You are on page 1of 2

Computational Multiscale Modeling (151-0520-00L) March 19, 2018

Spring 2018 Prof. Dennis M. Kochmann, ETH Zürich

Project #1

assigned: Wednesday, March 21st, 2018


due: Wednesday, April 11th, 2018, 17:00
please hand in to one of the TAs in LEE N203

Homogenization in Linear Elasticity:

Let us investigate the effective behavior of a two-phase, linear elastic composite material composed of a stiff
metal phase (1) and a soft polymer phase (2). Both phases are assumed isotropic with
E1 = 210 GPa E2 = 1 GPa
ν1 = 0.4 ν2 = 0.48

For simplicity, we consider a square-shaped RVE in 2D, which is discretized by the finite element method using
CST elements and linearized kinematics. Each element is assigned to either of the two materials. The RVE
is deformed by imposing boundary conditions on its perimeter. We will impose the three kinds of boundary
conditions discussed in class:

(1) affine displacement boundary conditions: u(x) = ε∗ x


(2) uniform traction boundary conditions: t(x) = σ ∗ n
(3) periodic boundary conditions: u+ = u− + ε∗ (x+ − x− )

For given boundary conditions, our code solves for the displacements of all inner nodes and, subsequently,
computes the average stresses and strains of the RVE by averaging over all elements.
The effective stiffness tensor can be obtained in a variety of ways, the simplest to implement is to impose
three linearly independent choices of ε̃ = {ε11 , ε22 , ε12 }, for each of which an average σ̃ = {σ11 , σ22 , σ12 } is
calculated (or vice-versa). Using those values, we may solve ε̃ = C∗ σ̃ for C∗ which is the effective stiffness
matrix in Voigt notation (note that for reasons of numerical noise we may not assume its symmetry).

Problem 1: laminate (20 points).

Let us compute the effective stiffness tensor of a laminate with equal volume fractions of v1 = v2 = 50%,
using each of the three types of boundary conditions. From composite theory1 , we know the exact answer for
the effective laminate stiffness tensor:
 
136.537 13.697 0
C∗ =  13.697 17.231 0  GPa
0 0 1.345

(a) Using each of the three types of BCs, calculate the effective stiffness tensor C∗ of the two-phase laminate.
Extract Young’s moduli E1 and E2 in the two directions perpendicular and along the laminate interfaces
(b) In terms of the directional Young’s moduli, which of the three BCs best approximates the exact result,
which over/underpredict the results?

1
See, e.g., G.W. Milton’s “The Theory of Composites” (Cambridge University Press, 2004), Section 9.2.

1
Computational Multiscale Modeling (151-0520-00L) March 19, 2018
Spring 2018 Prof. Dennis M. Kochmann, ETH Zürich

Problem 2: random composite, RVE convergence (80 points).

Next, let us consider a composite with random microstructure (i.e., for given volume fractions, elements are
randomly assigned to either of the two phases). Due to the random assignments, two realizations of the same
volume fraction but with different arrangements may yield different effective behavior.

(a) Let us demonstrate the convergence due to ensemble enlargement: for an RVE with 10 × 10 nodes
per side, calculate the effective C∗ , using periodic BCs only (we may conclude that the general behavior
is the same for all three types of BCs). Repeat the calculation 200 times. As representative moduli, plot
the evolution of the mean values of E1 and E2 vs. the increasing number of realizations (see slide 5 of
the project presentation). Does the effective stiffness tensor converge to isotropy?

(b) Let us demonstrate convergence due to sample enlargement: repeat the above calculation for RVEs
having 5 × 5, 10 × 10, . . . , 30 × 30 nodes per side (each with 50 realizations). Plot all obtained data for
E1 and E2 in a 2D map with all realizations on the y-axis and all sample sizes on the x-axis, as shown
in class (see slides 7/8 of the project presentation).

(b) For sufficiently high RVE resolution (say, 30 × 30 elements), we may assume isotropy and convergence
so that a single realization is sufficient. Plot the effective Young’s modulus E = E1 = E2 of the random
composite as a function of the volume fraction v1 ∈ [0, 1], and compare to the Reuss and Voigt bounds
(see slide 12 of the project presentation).

total: 100 points

General Hints and Remarks:

• You are encouraged to discuss the problems and coding tasks with each other. However, writing down
the solutions and code must be done by everyone on their own (projects with several names on them
will not be accepted; identical code among participants will not be accepted either). When handing in
your projects, you may be asked some quick questions about your code.

• For all problems, feel free to use Mathematica, Matlab, C++, etc. if it helps. If you do, please submit
a printout of your code to show your work.

• Please feel free to take full advantage of our office hours. We are happy to help!

• We are providing C++ code to run the simulations; you only need to use the code. Of course, you are
welcome to modify the code as you see fit (e.g., when running case studies with repetitions, etc.) But
all problems can, in principle, be solved without profound knowledge of C++.

• If you need help accessing the server, editing and compiling files, etc., have a look at the practical
tips found on the course website and/or contact us. Our code uses the library Eigen which defines,
e.g., Vector2D and has myriads of vector and tensor operations available (such as trace, determinant,
inverse, transpose, LU solvers, etc.) and has substantial online documentation, see e.g., here. Feel free
to take full advantage of those where needed.

• Your home directory (home/yourUsername/) has a folder CMM project1 which contains all files required
for this project. There, you will find executables (ending in .cc) for the two problems.

You might also like