GEOMETRIC NONLINEAR ANALYSIS

The approach for solving problems with geometric nonlinearity is presented. The
ESAComp solution relies on Elmer open-source computational tool [1] for multi-
physics problems. Elmer Shell Solver has been developed by Dr. Mikko Lyly from
CSC - Scientific Computing Ltd. and Dr. Petri Kere from Tampere University of
Technology / Department of Mechanics and Design. Their contribution is acknowl-
edged. This theory document is a quite direct quotation from their works [2–4]
and therefore, notation and conventions differ from the ones used in ESAComp.
The deviations between the presented approach and ESAComp implementation are
explained in the last section.

1 REISSNER-MINDLIN-VON KÁRMÁN TYPE SHELL FACET MODEL

The plate bending problem is formulated for a thin or moderately thick laminated
composite plate which in its undeformed configuration occupies the region Ω ×
(−t/2, t/2), where Ω ⊂ R2 is the midsurface and t > 0 is the laminate thickness.
The kinematical unknowns in the model are transverse deflection w, in-plane dis-
placement u = (ux , uy ), rotation of the middle surface β = (βx , βy ), and drilling
rotation ω. The plate is subjected to the in-plane load f = (fx , fy ) and the trans-
verse pressure g.

We will use standard notation of tensor calculus. Dyadic and index notation with
summation convention over repeated indices are used in parallel. Latin indices take
their values in the set {1, 2, 3} and Greek indices in the set {1, 2}.

1.1 Constitutive relation for a single layer

Let us denote by ei and ēj the cartesian basis vectors for the so called 123-coordinate
system of a single ply, and for the xyz-system of material coordinates common to

1
all plies, respectively. In the material coordinate system, i.e., the laminate coor-
dinate system, the layer system has been rotated by a positive counter clockwise
angle θ about the z-axis. Hence, we define the transformation matrix between the
two coordinate systems as T = Tij = ei · ēj , or
 
 cos θ sin θ 0 
 
T =
 − sin θ cos θ 0 

(1)
 
 
0 0 1

For linear orthotropic materials in plane stress state the constitutive relation for
each ply has the form
σ=Q: (2)
where σ = σij = σji is the second order stress tensor,  = ij = ji is the strain
tensor, and Q = Qijkl = Qjikl = Qijlk = Qklij is the fourth order tensor of elastic
stiffness coefficients. In the laminate coordinate system the constitutive equation is
written as
σ̄ = Q̄ : ¯ (3)
where σ̄ij = Tip Tjq σpq is the laminate stress, ¯ij = Tip Tjq pq is the laminate strain,
and Q̄ijkl = Tip Tjq Tkr Tls Qpqrs is the tensor of stiffness coefficients in the laminate
coordinate system.

The six independent non-zero components of Q are computed using the orthotropic
material engineering constants E1 , E2 , ν12 , ν21 = ν12 E2 /E1 , G12 , G23 , and G31 [5]
as
Q1111 = E1 /(1 − ν12 ν21 ), Q2222 = E2 /(1 − ν12 ν21 ),
Q1122 = ν12 E2 /(1 − ν12 ν21 ), (4)
Q1212 = G12 , Q2323 = G23 , Q3131 = G31

1.2 Kinematic relations for a laminate

The kinematic relations for a laminate are considered in the xyz-coordinate system.
For notational simplicity, laminate stresses and strains in the xyz-coordinate system
are in the following denoted without bar symbol.

Using the classical kinematic assumptions of Reissner, Mindlin, and Von Kármán
the laminate strain is obtained from

 = ε(u) + ϕ(u, w) − zε(β) (5)

and
3α = γα (w, β), 33 = 0 (6)

2
where z := x3 , ε is the linear strain tensor, ϕ is the nonlinear membrane strain
tensor, and γ the transverse shear strain vector, viz.

1
ε(u) = (∇u + ∇uT ) (7)
2
1
ϕ(u, w) = (∇ux ⊗ ∇ux + ∇uy ⊗ ∇uy + ∇w ⊗ ∇w) (8)
2
γ(w, β) = ∇w − β (9)

Let us note that the original Von Kármán strains [6] do not have the quadratic in-
plane displacement gradients in the membrane tensor ϕ.

1.3 Constitutive relations for a laminate

In plane-stress state, the laminate membrane stress resultants N (forces per unit
length) and bending moment resultants M (moments per unit length) are obtained
by integration of the stress resultants of all layers zk−1 < z < zk , k = 1, . . . , n,
over the thickness of the laminate as
Z t/2 X Z zk
N= σdz = σdz (10)
−t/2 k zk−1

Z t/2 X Z zk
M= σzdz = σzdz (11)
−t/2 k zk−1

Furthermore, the resultant transvere shear forces S are obtained from

Z t/2 X Z zk
S= σ3α dz = σ3α dz (12)
−t/2 k zk−1

Using the constitutive equation and the kinematic relations of Reissner, Mindlin,
and Von Kármán we get the following constitutive relations for the laminate

N (u, w, β) = A : [ε(u) + ϕ(u, w)] + B : ε(β) (13)

M (u, w, β) = B : [ε(u) + ϕ(u, w)] + D : ε(β) (14)

?
S(w, β) = A · γ(w, β) (15)

The tensors A, B, D, and A? , are defined according to the Classical Lamination

Theory (CLT) [5] as
X Z zk
(zk − zk−1 )Q̄(k)
X
A= Q̄ dz = (16)
k zk−1 k

3
 X Z zk 1X 2 2
B= (zk − zk−1
Q̄z dz = )Q̄(k) (17)
k zk−1 2 k
X Z zk 1
Q̄z 2 dz = (zk3 − zk−1
3
)Q̄(k)
X
D= (18)
k z k−1 3 k
X Z zk (k)
A?ij =
X
Q̄3i3j dz = (zk − zk−1 )Q̄3i3j (19)
k zk−1 k

where Q̄(k) defines the constitutive relation for linear orthotropic materials in plane
stress state for layer k in the laminate coordinate system.

1.4 The shell facet model

The functions u, w, β, ω are determined from the condition that they minimize the
potential energy of the plate. The energy is defined as

1Z
Π(u, w, β, ω) = N (u, w, β) : [ε(u) + ϕ(u, w)] dΩ
2 Ω
1Z 1Z
+ M (u, w, β) : ε(β) dΩ + S(w, β) · γ(w, β) dΩ (20)
2 ZΩ Z 2 Ω Z
+C [ω − rot(u)]2 dΩ − f · u dΩ − gw dΩ
Ω Ω Ω

where C > 0 is a penalty parameter for imposing the condition ω = rot(u) (see
[7]), and
∂ux ∂uy
rot(u) = − (21)
∂y ∂x
Substituting the constitutive equations in Eq. 20, we get

1Z Z
ε(u) : A : ε(u) dΩ + ε(u) : B : ε(β) dΩ
Π(u, w, β, ω) =
2 Ω Ω
1Z 1Z
+ ε(β) : D : ε(β) dΩ + γ(w, β) · A? · γ(w, β) dΩ (22)
2 Ω 2 Ω
Z
1Z
+C [ω − rot(u)]2 dΩ + ϕ(u, w) : A : ϕ(u, w) dΩ
Z Ω Z 2 Ω Z Z
+ ε(u) : A : ϕ(u, w) dΩ + ε(β) : B : ϕ(u, w) dΩ − f · u dΩ − gw dΩ
Ω Ω Ω Ω

2 GEOMETRICALLY NONLINEAR STRUCTURAL ANALYSIS

To obtain the load-displacement curve and to sudy the stability behaviour, the non-
linear equations are solved iteratively by Riks’ method with Crisfield’s elliptical

4
constraint for arc length [8–11]. The differential equilibrium equations of the min-
imization problem are obtained using standard variational calculus and integration
by parts. The linearized equations are then discretized by the finite element method.

The algorithm is based on the Newton iteration, which follows the principal equi-
librium path. The resulting linear system is solved by minimizing the energy, and if
there is no minimum, i.e., if there is a negative coefficient matrix, it will terminate.
All solutions presented in the paper are obtained for positive definite coefficient
matrices. Hence, in the post-buckling region the algorithm follows the principal
equilibrium path with the minimal stiffness.

Let the equilibrium equation be defined as

F (q, λ) ≡ λP (q) − R(q) = 0 (23)

where the unbalanced or residual force is denoted by F , P and R are the external
and internal loads, respectively, q = (ux , uy , w, βx , βy , ω) are the nodal displace-
ments of the FE-solution and λ is the load scaling factor serving as a control param-
eter of the system. In this work, the reference load P is assumed to be independent
of q.

Using the arc-length methods for solving the nonlinear equilibrium equations, the
load-displacement constraint G is added to the system


 F (q, λ)

=0
H(q, λ) =  (24)
 G(q, λ) =0

Solution of the nonlinear system is obtained by an incremental approach with the

elliptical constraint and the algorithm as follows:

(1) Given a solution (qe , λe ) to Eq. 24, define G(q, λ) = (q − qe )T W (q − qe ) +

c2 (λ−λe )2 −s2 , where W is a positive weighting matrix for the displacements,
c is a load scaling parameter, and s is the arc length.
(2) Linearize in the intermediate state (qi , λi ) as

 F (qi , λi ) + ∂F (qi , λi )(q

− qi ) + ∂F
(q , λi )(λ − λi ) = 0
∂q ∂λ i
H(q, λ) ≈ (25)
∂G ∂G
 G(q
i , λi ) + (qi , λi )(q − qi ) + (q , λi )(λ − λi ) = 0

∂q ∂λ i

or

 F (qi , λi ) + K(q

− qi ) + P (λ − λi ) = 0
 G(q
i , λi ) + 2(qi − qe )T W (q − qi ) + 2c2 (λi − λe )(λ − λi ) = 0


5
 where K = ∂F ∂q
(qi , λi ) is the tangential to the equilibrium path at an equilib-
rium state.
(3) Solve (q, λ) from Eq. 25. If |H(q, λ)| ≤ δ goto Step 4 else set (qi , λi ) = (q, λ)
and goto Step 2.
(4) Set (qe , λe ) = (q, λ) and goto Step 1.

3 FE-IMPLEMENTATION

In the FE-implementation bilinear stabilized MITC plate elements [12,13] are used.
The shear energy term is modified as

1 t2 Z
γh (w, β) · A? · γh (w, β) dΩ (26)
2 t2 + αh2 Ω
where α > 0 is a numerical stabilization parameter, h is the mesh parameter, i.e.,
the largest side length, and γh is the reduced shear strain [12,13]. For the bilinear
quadrilateral element used in the simulation, the reduced shear is defined locally
such that  
a
 K + c y
K 
γh (w, β)|K =   (27)
bK + d K x
for every element K. The parameters aK , bK , cK , and dK are determined from the
condition Z
[γ(w, β) − γh (w, β)] · τ ds = 0 (28)
e
for every edge e of K. Here τ is the tangent to the edge. In the computation, the
in-plane forces N and the bending moments M are obtained consistently from the
constitutive equations. Shear forces are computed from

t2
S(w, β) = A? · γh (w, β) (29)
t2 + αh2

4 ESAComp IMPLEMENTATION

In ESAComp implementation the reference surface is not limited to the mid-surface

of the shell. It is determined from the problem definition.

Instead of using CLT the computation of the out-of-plane shear stress distribu-
tion and stiffness (see equation 19) is based on the theory developed at German
Aerospace Center, DLR (Rohwer and Rolfes). This is described in other theoretical
documentation of ESAComp.

6
Instead of using Crisfield’s elliptical constraint for arc length the load controlled
method is applied. In the load controlled method W =0 and c=1 (see equation 24).

Shear Stabilization Parameter alpha (see equation 26) and Drilling Stabilization
Parameter C (see equation 20) have fixed values of 0.2. Those can be manually
adjusted from template files (*.sif).

The analysis approach supports large deflection but not large rotation.

References

[1] Elmer Models Manual, CSC - Scientific Computing Ltd., 2007, Elmer web site
www.csc.fi/elmer.
[2] P. Kere, M. Lyly, Reissner-Mindlin-von Kármán type shell facet model for buckling
simulation of imperfect cylindrical composite shells, 2009. In: Ambrosio, J. et.
al. (eds). 7th EUROMECH Solid Mechanics Conference, Lisbon, Portugal, 7-11
September 2009, pp. 369-370.
[3] P. Kere, M. Lyly, On Post-Buckling Analysis and Experimental Correlation of
Cylindrical Composite Shells with Reissner-Mindlin-Von Kármán Type Facet Model,
Computers & Structures, vol. 86, pp. 1006–1013, 2008.
[4] P. Kere, M. Lyly., Reissner-Mindlin-Von Kármán Type Shell Facet Model for Buckling
Simulation of Imperfect Cylindrical Composite Shells, Mechanics of Advanced
Materials and Structures, Vol. 18, pp. 115-124, 2011.
[5] J. N. Reddy, Mechanics of Laminated Composite Plates and Shells - Theory and
Analysis, CRC Press, Florida, 2004.
[6] T. von Kármán, H. S. Tsien, The Buckling of Thin Cylindrical Shells under Axial
Compression, Journal of the Aeronautical Sciences, vol. 8, pp. 303–312, 1941.
[7] T. J. R. Hughes, F. Brezzi, On drilling degrees of freedom, Computer Methods in
Applied Mechanics and Engineering, vol. 72, pp. 105–121, 1989.
[8] R. Kouhia, On the Solution of Non-linear Finite Element Equations, Computers and
Structures, vol. 44, pp. 243–254, 1992.
[9] E. Ramm, Strategies for Tracing the Nonlinear Response Near Limit Points, Nonlinear
Finite Element Analysis in structural Mechanics, Springer Verlag, Bochum, 1981.
[10] W. C. Rheinboldt, Numerical Analysis of Parametrized Nonlinear Equations, Wiley,
New York, 1986.
[11] E. Riks, The Application of Newton’s Method to The Problem of Elastic Stability,
Journal of Applied Mechanics, vol. 39, pp. 1060–1065, 1972.
[12] F. Brezzi, M. Fortin, R. Stenberg, Error analysis of mixed-interpolated elements for
Reissner-Mindlin plates, Math. Models Methods Appl. Sci., vol. 1, pp. 125–151, 1991.

7
[13] M. Lyly, On the connection between some linear triangular Reissner-Mindlin plate
bending elements, Numer. Math., vol. 85, pp. 77–107, 2000.