Numerical modelling of thin-walled Z-columns made of general laminates subjected to

uniform shortening
Andrzej Teter, and Zbigniew Kolakowski

Citation: AIP Conference Proceedings 1922, 080002 (2018); doi: 10.1063/1.5019073

View online: https://doi.org/10.1063/1.5019073
View Table of Contents: http://aip.scitation.org/toc/apc/1922/1
Published by the American Institute of Physics

Articles you may be interested in

Design solutions for improving the lowest buckling loads of a thin laminate plate with notch
AIP Conference Proceedings 1922, 080004 (2018); 10.1063/1.5019075

Effective elastic properties of sintered materials with branched cracks

AIP Conference Proceedings 1922, 030008 (2018); 10.1063/1.5019042

New formulation of the discrete element method

AIP Conference Proceedings 1922, 030009 (2018); 10.1063/1.5019043

Numerical/phenomenological model for fatigue life prediction of hybrid laminates

AIP Conference Proceedings 1922, 050001 (2018); 10.1063/1.5019055

Multimaterial topology optimization of contact problems using phase field regularization

AIP Conference Proceedings 1922, 020004 (2018); 10.1063/1.5019031

Buckling and limit states of composite profiles with top-hat channel section subjected to axial compression
AIP Conference Proceedings 1922, 080001 (2018); 10.1063/1.5019072
 Numerical Modelling of Thin-Walled Z-Columns Made of
General Laminates Subjected to Uniform Shortening
Andrzej Teter1,a) and Zbigniew Kolakowski2,b)
1
Faculty of Mechanical Engineering, Department of Applied Mechanics, Lublin University of Technology,
Nadbystrzycka 36, 20-618 Lublin, Poland
2
Department of Strength of Materials (K12), Lodz University of Technology, Stefanowskiego 1/15,
90-924 Lodz, Poland
a)
Corresponding author: a.teter@pollub.pl
b)
zbigniew.kolakowski@p.lodz.pl

Abstract. The numerical modelling of a plate structure was performed with the finite element method and a one-mode
approach based on Koiter’s method. The first order approximation of Koiter’s method enables one to solve the eigenvalue
problem. The second order approximation describes post-buckling equilibrium paths. In the finite element analysis, the
Lanczos method was used to solve the linear problem of buckling. Simulations of the non-linear problem were performed
with the Newton-Raphson method. Detailed calculations were carried out for a short Z-column made of general laminates.
Configurations of laminated layers were non-symmetric. Due to possibilities of its application, the general laminate is very
interesting. The length of the samples was chosen to obtain the lowest value of local buckling load. The amplitude of initial
imperfections was 10% of the wall thickness. Thin-walled structures were simply supported on both ends. The numerical
results were verified in experimental tests. A strain-gauge technique was applied. A static compression test was performed
on a universal testing machine and a special grip, which consisted of two rigid steel plates and clamping sleeves, was used.
Specimens were obtained with an autoclave technique. Tests were performed at a constant velocity of the cross-bar equal
to 2 mm/min. The compressive load was less than 150% of the bifurcation load. Additionally, soft and thin pads were used
to reduce inaccuracy of the sample ends.

INTRODUCTION
A general laminate has many layers and the arrangement of its layers is non-symmetric. The main disadvantage of
the general laminate is such that mechanical coupling effects can occur. Different types of couplings between
extension/compression, shearing, bending and twisting can take place [1-2]. It causes that in-plane load induces out-
of-plane deformation and vice versa. This is very important during the manufacturing process of laminated thin-walled
structures. A special type of the general laminate (the so-called hygro-thermally curvature-stable laminate), which is
not warping during manufacturing, is developed. A general theory of hygro-thermally curvature-stable laminates is
presented in Vannucci et al. [3-4].
In the literature one can find a few studies on the effect of mechanical couplings on the behaviour of composite
structures [5-9]. Haynes and Armanios [5] examine the laminates that produce the maximum extension-twist coupling.
They show a few stacking sequences of laminate layers to demonstrate improvement in coupling. Loughlan [6]
discusses an influence of anti-symmetric angle-ply laminates on the compressive stability. The buckling and post-
buckling behaviour of plates made of coupling laminates is studied in Teter et al. [9]. The Koiter’s theory is used to
determine the post-buckling equilibrium behaviour. Summing up the results, the most interesting observations can be
made about the laminate with bending-twisting and extension-shearing couplings. Some effects of mechanical
coupling on the post-buckling behaviour or vibration analyses of composite laminated structures are studied in the
paper by de Goeij et al. [7]. The stability analysis of thin-walled composite beams with symmetric and arbitrary
laminations subjected to a compressive force is performed by Kim et al. [8]. York [1-2,10] discusses a number of

Computer Methods in Mechanics (CMM2017)

AIP Conf. Proc. 1922, 080002-1–080002-11; https://doi.org/10.1063/1.5019073
Published by AIP Publishing. 978-0-7354-1614-7/$30.00

080002-1
stacking sequence configurations of layers for general laminates with different types of mechanical couplings.
He proposes the hygro-thermally curvature-stable laminates with non-standard ply angle orientations.
According to the classical laminate theory [11], the constitutive equation can be written as:

N  ε 
 K   (1)
M
  κ 
and
N   A B  ε 
     (2)
M   B D κ 
The stiffness matrix K defines a relationship between section forces N / moments M and deformation
 / curvature κ. The K-matrix can be divided into three submatrices, namely: the extensional (i.e., A), coupling
(i.e., B) and bending (i.e., D) one, which can be written as:

 A11 A12 A16   B11 B12 B16   D11 D12 D16 

A  
A22 A26  B    B22 B26   D   D 22 D 26  (3)
 Sym. A66   Sym. B66   Sym. D66 
Each element of the stiffness submatrices can be calculated when mechanical properties and a sequence of the
laminate plies are known. The elements of the A-submatrix denoted as A16 and A26 (Eq. 3) describe separate in-plane
coupling effects. In this case, an interaction between shearing and extension or compression / extension occurs [10].
In contrast, the elements of the D-submatrix denoted as D16 and D26 (Eq. 3) describe separate out-plane coupling
effects. Here, one can observe an interaction between bending and twisting [8].
If the elements of the B-submatrix (Eq. 3) are non-zero, in-plane and out-plane coupling effects can be noticed.
There are many types of coupling effects [1-2, 12].

NUMERICAL MODELLING
Thin-walled structures can exhibit numerous different buckling modes depending on their length. If a structure is
short, local buckling takes place. It is the first buckling mode, and the corresponding buckling stress is the lowest.
Other forms of buckling are possible but buckling loads will be higher. With longer columns, coupling buckling may
occur if two or more eigenvalue loads are nearly identical. Finally, one can distinguish global buckling which is typical
of long structures. The numerical modelling of the buckling behaviour of plate structures can be performed with the
analytical-numerical method (ANM) [13-15] based on Koiter’s asymptotic method (Fig. 1a) or the finite element
method [16-17] (Fig. 1b).

(a) (b)
FIGURE 1. Modeling of a Z-column subjected to uniform shortening: (a) ANM model; (b) FE model

080002-2
 Both the methods allow us to determine static and dynamic buckling stresses as well as the post-buckling
equilibrium path of a thin-walled structure subjected to various types of loads. The numerical results should then be
verified in experimental tests performed on real structures (Fig. 2) [18-21].

One-mode approach based on Koiter’s method (ANM)

The analytical-numerical method is based on Koiter’s asymptotic method [13-15]. An eigenvalue problem of
perfect structures and a post-buckling analysis were performed. In this case a plate model was used for all walls of the
thin-walled structure (Fig. 1a). The column is built of plates, which are connected and supported at both ends. With
variational principles, differential equations of equilibrium can be obtained. The solution to these equations should
satisfy the initial conditions, the kinematic and static continuity conditions at the junctions of adjacent plates, and the
boundary conditions. If the number of interacting buckling modes is one, the solution is a one-mode approach
(i.e., uncoupled buckling). In this case, the dimensionless post-buckling equilibrium path can be written as [13]:

2 3
 P w  w  w P  wo 
1  
 P  t  b  t   c  t   P  t  (4)
 cr  cr  
where: w is an amplitude of the buckling mode (i.e., maximum deflection), t – a wall thickness, Pcr, P and wo are a
bifurcation load, a compression load and an amplitude of the initial deflection corresponding to the buckling mode
(i.e., maximum initial deflection), respectively, b is a coefficient in the first order approximation, whereas c is a
coefficient in the second approximation.
The post-buckling equilibrium path is a plot of compressive loads versus the amplitude of deflections. The
coefficients: b and c are known. More details can be found in [14-15]. For plate structures, the coefficient c is always
positive, so the post-buckling paths have to be stable. If the initial imperfections are 0, the dimensionless post-buckling
equilibrium path can be written as:

2
 w  w P
1 b    c    (5)
t  t  Pcr
For Z-columns, the coefficient b is always 0, thus the post-buckling paths are symmetric:

2
 w P
1 c    (6)
t  Pcr

Finite element method (FEM)

The numerical simulations were performed with the Abaqus system [17]. The FE-model of the column under
analysis (Fig. 1b) was composed of multi-layered shell finite elements with 8 nodes. This type of elements enabled
one to definite the laminate lay-up by an independent definition of the layer orientation and their mechanical
properties. The elements had six degrees of freedom at each node. A reduced integration technique was applied. An
FE-model of the Z-column was built with a structural square finite element mesh. To discuss the mechanical coupling
effects, the special stacking sequences of the laminates were chosen. First laminate lay-up was with a separate in-
plane coupling effect. Second one was with a separate in-plane coupling and separate out-plane coupling effects. And
the last one was with a separate out-plane coupling effect. Individual layers of the laminate were made according to
the Layup-ply technique [17]. The boundary conditions were as follows: the edges of bottom sections of the column
can only rotate around themselves, the edges of top sections of the column can rotate and move to shorten the column.
In addition, the nodes from the top end of the column have to have the same displacements. The edge load was applied
to the top section of the column. The laminate is described by properties of the orthotropic material in a two-
dimensional state of strain. The FEM analysis involved solving the eigenvalue problem of buckling with the Lanczos
method. Next, post-buckling paths were determined with a nonlinear buckling analysis of the model with an initiated
geometrical imperfection reflecting the lowest local buckling mode with an amplitude of 10% of the wall thickness.
To this end, the Newton-Raphson method was employed, because post-buckling equilibrium paths of the plate
structures have always to be stable.

080002-3
 Experimental verification
The experimental investigations were conducted on a universal testing machine with a special grip made of two
identical parts (Fig. 2a).

2 1
2

(a) (b)
FIGURE 2. Experimental investigations of Z-columns made of general laminates subjected to uniform shortening:
(a) test stand, (b) sample and sensors

One part consisted of a rigid steel plate and a clamping sleeve. The speed of the upper cross beam in the testing
machine was maintained constant at 2 mm/min. The samples were made with an autoclave technique. The chosen
laminates were hygro-thermally curvature-stable laminates, which were not warping during manufacturing [1]. The
compressive loads and strains in the direction of load were measured at the points where deflections were the highest.
A strain-gauge technique was used. Sensors were located on the opposite sides of the column wall in the region of the
highest half-wave amplitude determined with the FE analysis (Fig. 2b). The compressive load was less than 150% of
the bifurcation buckling load. To determine the post-buckling path, we assumed that the deflection was proportional
to a difference in strains (i.e., 1, 2 - Fig. 2b) in each point:

w ~ 1   2 (7)
and
w 1   2
 (8)
wmax  1   2 max
Because in our studies, transverse deflections cannot be measured, so any other parameter of deflections can be
plotted instead. One such parameter is the difference in strains (Eq. 7) in the load direction on the two sides of the
wall. That difference will be zero at low loads and will be large above buckling load. More details can be find in [22].

080002-4
 RESULTS
All tests were performed on thin-walled Z-columns made of a carbon-epoxy laminate under uniform shortening.
The dimensions of their cross-section are shown in Fig. 3. The structures were simply supported at both ends.

FIGURE 3. Dimensions of the cross-section in mm

The stacking sequences of the laminate were: [60,0,-60,60,-605,(0,-60)3,602,-60]T (denoted as Case 1),
[60,0,-602,605,(0,60)2,0,-602,602]T (denoted as Case 2), [60,02,-602,603,-602,03,-602,0,602]T (denoted as Case 3),
respectively. The first case, i.e., the laminate with a separate in-plane coupling effect and stiffness submatrices, can
be written as:

 A11 A12 A16   D11 D12 0 

A   A22 A26  B0 D   D 22 0  (9)
 Sym. A66   Sym. D66 
The second case is the laminate with a separate in-plane coupling and separate out-plane coupling effects. The
stiffness submatrices are:

 A11 A12 A16   D11 D12 D16 

  D   D 22 D 26 
A A22 A26  B0 (10)
 Sym. 
A66  
 Sym. 
D66 
The last case, i.e., the laminate with a separate out-plane coupling and stiffness submatrices, can be written as:

 A11 A12 0   D11 D12 D16 

A   A22 0  B0 D   D 22 D 26  (11)
 Sym. A66   Sym. D66 
The lengths of the specimens were: 270 mm (Cases 1 and 2) and 330 mm (Case 3), correspondingly. The length
of the samples was chosen to obtain the lowest value of local buckling load. The amplitude of the initial imperfection
related to the first buckling mode was equal to 10% of the column wall thickness. The mechanical properties of the
carbon-epoxy laminate were as follows: Young’s modulus along the fiber direction (i.e., direction 1) was 170 GPa
and along the transverse direction of the fibers (i.e., direction 2) – 7.6 GPa, respectively; Poisson’s ratio in plane 1-2
was 0.36; shear modulus in plane 1-2 was 3.52 GPa. The dimensions of the samples and the mechanical properties of
the laminate did not change during the experimental investigations.

TABLE 1. Lowest bifurcation loads in N

Method Case 1 Case 2 Case 3

Koiter’s method 1941 1831 1639
Finite element method 1913 1856 1744

080002-5
 FIGURE 4. Post-buckling equilibrium paths: (a) perfect structure; (b) real structures with initial imperfections

The first step involved determination of bifurcation loads (Fig. 4). A numerical analysis was performed to solve
the eigenvalue problem for the perfect structure (i.e., initial imperfections of the structures did not exist). Solving
eigenvalue problem, the buckling modes can be determined. But either displacements and stresses are unknown,
because solutions contain the unknown constant value. The numerical solutions are presented in Table 1. For all cases,
the buckling mode had three buckles (Fig. 1b). A good agreement of the results was achieved. The change in the
bifurcation loads was equal to a few percent. The bifurcation loads were almost the same in cases 1 and 2. In case 3,
the bifurcation load was the lowest. Thus, a separate out-plane coupling effect is the most dangerous.
The second step consisted in conducting a post-buckling analysis. Any imperfections can be entered to numerical
models of the real structures after determining the bifurcation loads. Solving the post-buckling problem, both
displacements and stresses can be found. In Fig. 5, the numerical solutions are compared in a dimensionless form. In
the numerical solutions, the amplitude of deflections for each buckle was the same. An agreement of the obtained
results was high for a low overload parameter (denoted as P/Pcr in Figs. 5-6), less than 200% of the bifurcation load.
If the overload parameter was high, the error increased because the analytical-numerical method (ANM) is a one-
mode approach, whereas the finite element method (FEM) is a multi-mode approach. To reduce the risk of the structure
damage and to compare both the numerical methods, an experimental verification was carried out for the compressive
load less than 150% of the bifurcation load.

(a) (b)
FIGURE 5. Numerical post-buckling paths in a dimensionless form: (a) Case 1 and 2, (b) Case 3

In Fig. 6, an influence of coupling effects on the post-buckling behaviour obtained with both the numerical methods
is presented. In all the cases under consideration, the equilibrium paths were stable. A separate in-plane coupling effect
was the most dangerous, because the stiffness of the post-buckling equilibrium path was the lowest.

080002-6
 (a) (b)
FIGURE 6. Influence of coupling effects on the post-buckling behavior (a) ANM results, (b) FE results

The real samples are not perfect, so initial imperfections are not 0 and the bifurcation effect did not exist (Fig. 4 –
line (b)). It was possible to determine an approximate buckling load with the following methods: the vertical tangent
method (P1 in Figs. 7-8), the averaged strain method (P2 in Figs. 7-8), the P-w method (P3 in Figs. 7-8), the P-w2
method (P4 in Figs. 7-8), the inflection point method (P5 in Figs. 7-8) and Koiter’s method (P6 in Figs. 7-8) [19-21].
Approximate buckling loads have to be lower than the bifurcation load (Fig. 4). If the amplitude of initial imperfections
is low, the errors of approximate buckling loads are low. Otherwise, the errors grown very fast. Hence, the samples
should be as perfect as possible.
The vertical tangent method (P1 in Figs. 7-8) can be applied if the strain-gauge technique was used in the
experimental investigations. Strains were measured as a function of compressive loads at a chosen point on the
opposite sides of the column wall. Thus, a plot of compressive loads versus the average strain can be drawn. Using
this plot, an approximate buckling load can be found as the compressive load corresponds to the minimum value of
the average strain.
In the second method (i.e., the averaged strain method (P2 in Figs. 7-8)), the approximate buckling load
corresponds to the point of intersection of two lines on the plot of compressive loads versus the average strain. The
first line approximated the pre-buckling state, whereas the second one approximated the post-buckling state. It is
important that the overload parameter (i.e., a ratio of the compressive load to the bifurcation load) should be low. If
the overload parameter is high, the error in the buckling load increased rapidly.
If the experimental post-buckling path is known, an approximate buckling load can be found in a different way
with: the P-w method (P3 in Figs. 7-8), the P-w2 method (P4 in Figs. 7-8), the inflection point method (P5 in Figs. 7-
8) and Koiter’s method (P6).
In the P-w method (P3 in Figs. 7-8), the part of the path for the compressive load higher than the bifurcation load
is approximated with a quadratic function. A free term of this function is the approximate buckling load. This method
was not good enough, because buckling loads can vary extensively.
In the P-w2 method (P4 in Figs. 7-8), the post-buckling path is presented as a plot of compressive loads versus the
square amplitude of deflections. A free term of the approximate linear function is the approximate buckling load. This
method is perfect because approximate buckling loads are almost the same.
In the inflection point method (P5 in Figs. 7-8), the buckling load corresponds to the inflection point of the
approximate cubic function. This method can overestimate the buckling load.
In the last method (P6 in Figs. 7-8), a free term of the approximate quadratic function is equal to the approximate
buckling load. This function has to be symmetric. The results for this method and the P-w2 method were the same.
The buckling loads were determined with all experimental methods for 12 samples. In each case 4 samples were
considered. The results are shown in Fig.7. The dark and bright lines are the bifurcation loads obtained with the ANM
and FEM, respectively. For all cases, the values of median and its errors were calculated (Fig. 8). A good agreement
of the results was attained.

080002-7
(a)

(b)

(c)

FIGURE 7. Approximate buckling loads obtained with all the experimental methods

080002-8
 (a)

(b)

(c)

FIGURE 8. Medians of the approximate buckling loads and their errors for all the cases under investigation

080002-9
 A separate out-plane coupling effect is the most dangerous. Therefore, approximate buckling loads of the samples
were the lowest. The trend lines were horizontal. The median errors were smaller than 26%. Additionally, the
experimental buckling loads were always lower than the bifurcation ones. In the authors’ opinion, the median errors
were slight because the buckling loads are a low estimation of the bifurcation loads. The samples were not perfect.
When the amplitude of imperfections was low, the errors were smaller. The second problem was the thickness of the
samples. If the thickness was changed inconsiderably, the buckling loads changed considerably (Fig. 9).
The experimental post-buckling paths were compared to the theoretical ones. The rigidity factor c (Eq. 2) of the
post-buckling paths was compared. Many samples overestimated the rigidity factor. All experimental and numerical
post-buckling equilibrium paths were stable. A separate in-plane coupling effect was the most dangerous. Therefore,
the stiffness of the samples was the lowest. Further theoretical and experimental investigations of this problem are
necessary.

FIGURE 9. Bifurcation loads versus wall thickness

CONCLUSIONS
An application of the analytical-numerical method based on Koiter’s asymptotic method provides very effective
solutions to all types of thin-walled structures subjected to various types of loads. All static and dynamic buckling
analyses of thin-walled structures can be conducted with this method. Moreover, in contrast to the FEM, ANM
calculations can be performed much faster and easier, and the results show a satisfactory accuracy. Nonetheless, the
visualization of results is much easier in the FEM. Given that the results obtained with the two methods and the
experimental results were similar, it can be concluded that the proposed modelling technique yields accurate results.
Different types of couplings had a significant influence on the buckling behaviour of structures under compression.
The lowest local buckling loads were achieved for the laminate with a separate out-plane coupling effect. All
equilibrium paths were stable. The post-buckling path was the most sensitive to a separate in-plane coupling effect.

ACKNOWLEDGMENTS
The investigations discussed in this paper were financially supported by the Ministerial Research Project No.
DEC-2013/11/B/ST8/04358 financed by the Polish National Science Centre.

REFERENCES
1. C.B. York, Compos. Part A 44, 140–148, (2013).
2. C.B. York, J. Aircraft 46(4), 1114-1125, (2009).
3. P. Vannucci and A. Vincenti, Compos. Struct. 79(3), 454-466, (2007).
4. E. Valot, P. Vannucci and G. Verchery, Compos. Struct. 60, 413–429, (2003).
5. R.A. Haynes and E.A. Armanios, AIAA J. 48(12), 2954-2961, (2010).

080002-10
6. J. Loughlan, Compos. Struct. 57, 473–482, (2002).
7. W.C. de Goeij, M.J.L. van Tooren and A. Beukers, Appl. Energ. 63, 191-207, (1999).
8. N.I. Kim, D.K. Shin and M.Y. Kim, Int. J. Mech. Sci. 49, 950–969, (2007).
9. A. Teter and Z. Kolakowski, Compos. Struct. 166, 68–76, (2017).
10. C.B. York, Compos. Struct. 120, 472–482, (2015).
11. R.M. Jones, Mechanics of Composite Materials, 2nd ed. (Taylor & Francis, London, 1999), pp. 190-277.
12. A. Teter, R.J. Mania and Z. Kolakowski, Compos. Struct. 180, 140-147, (2017).
13. Z. Kołakowski and A. Teter, Acta Mechanica et Automatica 10(2), 141-149, (2016).
14. T. Kubiak, Static and dynamic buckling of thin-walled plate structures, (Springer, Verlag, London, 2013),
pp. 47-67.
15. A.M.A. van der Heijden, W.T. Koiter’s Elastic Stability of Solids and Structured. (Cambridge University Press,
2009), pp. 230.
16. E.J. Barbero, Finite element analysis of composite materials, (CRC Press, Taylor & Francis Group, 2008),
pp. 107-119.
17. Abaqus, HTML Documentation, (2016).
18. M. Urbaniak, A. Teter and T. Kubiak, Compos. Struct. 134, 199–208, (2015).
19. P. Rozylo, A. Teter A., H. Debski, P. Wysmulski and K. Falkowicz, Appl. Compos. Mater. 24(5), 1251-1264,
(2017).
20. H. Debski, T. Kubiak and A. Teter, Thin Wall. Struct. 71, 147–154, (2013).
21. H. Debski, A. Teter. and T. Kubiak, Compos. Struct. 118, 28–36, (2014).
22. R.M. Jones, Buckling of bars, plates and shells, (Bull Ridge Publishing, Blacksburg, Virginia, 2006), pp. 308-
319.

080002-11