CompositeStructures29 (1994) 213-218

© 1994 Elsevier Science Limited

Printed in Great Britain. All rights reserved
0263-8223/94/$7.00
ELSEVIEP

Non-linear finite element modelling of an

integrally stiffened composite panel
Murray L. Scott
Cooperative Research Centre for Aerospace Structures, 506 Lorimer Street, Fishermens Bend, Victoria, 3207, Australia

The progressive introduction of advanced fibre composite materials into

aerospace structures has enabled significant performance improvements to be
achieved. Structures using these materials are generally designed so that their
behaviour is essentially linearly elastic through to failure. The realisation of
further significant weight reductions in light-weight aerospace structures is
highly dependent on design technologies which will enable postbuckling
stiffened panels to be utilised in primary structures.
The capabilities of the finite element code, MSC/NASTRAN, to predict the
onset of buckling and subsequent postbuckling behaviour of a blade-stiffened
fibre composite panel have been investigated. The panel consists of T300/914
carbon/epoxy unidirectional tape and is reinforced by integral stiffeners. Basic
modelling techniques for the efficient analysis of such postbuckling assemblies
are presented. It is concluded that the prediction of their performance requires
specialist modelling skills and a sound understanding of the behaviour of the
composite materials used in the structure.

INTRODUCTION been investigated both analytically and experi-

mentally by a number of authors. ~-6 The practical
Aircraft structures typically make extensive use of case of the behaviour of structures experiencing a
stiffened panels in a semi-monocoque construc- bending-buckling response has received less
tion in order to achieve optimum structural per- attention and is the subject of this paper. A feature
formance. The incorporation of advanced fibre of the panel used for the investigation is the
composite materials in primary structures of manner in which the stiffeners 'run-out', or term-
modem commercial aircraft presents new chal- inate. Often the design constraints are such that
lenges for designers of these components. As lami- the stiffeners need to be terminated to avoid inter-
nated composites are highly anisotropic, their use ference with surrounding structure. A common
introduces more variables which must be method of overcoming this detail design problem
considered when attempting to obtain optimum is to run-out or taper the stiffener height where
structural efficiency; producibility must also be necessary. As a consequence of selecting such a
considered early in the design stage. In this analy- configuration, only the skin can be loaded directly
tical study, the behaviour of a postbuckling blade- and the run-out angle of the stiffener introduces
stiffened panel has been investigated. The design an eccentricity in the centroidal plane of the
of the panel is representative of relatively lightly panel; this results in bending deformations from
loaded aircraft skin construction, manufactured as the onset of compressive loading.
a co-cured structure from unidirectional T300/ The laminate design for the panel was con-
914 carbon fibre/epoxy preimpregnated tape. strained by the criterion that the stacking se-
The configuration of the stiffeners and skin is quence for the stiffeners should be symmetrical.
designed to allow the structure to operate in the This resulted in a skin that alternates between a
postbuckled regime. symmetrical and an unsymmetrical laminate, as
The phenomenon of postbuckling of stiffened shown in Fig. 1. This departure from conventional
panels has traditionally been studied in panels symmetrical composite construction is necessary,
where the skin and stiffeners are uniformly as co-cured stiffeners are manufactured by
loaded. This aspect of composite structures has extending the first three skin pries to the sides of
213
 214 Murray L. Scott

I ~ ,I Explply
°dedtop
layupSeCti°nview of
, ..... . /~2,mm SKIN STIFFENER
PROP 1 PROP 2 PROP 3 PROP 4

<U/ 45 -45
-45
45
-45
45
45
-45
90
~P/ 7 ~j/ff474mm
90 90 90
02 02 010 010
90 90 90 90
-45 -45 45 -45
45 45 -45 45

/ 672mm . . . . . . . . . . //

Fig. 1. Blade-stiffened panel ply lay-up configuration.

the blade-stiffeners. This design allows the stiffe- MAT8 data entries which are based on two-
ners to become effectively integral, but the con- dimensional orthotropic assumptions consistent
sequence of this is that the skin lay-up must with classical laminate t h e o r y / A n ultimate com-
alternate across the panel between each stiffener. pressive loading of 200 N/mm was applied to the
The panel dimensions were determined by a short edges of the panel as a distributed line load-
requirement to be compatible with structural test ing in the plane of the skin. This loading was
rigs, which have 'working' dimensions of 474 mm applied to the shorter edges via 'dummy' bar
in width and 672 mm in length. The angle of the elements, which were assigned with low section
stiffener run-out selected is 20 °, with a stiffener and stiffness properties; this is necessary as the
height of 25 mm and lateral spacing of 79 mm. CQUAD4 element does not support in-plane line
The advent of finite element codes with specific loads.
capabilities applicable to orthotropic composite Clamped edge boundary conditions were
laminates has permitted structures to be investi- applied to the blade-stiffened panel and kinematic
gated in greater detail than ever before. The abil- constraints, known as Multi-Point Constraints
ity to post-process analytical data such as ply (MPCs) in MSC/NASTRAN, were applied to
strains and failure indices using advanced model- ensure that the loaded edges were uniformly
ling packages, offers designers the potential to shortened under the compressive loading, MPCs
efficiently obtain maximum structural perform- were additionally used on the unloaded edges to
ance. The key objective of this work is to enable simulate movable edges that remain parallel, as
commercial airframe components to safely shown in Fig. 2.
operate in their postbuckled states for extended
periods. In this study, the general purpose finite
element code, MSC/NASTRAN, has been uti- LINEAR ANALYSIS OF MODEL
lised to investigate the bending-buckling response
well into the postbuckled regime for the panel A linear static analysis was performed to obtain
previously described. Linear static and linear an appreciation for the magnitude of out-of-plane
buckling analyses have been performed, together bending deflections expected for the ultimate
with geometric non-linear analyses in order to loading of 200 N/mm. The value for end shorten-
compare the structural behaviour predicted by ing experienced by the panel was 1.94 mm; the
these different analytical techniques. result for the maximum out-of-plane deflection
was 10-1 mm, occurring offset from the centre of
the panel. When the out-of-plane displacemems
THE FINITE ELEMENT MODEL are compared to the panel skin thickness of
1-04 mm, the requirement for a geometric non-
The blade-stiffened panel under investigation was linear analysis becomes self evident.
modelled using 1676 four-noded, two-dimen- The results from an eigensolufion linear buck-
sional shell elements, known as CQUAD4 ele- ling analysis predicted the first eigenvalue or
ments in MSC/NASTRAN. The carbon/epoxy lowest buckling load factor to be 0.340 of ultim-
lay-up was represented by the PCOMP and ate, with the corresponding eigenmode of 10 half-
 Non-linear finite element modelling 215

Constrainedin their torsional stiffness, with the result that the

all 6 degreesof stiffeners and skin interact.
freedom.',,~
C.Ey

/
< k:
)2--
NON-LINEAR ANALYSES OF MODEL

An obvious initial step in evaluating the geometric

C.Ey non-linear capability of MSC/NASTRAN was an
investigation of the structural response without
C.E.x/y - ClampededgewithMulti-PointConstraintsrelatingthe
imperfections or disturbances to induce buckling.
x/y translationfreedomas an equalitybetweennodes.
The non-linearity was introduced by the eccen-
Fig. 2. Loadingand boundaryconditionsapplied to panel. tricity of the centroidal plane of the panel under
compression. This was investigated to allow a
direct comparison with the behaviour predicted
with the linear static analysis. The results from the
analysis showed that the complexity of the model
and loading was significant enough to introduce a
postbuckled solution. The maximum out-of-plane
deflection was 23.5 mm, whilst the end shortening
displacement reached was 4.58 mm. When com-
pared to the values from the linear static analysis,
the difference is significant.
A geometric non-linear analysis was performed
with initial imperfections to represent the first
linear buckling mode. The imperfections were
provided by disturbing forces applied at locations
where the peak buckling amplitudes were pre-
dicted from the linear buckling analysis. The
Fig. 3. Linearbucklingmode shape for panel. forces were applied such that their resulting dis-
placements only caused a deformation of 1% of
wavelengths between the stiffeners, as illustrated the skin thickness. The loading history was orga-
in Fig. 3. The alternating symmetrical and unsym- nised into three load steps; the disturbing forces
metrical ply stacking sequence between the stif- where applied in the first load step and removed
feners resulted in a non-uniform buckling mode from the following two load steps. Although
shape. applying forces to cause initial imperfections
The same model was analysed with a com- rather than offsetting nodes (GRID is the data
pletely symmetrical skin to compare the results entry in MSC/NASTRAN) induces stresses into
for buckling load factor and mode shape. The the structure, the stresses are small compared to
mode shape generated was close to the original those introduced by the compressive load and
shape, except that the mode displayed symmetry. produce no adverse effect on the response of the
The buckling load factor was 0.337, lower than structure. The use of offset nodes as the initial
that for the alternating unsymmetrical laminate imperfections resulted in the same response as
configuration. This indicates that the extensional- that for a disturbing force and also for the solution
bending coupling stiffness coefficients are signifi- without imperfections.
cant in modifying the buckling load factor and The results of the analysis clearly demonstrated
assist in increasing the buckling load factor for the that the structure exhibited a bending-buckling
unbalanced configuration. response. The deformed shape and out-of-plane
The linear buckling mode shape demonstrated contours shown in Figs 4 and 5, not only repre-
that the response of the stiffeners is dependent on sent a panel that is undergoing bending which
the behaviour of the skin. A consequence of the causes the greatest deformation, but also buckling
stiffener run-outs is the absence of any lateral which is indicated by the small irregularities in the
support at their ends, which would be present if response. The contours indicate a structural
the stiffeners extended uniformly for the full response that is symmetric about a diagonal; this
distance. This lack of restraint primarily reduces behaviour could be expected, as the stiffeners
216 Murray L. Scott

Fig. 6. Panel lower surface stress distribution for shell

Fig. 4. Panel deformed plot from the non-linear analysis. elements.

~24

190

173

156

139

122

714

$44

375

355

Fig. 5. Panel out-of-plane displacement contours from Fig. 7. Panel top surface stress distribution for shell
non-linear analysis. elements,

have a relatively low torsional stiffness. Small analysis highlights the differences in the assump-
irregularities in the contours are present due to tions used in the numerical analyses. Performing a
the skin alternating between a symmetrical and postbuckling analysis introduces many factors
unsymmetrical ply stacking sequence. which need to be considered in the analysis of a
Examination of the distribution of stresses in structure. The sensitivity to imperfections and
the x-direction for the outer surfaces of the covergence criteria can all have significant effects
elements shown in Figs 6 and 7, provides an indi- on the computational time and the load path that
cation of the change in stress throughout the skin the structure adopts.
and stiffeners in the postbuckled state. The buckl- For a flat plate under compressive loading, the
ing is discernible by the local tensile stresses postbuckling response depends on the initial
produced by buckling peaks. The results show the imperfection in the prebuckled solution. The
skin to be largely in compression as expected, bifurcation point for this load path is therefore
since the loading is via the skin. The stiffeners sensitive to the imperfections used. The path
carry tension loads due to the bending deforma- followed by the structure is the one that corre-
tion associated with the offset centroidal plane of sponds to the lowest potential energy. The blade-
the structure. The response of the structure dis- stiffened panel investigated was analysed both
plays symmetry about a diagonal. with initial disturbances to reproduce the first lin-
ear eigenvalue and also without imperfections.
The solutions both converged to the same
DISCUSSION OF RESULTS response, indicating that geometry and loading for
the blade-stiffened panel were numerically suffi-
The prediction of a bending-buckling structural cient to introduce 'imperfections' for a postbuck-
response using linear compared to non-linear ling solution. The solution with the disturbing
 Non-linear finite element modelling 217

forces applied converged with fewer iterations stiffness is correct for any general small displace-
than that for the undisturbed analysis. ment. The load-deflection curve for the non-
Newton's method for non-linear analysis linear analysis shown in Fig. 9, predicts a
requires the solution of non-linear equilibrium bifurcation point close to that predicted with the
equations at increments along the structural load- linear buckling solution. This indicates that for the
ing history. As Newton's method has a finite configuration considered, a linear eigensolution is
radius of convergence, too large an increment can sufficient to estimate the onset of buckling. From
prevent the solution from converging, as the the load-deflection curve for the non-linear ana-
equilibrium state being sought is too far from the lysis, a non-linearity in the curve can be detected
initial state. The automatic (AUTO) stiffness in both the prebuckled and postbuckled states.
update method s provided in MSC/NASTRAN Performing a non-linear eigensolution using the
was selected, so that any diverging solution differential stiffness matrices in the vicinity of
between selected increments would automatically instability would offer a refined estimate to the
reiterate with a smaller increment. The conver- buckling point, however this is still only a numeri-
gence criteria selected changed with the loading cal point of instability. The significant change in
states of the structure. For the prebuckling solu- tangential stiffness on the load-deflection curve
tion, the residual load convergence criterion was corresponds to the imperfections producing signi-
selected, whereas for the postbuckling or 'soften- ficant local out-of-plane displacements. The point
ing' solution the displacement error criterion was of bifurcation was judged to be at 38% of ulti-
selected. mate, which is 4% higher that that predicted from
The cross-sections of the deformed shapes a linear buckling analysis.
predicted for the linear static and non-linear The influence of the alternating symmetrical
analysis are plotted in Fig. 8. As can be observed, and unsymmetrical skin stacking sequence was
not only is the magnitude different, but so is the clearly evident in the response of the panel for
shape of the panel when subjected to a compres- both linear and non-linear analyses. The interac-
sive loading. The non-linear solution predicted a tion between extensional and bending stiffnesses
much higher out-of-plane displacement due to the is related by the constitutive relationship com-
shift in the centroidal plane as the loading is monly represented for composite materials as
updated. This shift as the load increases naturally follows:
adds to the overall bending deformation. The
linear solution involves only a constant distance
between the centroidal and loading planes, thus
underestimating the total deformation of the
panel.
The comparison of the deformations between For problems where two-dimensional assump-
linear and non-linear analyses clearly demon- tions are appropriate, these matrices relate three
strates the discrepancies in assuming small dis- components of strains (e) and curvatures (lc)to
placements through to ultimate loading. The three components of loads (N) and moments (M)
linear eigensolution assumes that the differential per unit length. The extensional (A),

Linear Static 2OO

25 !
/ ~, " ,, <~ Geornetdc Nonlinear 180
5
160

,\ 140
"Q,
f~ 120
/ 15
/
/i Nx(N'mm)too
80

/ /I 6O

¢ I 5 4O

2O

0,
254) 200 158 -10o 50 o 50 lOO 150 2o0 250 0 05 1 1.5 2 2.5 3 3.5 4 4.5 5
Distance along cross section Y (ram)
Endshortening(ram)
Fig. 8. Panel out-of-plane displacements at mid-plane from Fig. 9. Panel load vs end shortening for geometric non-
ana]yses. linear analysis.
218 Murray L. Scott

extensional-bending coupling (B) and bending path, which resulted in a bending-buckling

(D) matrices are defined as follows: response. The alternating symmetrical and
n unsymmetrical skin laminate does influence the
results for both the linear and non-linear analyses,
Aij = E ( 6 i j ) k ( h k - - h k - , )
k=l
but only to a small extent as evidenced by the
minor irregularities in the response. Nevertheless,
n it has been demonstrated that the prediction of
Bij-2-! 51(Qi/)k(hk_hk_12 ) postbuckling performance requires both specialist
modelling skills and a sound understanding of the
orthotropic behaviour of the composite materials
1 " (Q/j)~(hk_h/~) used in the structure.
k=l

The A, B and D matrices involve the ply stiffness ACKNOWLEDGEMENTS

matrix, Q ij, in a global coordinate system with the
upper and lower surfaces of each ply, h k and h k- J, The author wishes to gratefully acknowledge the
being measured from the mid-plane of the lami- significant assistance he has received from his
nate cross-section. From the definition of the B colleagues at the Cooperative Research Centre
matrix, the coefficients are all zero for mid-plane for Aerospace Structures and in particular the
symmetrical laminates. The presence of a non- computing efforts of Christian Trentin.
zero extension-bending stiffness matrix implies
that the mid-plane normal and shear forces result
in not only mid-plane deformations, but also REFERENCES
twisting and bending curvatures. These effects
were evident in the response predicted from the 1. Bushnell, D., Computerized Buckling Analysis of Shells.
finite element model, although they were small in Kluwer, Dordrecht, 1989.
2. Elishakoff, I., Arbocz, J., Babcock, C. D. & Libai, A.
magnitude when compared to the overall panel (Eds), Buckling of Structures -- Theo~ and Experiment.
behaviour. Similar findings have been reported by Elsevier Science, Amsterdam, 1988.
other researchers.9- ~ 3. Palazotto, A. N. & Dennis, S. T., Nonlinear Analysis o]
Shell Structures, American Institute of Aeronautics and
Astronautics, Inc., AIAA Education Series, Wash-
ington, 1992.
CONCLUSIONS 4. Kumar, A., Postbuckling analysis of orthotropic rectan-
gular plates under compression. J. Aeronautical Socie~"
of India, 41 (2)(1989) 191-203.
The finite element modelling of a fibre composite 5. Stein, M., Postbuckling of eccentric open section
blade-stiffened panel using both linear and non- stiffened composite panels: American Institute of Aero-
linear analytical techniques to investigate the nautics Conference Paper, AIAA 88-2215, Washington,
1988.
structural response under a compressive load, has 6. Reddy, J. N., On refined computational models of com-
clearly demonstrated the significant differences in posite laminates. Int. J. Numerical Methods in Engineer-
the predicted responses. For the configuration ing, 27 (2)(1989) 361-82.
7. Schaeffer, H. G., MSC/NASTRAN Primer." Static and
considered, the results from the linear buckling Normal Modes Analysis, PDA Engineering, Costa Mesa,
analyses correlate well with the bifurcation point 1982.
predicted from the non-linear analysis. However, 8. Lee, S. H. (Ed.), MSC/NASTRAN ttandbook for Non-
the postbuckling behaviour of the panel cannot be linear Analysis, MacNeal-Schwendler Corporation, Los
Angeles, 1992.
modelled by a linear eigenvalue solution, as the 9. Starnes, J. H., Dickson, J. N. & Rouse, M., Postbuckling
overall bending response of the panel cannot be behavior of graphite-epoxy panels. Proc. ACEE Com-
predicted. The linear static analysis significantly posite Structures Technology Conference, Seattle, 13-15
August, 1984, pp. 137-59.
underestimated the out-of-plane displacements 10. Wiggenraad, J. F. M., The postbuckling behaviour of
and predicted a different response to the non- blade-stiffened panels loaded in compression. National
linear analysis. Aerospace Laboratory Report, NLR MP 850219 U,
Amsterdam, 1985.
Investigations into the imperfection sensitivity 11. Geier, B., Buckling and postbuckling behaviour of com-
for the non-linear analysis indicated that the struc- posite panels. Int. J. Numerical Methods in Engineering,
ture numerically converges to a 'natural' load 27 (1989) 403-27.