Non-linear Elasticity

Written by for Version Revised by Revision date Revised version
@rsanfer 7.0.2 @rsanfer 2020-01-28 7.0.2
At one glance:
Solver:
Elasticity
Uses:
SU2_CFD
Prerequisites:
Linear Elasticity
Complexity:
Intermediate

Goals

Once completed the tutorial on Linear Elasticity, we can move on to non-linear problems. This document will guide you throught the following capabilities of SU2:

  • Setting up a non-linear structural problem with large deformations
  • Non-linear (hyperelastic) material law
  • Follower boundary conditions
  • Incremental loads

In this tutorial, we use the same problem definition as for the linear elastic case: a vertical, slender cantilever, clamped in its base, and subject to a horizontal, follower load \(P\) on its left boundary. This is shown in Fig. 1.

ProblemSetup

Resources

You can find the resources for this tutorial in the same structural_mechanics/cantilever folder in the Tutorials repository. You can reuse the mesh file mesh_cantilever.su2 from the Linear Elasticity tutorial, but you will need a new config file, config_nonlinear.cfg.

Background

SU2 has been designed using a finite-deformation framework\(^1\) to account for geometrical and material non-linearities.

We can write the non-linear structural problem via the residual equation

\[\mathscr{S}(\mathbf{u}) = \mathbf{T}(\mathbf{u}) - \mathbf{F}_b - \mathbf{F}_{\Gamma}(\mathbf{u})\]

which has been obtained from the weak formulation of the structural problem defined using the principle of virtual work and discretized using FEM. The state variable \(\mathbf{u}\) corresponds to the displacements of the structure with respect to its reference configuration \(\mathbf{R}\). \(\mathbf{T}(\mathbf{u})\) are the internal equivalent forces, \(\mathbf{F}_b\) are the body forces and \(\mathbf{F}_{\Gamma}(\mathbf{u})\) are the surface forces applied over the external boundaries of the structure.

We can linearize the structural equation as

\[\mathscr{S}(\mathbf{u}) + \frac{\partial \mathscr{S}(\mathbf{u})}{\partial \mathbf{u}} \Delta \mathbf{u} = \mathbf{0}\]

where the tangent matrix is \(\mathbf{K} = \partial \mathscr{S}(\mathbf{u})/\partial \mathbf{u}\). The linearized problem is solved using a Newton method until convergence. Further description of the approach adopted in SU2 can be found in a paper by Sanchez et al\(^2\).

Mesh Description

The cantilever is discretized using 1000 4-node quad elements with linear interpolation. The boundaries are defined as follows:

Mesh

Configuration File Options

We start the tutorial by definining the structural properties. In this case, we will solve a geometrically non-linear problem, with a Neo-Hookean material model.

GEOMETRIC_CONDITIONS= LARGE_DEFORMATIONS
MATERIAL_MODEL= NEO_HOOKEAN

We adopt a plane strain formulation for the 2D problem, and reduce the stiffness of the cantilever by 2 orders of magnitude as compared to the Linear Elasticity tutorial, so that the beam undergoes large deformations:

ELASTICITY_MODULUS=5E7
POISSON_RATIO=0.35
FORMULATION_ELASTICITY_2D = PLANE_STRAIN

Now, we define the Newton-Raphson strategy to solve the non-linear problem. We set the solution method and maximum number of subiterations using

NONLINEAR_FEM_SOLUTION_METHOD = NEWTON_RAPHSON
INNER_ITER = 15

Then, the convergence criteria is set to evaluate the norm of the displacement vector (RMS_UTOL), the residual vector (RMS_RTOL) and the energy (RMS_ETOL). We set the convergence values to -8.0, which is the log10 of the previously defined magnitudes.

CONV_FIELD= RMS_UTOL, RMS_RTOL, RMS_ETOL
CONV_RESIDUAL_MINVAL= -8,

Finally, we define the boundary conditions. In this case, the load applied to the left boundary will be an uniform, follower pressure, that is, its direction will change with the changes in the normals of the surface. This is done as

MARKER_CLAMPED = ( clamped )
MARKER_PRESSURE = ( left, 1E3, upper, 0, right, 0)

Running SU2

Follow the links provided to download the config and mesh files. Execute the code with the standard command

$ SU2_CFD config_nonlinear.cfg

which will show the following convergence history:

+----------------------------------------------------------------+
|  Inner_Iter|      rms[U]|      rms[R]|      rms[E]|    VonMises|
+----------------------------------------------------------------+
|           0|   -1.454599|   -0.001635|   -2.082200|  1.1259e+06|
|           1|   -2.406133|    3.664361|   -0.038068|  1.1224e+06|
|           2|   -3.623182|    2.288350|   -2.827362|  1.1053e+06|
|           3|   -3.096627|   -0.433125|   -5.369087|  1.1229e+06|
|           4|   -4.547761|    0.493533|   -6.401479|  1.1222e+06|
|           5|   -5.306858|   -2.420278|   -9.740703|  1.1221e+06|
|           6|   -6.878854|   -3.854770|  -12.877010|  1.1221e+06|
|           7|   -8.478641|   -6.065417|  -16.078409|  1.1221e+06|
|           8|  -10.081222|   -7.667567|  -19.283559|  1.1221e+06|
|           9|  -11.683669|   -8.697378|  -22.488189|  1.1221e+06|

The code is stopped as soon as the values of rms[U], rms[R] and rms[E] are below the convergence criteria set in the config file. The displacement field obtained in nonlinear.vtk is shown below:

Nonlinear Results

Increasing the load

The magnitude of the residual can limit the convergence of the solver, particularly for those cases in which the problem is very non-linear. Say, for example, that we multiply the load in the left boundary by 4, using

MARKER_PRESSURE = ( left, 4E3, upper, 0, right, 0)

Running SU2 now would result in the divergence of the solver,

+----------------------------------------------------------------+
|  Inner_Iter|      rms[U]|      rms[R]|      rms[E]|    VonMises|
+----------------------------------------------------------------+
|           0|   -0.852539|    0.600425|   -0.878080|  4.5753e+06|
|           1|   -1.130461|    4.628370|    2.207826|  4.4618e+06|
|           2|   -2.198592|    4.658260|    1.539054|  2.8742e+08|
|           3|        -nan|        -nan|        -nan|  0.0000e+00|
|           4|        -nan|        -nan|        -nan|  0.0000e+00|

due to a large imbalance between the internal and external loads. It can be observed that the order of magnitude of the residual vector grows above 104. To solve this problem, we have incorporated an incremental approach to SU2, that applies the load in linear steps,

\[\mathbf{F}_i = \frac{i}{N}\mathbf{F}\]

where \(i\) is the step and \(N\) the maximum number of increments allowed. To apply the load in increments, we need to use the following commands

INCREMENTAL_LOAD = YES
NUMBER_INCREMENTS = 25
INCREMENTAL_CRITERIA = (2.0, 2.0, 2.0)

where \(N\) = 25, and the criteria to apply the increments is that either one of rms[U] > 2.0, rms[R] > 2.0 or rms[E] > 2.0 is met. We add an option to print out the load increment to the screen output:

SCREEN_OUTPUT = (INNER_ITER, LOAD_INCREMENT, RMS_UTOL, RMS_RTOL, RMS_ETOL, VMS)

If we now run the code,

+-----------------------------------------------------------------------------+
|  Inner_Iter|     Load[%]|      rms[U]|      rms[R]|      rms[E]|    VonMises|
+-----------------------------------------------------------------------------+
|           0|     100.00%|   -0.852539|    0.600425|   -0.878080|  4.5753e+06|
|           1|     100.00%|   -1.130461|    4.628370|    2.207826|  4.4618e+06|

Incremental load: increment 1
+-----------------------------------------------------------------------------+
|  Inner_Iter|     Load[%]|      rms[U]|      rms[R]|      rms[E]|    VonMises|
+-----------------------------------------------------------------------------+
|           0|       4.00%|   -2.250479|   -0.797515|   -3.673960|  1.7944e+05|
|           1|       4.00%|   -4.005792|    2.096002|   -3.207727|  1.7948e+05|
|           2|       4.00%|   -5.555428|   -0.881184|   -9.113457|  1.7949e+05|
|           3|       4.00%|   -7.611025|   -4.228316|  -14.254076|  1.7949e+05|
|           4|       4.00%|  -10.343696|   -7.532184|  -19.528442|  1.7949e+05|
|           5|       4.00%|  -13.138042|   -8.791438|  -24.949847|  1.7949e+05|

given that 4.628370 > 2.0 and 2.207826 > 2.0, the load is applied in \(\Delta \mathbf{F} = 1/25 \mathbf{F}\), completing the simulation in the 25th step,

Incremental load: increment 25
+-----------------------------------------------------------------------------+
|  Inner_Iter|     Load[%]|      rms[U]|      rms[R]|      rms[E]|    VonMises|
+-----------------------------------------------------------------------------+
|           0|     100.00%|   -2.188244|   -0.786306|   -3.616397|  4.2701e+06|
|           1|     100.00%|   -3.417010|    2.241142|   -2.911125|  4.2325e+06|
|           2|     100.00%|   -2.802795|   -0.383540|   -4.853108|  4.1931e+06|
|           3|     100.00%|   -3.461966|    1.059169|   -5.203155|  4.2102e+06|
|           4|     100.00%|   -4.053497|   -0.412173|   -7.024221|  4.2103e+06|
|           5|     100.00%|   -4.505028|   -1.211625|   -8.188042|  4.2090e+06|
|           6|     100.00%|   -5.044900|   -2.434771|   -9.288963|  4.2093e+06|
|           7|     100.00%|   -5.572267|   -3.418852|  -10.345812|  4.2092e+06|
|           8|     100.00%|   -6.099258|   -4.140827|  -11.400141|  4.2093e+06|
|           9|     100.00%|   -6.625915|   -4.693663|  -12.453409|  4.2093e+06|
|          10|     100.00%|   -7.152645|   -5.224056|  -13.506885|  4.2093e+06|
|          11|     100.00%|   -7.679350|   -5.750731|  -14.560291|  4.2093e+06|
|          12|     100.00%|   -8.206063|   -6.277523|  -15.613718|  4.2093e+06|
|          13|     100.00%|   -8.732773|   -6.804154|  -16.667138|  4.2093e+06|
|          14|     100.00%|   -9.259484|   -7.329682|  -17.720560|  4.2093e+06|
|          15|     100.00%|   -9.786195|   -7.846390|  -18.773983|  4.2093e+06|
|          16|     100.00%|  -10.312907|   -8.293905|  -19.827405|  4.2093e+06|

The displacement field is now

Nonlinear Incremental Results

References

\(^1\) Bonet, J. and Wood, R.D. (2008), Nonlinear Continuum Mechanics for Finite Element Analysis, Cambridge University Press

Attribution

If you are using this content for your research, please kindly cite the following reference (reference \(^2\) in the text above) in your derived works:

Sanchez, R. et al. (2018), Coupled Adjoint-Based Sensitivities in Large-Displacement Fluid-Structure Interaction using Algorithmic Differentiation, Int J Numer Meth Engng, Vol 111, Issue 7, pp 1081-1107. DOI: 10.1002/nme.5700

This work is licensed under a Creative Commons Attribution 4.0 International License
Creative Commons Licence


Improve this page