ld effo¢_, of n"lOOlur41cI r'rlodol ond boundor"y imogrfect_O_l
Coltech _sotrop_c lllhell A--8
.._..(_..._. gw;mpb.4,.gwlrnO c -- _xpgrimentoI
- • - , l ,. J i
o _ "a-- •
U "'...
_. ....
•v ",'f" O
o q 1
I
d i . , . i , r , I , * , I , i , ,
0.00 0.02 0.04 0.08 0.08 O. 10
Amplitude of the bounclory imperfection -- xibound
Combined effect of measured modal and boundary imperfections.
2423