| 1 |
"""This demo uses PETSc's TAO solver for nonlinear (bound-constrained) |
| 2 |
optimisation problems to solve a buckling problem in FEniCS. |
| 3 |
We consider here a hyperelastic beam constrained in a box |
| 4 |
under axial compression. |
| 5 |
|
| 6 |
The box is designed such that this beam will lose stability and move |
| 7 |
upwards (and not downwards) in order to minimise the potential energy.""" |
| 8 |
|
| 9 |
# Copyright (C) 2014 Tianyi Li |
| 10 |
# |
| 11 |
# This file is part of DOLFIN. |
| 12 |
# |
| 13 |
# DOLFIN is free software: you can redistribute it and/or modify |
| 14 |
# it under the terms of the GNU Lesser General Public License as published by |
| 15 |
# the Free Software Foundation, either version 3 of the License, or |
| 16 |
# (at your option) any later version. |
| 17 |
# |
| 18 |
# DOLFIN is distributed in the hope that it will be useful, |
| 19 |
# but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 20 |
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 21 |
# GNU Lesser General Public License for more details. |
| 22 |
# |
| 23 |
# You should have received a copy of the GNU Lesser General Public License |
| 24 |
# along with DOLFIN. If not, see <http://www.gnu.org/licenses/>. |
| 25 |
# |
| 26 |
# First added: 2014-07-19 |
| 27 |
|
| 28 |
|
| 29 |
from dolfin import * |
| 30 |
import matplotlib.pyplot as plt |
| 31 |
|
| 32 |
|
| 33 |
if not has_petsc(): |
| 34 |
print("DOLFIN must be compiled at least with PETSc 3.6 to run this demo.") |
| 35 |
exit(0) |
| 36 |
|
| 37 |
# Read mesh and refine once |
| 38 |
mesh = Mesh("../buckling.xml.gz") |
| 39 |
mesh = refine(mesh) |
| 40 |
|
| 41 |
# Create function space |
| 42 |
V = VectorFunctionSpace(mesh, "Lagrange", 1) |
| 43 |
|
| 44 |
# Create solution, trial and test functions |
| 45 |
u, du, v = Function(V), TrialFunction(V), TestFunction(V) |
| 46 |
|
| 47 |
# Elasticity parameters |
| 48 |
E, nu = 10.0, 0.3 |
| 49 |
mu = Constant(E/(2.0*(1.0+nu))) |
| 50 |
lmbda = Constant(E*nu/((1.0+nu)*(1.0-2.0*nu))) |
| 51 |
|
| 52 |
# Compressible neo-Hookean model |
| 53 |
I = Identity(mesh.geometry().dim()) |
| 54 |
F = I + grad(u) |
| 55 |
C = F.T*F |
| 56 |
Ic = tr(C) |
| 57 |
J = det(F) |
| 58 |
psi = (mu/2)*(Ic-2)-mu*ln(J)+(lmbda/2)*(ln(J))**2 |
| 59 |
|
| 60 |
# Surface force |
| 61 |
f = Constant((-0.08, 0.0)) |
| 62 |
|
| 63 |
# The displacement u must be such that the current configuration |
| 64 |
# doesn't escape the box [xmin, xmax] x [ymin, ymax] |
| 65 |
constraint_u = Expression(("xmax-x[0]", "ymax-x[1]"), xmax=10.0, ymax=2.0, degree=1) |
| 66 |
constraint_l = Expression(("xmin-x[0]", "ymin-x[1]"), xmin=0.0, ymin=-0.2, degree=1) |
| 67 |
u_min = interpolate(constraint_l, V) |
| 68 |
u_max = interpolate(constraint_u, V) |
| 69 |
|
| 70 |
# Symmetry condition (to block rigid body rotations) |
| 71 |
class Left(SubDomain): |
| 72 |
def inside(self, x, on_boundary): |
| 73 |
return on_boundary and near(x[0], 0) |
| 74 |
class Right(SubDomain): |
| 75 |
def inside(self, x, on_boundary): |
| 76 |
return on_boundary and near(x[0], 10) |
| 77 |
boundaries = MeshFunction("size_t", mesh, mesh.topology().dim()-1) |
| 78 |
boundaries.set_all(0) |
| 79 |
left = Left() |
| 80 |
left.mark(boundaries, 1) |
| 81 |
right = Right() |
| 82 |
right.mark(boundaries, 2) |
| 83 |
ds = Measure('ds', domain=mesh, subdomain_data=boundaries) |
| 84 |
bc = DirichletBC(V, Constant([0.0, 0.0]), boundaries, 1) |
| 85 |
bc.apply(u_min.vector()) |
| 86 |
bc.apply(u_max.vector()) |
| 87 |
|
| 88 |
# Variational formulation |
| 89 |
elastic_energy = psi*dx - dot(f, u)*ds(2) |
| 90 |
grad_elastic_energy = derivative(elastic_energy, u, v) |
| 91 |
H_elastic_energy = derivative(grad_elastic_energy, u, du) |
| 92 |
|
| 93 |
# Define the minimisation problem by using OptimisationProblem class |
| 94 |
class BucklingProblem(OptimisationProblem): |
| 95 |
|
| 96 |
def __init__(self): |
| 97 |
OptimisationProblem.__init__(self) |
| 98 |
|
| 99 |
# Objective function |
| 100 |
def f(self, x): |
| 101 |
u.vector()[:] = x |
| 102 |
return assemble(elastic_energy) |
| 103 |
|
| 104 |
# Gradient of the objective function |
| 105 |
def F(self, b, x): |
| 106 |
u.vector()[:] = x |
| 107 |
assemble(grad_elastic_energy, tensor=b) |
| 108 |
|
| 109 |
# Hessian of the objective function |
| 110 |
def J(self, A, x): |
| 111 |
u.vector()[:] = x |
| 112 |
assemble(H_elastic_energy, tensor=A) |
| 113 |
|
| 114 |
# Create the PETScTAOSolver |
| 115 |
solver = PETScTAOSolver() |
| 116 |
|
| 117 |
# Set some parameters |
| 118 |
solver.parameters["method"] = "tron" |
| 119 |
solver.parameters["monitor_convergence"] = True |
| 120 |
solver.parameters["report"] = True |
| 121 |
|
| 122 |
# Uncomment this line to see the available parameters |
| 123 |
# info(parameters, True) |
| 124 |
|
| 125 |
# Parse (PETSc) parameters |
| 126 |
parameters.parse() |
| 127 |
|
| 128 |
# Solve the problem |
| 129 |
solver.solve(BucklingProblem(), u.vector(), u_min.vector(), u_max.vector()) |
| 130 |
|
| 131 |
# Save solution in XDMF format if available |
| 132 |
out = XDMFFile(mesh.mpi_comm(), "u.xdmf") |
| 133 |
if has_hdf5(): |
| 134 |
out.write(u) |
| 135 |
elif MPI.size(mesh.mpi_comm()) == 1: |
| 136 |
encoding = XDMFFile.Encoding.ASCII |
| 137 |
out.write(u, encoding) |
| 138 |
else: |
| 139 |
# Save solution in vtk format |
| 140 |
out = File("u.pvd") |
| 141 |
out << u |
| 142 |
|
| 143 |
# Plot the current configuration |
| 144 |
plot(u, mode="displacement", wireframe=True, title="Displacement field") |
| 145 |
plt.show() |