23rd - 26rd June 2014 9th OpenFOAM Workshop

University of Zagreb
Croatia

Nonlinear Solid Mechanics

in OpenFOAM

Philip Cardiff

Željko Tuković
Aleksandar Karač
Alojz Ivanković

School of Mechanical & Materials Engineering

University College Dublin
Irish Centre for Composites Research
!
!

!
!
Background &
Motivation
Background

Finite Element method is not the only way!

Finite Volume method can be applied to solids;

it has slowly been gaining momentum since
early 90’s.

FV method developed for nonlinearity, so it

Finite
should be suitable! Elephant
Mesh
 I Z
n·Dd Q d⌦ = 0
⌦
2
@ u
⇢ 2 =r·
@t
@T 2
⇢C = kr T
Linear Finite Volume @t

Solid Mechanics
Governing Equations

Strong Form of Governing Equation Momentum

Z Z Z
@ @u
⇢ d⌦ = r· d⌦ + ⇢f b d⌦
@t ⌦ @t ⌦ ⌦

Finite Volume Method directly uses strong form

 Weak Form of Governing Equation Momentum

Z ✓ ◆ Z Z
@ @u
⇢ · w d⌦ = · rw d⌦ + ⇢f b · w d⌦
@t ⌦ @t ⌦ ⌦
Z
+ t·w d
t
 Weak Form of Governing Equation Momentum

weighting function
Z ✓ ◆ Z Z
@ @u
⇢ · w d⌦ = · rw d⌦ + ⇢f b · w d⌦
@t ⌦ @t ⌦ ⌦
Z
+ t·w d
Finite Element Method uses weak form t
Development of the Mathematical Model

Isotropic Hooke’s law assuming small strains

1 T
= 2µ✏ + tr(✏)I ✏= ru + ru
2

T
= µru + µru + tr(u)I
Development of the Mathematical Model

Insert constitutive relation into governing equation

Z Z Z
@ @u ⇥ ⇤
⇢ d⌦ = r · µru + µruT + tr(ru)I d⌦ + ⇢f b d⌦
@t ⌦ @t ⌦ ⌦

{
 Z Z Z Z
@ @u ⇥ ⇤
⇢ d⌦ = µr2 u d⌦ + r · µruT + tr(ru)I d⌦ + ⇢f b d⌦
@t ⌦ @t ⌦ ⌦ ⌦
 laplacian term implicitly approximated
using compact molecule
Z Z Z Z
@ @u ⇥ ⇤
⇢ d⌦ = µr2 u d⌦ + r · µruT + tr(ru)I d⌦ + ⇢f b d⌦
@t ⌦ @t ⌦ ⌦ ⌦
 laplacian term implicitly approximated
using compact molecule
Z Z Z Z
@ @u ⇥ ⇤
⇢ d⌦ = µr2 u d⌦ + r · µruT + tr(ru)I d⌦ + ⇢f b d⌦
@t ⌦ @t ⌦ ⌦ ⌦

Z Z
2
(2µ + )r u d⌦ 2
(µ + )r u d⌦
better convergence with
⌦ ⌦ over-relaxed approach
FEM & FVM Comparison
FVM
Pros
attractively simple
strongly conservative
memory efficient
straight-forward nonlinearity
suitable for FSI/cracks
general polyhedral cells
for solids
Cons
segregated solution not
always efficient
higher order elements are
more difficult
less development
 FEM
Pros
efficient for linear problems
higher order elements
plethora of development
inclusion of discontinuities
Cons
weakly conservative
memory hungry
different shape functions
for every shape element
FSI
 I Z
n·Dd Q d⌦ = 0
⌦
2
@ u
⇢ 2 =r·
@t
@T 2
⇢C = kr T
Nonlinear Finite Volume @t

Solid Mechanics
Different Types of Nonlinearity

Three Types

Geometric System of Linear Equations

Material Ax = b linear

Loading A(x) x = b(x) nonlinear

Geometric
large strains, changes in integration areas/volumes
engineering stress & strain
undeformed deformed
F l L
s = ✏s =
A L
F
2nd Piola-Kirchhoff stress & Green strain
F
FL l2 L2
l S= ✏=
L A l 2L
true stress & strain

A a F l
= ✏ = ln
a L
Material
nonlinear stress strain curve e.g. plasticity

nonlinear elastic elastoplastic

Loading
e.g. contact

displace top of block

force

displacement
Solution Procedure

Update nonlinear terms

Segregated Solution
do
solve (Aui = b) for i=1:3
update A(ui) and b(ui)
while A(ui) && b(ui) not converged

No change to solution procedure

Example: elasticPlasticSolidFoam
do
solve
(
rho*fvm::ddt(DU)
==
fvm::laplacian(2*mu + lambda, DU)
+ fvc::div
(
mu*gradDU.T() + lambda*tr(gradDU)*I
- (mu + lambda)*gradDU
- DEpsilonP
)
);

update gradDU
update DEpsilonP // plastic strain increment
while !converged
Geometrically Nonlinear Mathematical Models

Two general methods to deal with geometric nonlinearity:

Total Lagrangian elasticNonLinTLSolidFoam

Updated Lagrangian elasticNonLinULSolidFoam

Choice between the two approaches typically depends on

the constitutive model
Total Lagrangian Mathematical Model

Linear momentum equation neglecting inertia and body forces

r· =0
equivalent to
I
n· d| | = 0
 1 T Cauchy stress in terms of
= F ·S·F 2nd Piola Kirchhoff stress
J

T
= JF · o Nanson’s formula
Linear momentum equation may be written in terms of Cauchy
stress or 2nd Piola-Kirchhoff stress

I I ⇣ ⌘
T
n· d| | = no · S · F d| o| =0
o
Constitutive Law

St.Venant-Kirchhoff law i.e. Large strain/rotation form of Hooke’s law

S = 2µE + tr(E)I

Green strain

1 T
E= ru + ru + ru · ru
2
Total Lagrangian Mathematical Model

2
@ u 2
⇢ =(2µ + )r u
@t
T
+ r · (µru + tr(ru)I (µ + )ru)
✓ ◆
T 1 T
+ r · µru · ru + tr(ru · ru )I
2
+ r · (S old · ru)
elasticNonLinTLSolidFoam

fvVectorMatrix UEqn
(
rho*fvm::d2dt2(U)
==
fvm::laplacian(2*mu + lambda, U)
+ fvc::div
(
- (mu + lambda)*gradU
+ mu*gradU.T()
+ mu*(gradU & gradU.T())
+ lambda*tr(gradU)*I
+ 0.5*lambda*tr(gradU & gradU.T())*I
+ (sigma & gradU)
)
);
Test Cases
Plasticity

elasticPlasticSolidFoam
Frictional Contact

elasticSolidFoam
 angular mesh to the Delaunay dual mesh using OpenFOAM utility polyDualMesh. The mesh contains 50,247 cells with
approximate execution times of 30 to 40 s. The stress results for the polygonal mesh are shown in Fig. 5(b) and are shown
to agree closely with the analytical solutions.
P. Cardiff et al. / Comput. Methods Appl. Mech. Engrg. 268 (2014) 318–335 In all previous test cases, the explicit divergence of stress field has been calculated using the full gradient larger compu-

Large Rotations & Orthotropic Elasticity

tational molecule. Initially, however, as discussed in Appendix (A), this term has been calculated using the Laplacian operator
s the applied distant load in the positive x direction, h is the angle around the circumference of theemploys
which hole with 0! on
a compact computational module, but it has been found that the solution convergence may be poor. For the
ve x axis. Parameters k and n are given respectively by: current test case, execution time is approximately 450 s and requires 2,200 outer iterations, compared with 50 s and 30 outer
sffiffiffiffiffi iterations when employing the full gradient larger computational molecule.
Ex
; ð41Þ
Ey 4.2. Rotating orthotropic plate with a pressurised hole

elasticOrthoNonLinULSolidFoam
To examine that the developed updated Lagrangian procedure correctly rotates the constitutive stiffness tensor and the
stress and strain tensors, a rotating circular plate with a pressurised hole is considered. As the plate is rotated, the location of
the maximum hoop stress rotates with the corresponding rotating principal material directions. The test case geometry,
shown in Fig. 6(a), consists of a circular plate containing a circular hole with the ratio of outer radius to inner radius of
100:1. The mesh, shown in Fig. 6(b), contains 160,000 hexahedral cells and has been generated using OpenFOAM meshing
utility blockMesh. The mesh is graded towards the hole to reduce the total number of cells required. Plane stress conditions
are assumed.
The hole is subjected to a pressure of 1 MPa and the outer plate surface is rotated through 180! in increments of 1!. The
displacement increment for boundary face f is:
0 1
cos h " sin h 0
B C
duf ¼ @ sin h cos h 0 A # Cf " Cf ð43Þ
0 0 1
where h is the increment of rotation, and C f is the positional vector of the boundary face centre.

Fig. 7. Rotating orthotropic plate with pressurised hole – hoop stress for different

Fig. 8. Rotating orthotropic plate with pressurised hole – stress distribution at 0! rotation (in MPa).
elasticOrthoNonLinULSolidFoam
Summary & Conclusions
Summary & Conclusions

OpenFOAM solidMechanics solvers capable of all forms of linearity with no

change to underlying segregated solution procedure

Complex nonlinear problems treated in a implicit manner

Method is very memory efficient for large meshes, and shows good parallel speedup

Future Steps
Consider the use of newly developed block-coupled solid solver for nonlinearity with
Newton iterations, like in FE;
Clean integration of all methods into new FSI framework.
9th OpenFOAM Workshop University of Zagreb
23rd - 26rd June 2014 Croatia

Nonlinear Solid Mechanics

in OpenFOAM

Philip Cardiff

Željko Tuković
Aleksandar Karač
Alojz Ivanković

School of Mechanical & Materials Engineering

University College Dublin
Irish Centre for Composites Research
!
!

!
!