Interface fracture in particle composites

A ceramic matrix with ductile particles in it can be tougher than either constituent alone, and the reason is what happens at the boundaries between them. A crack meeting a particle can go around the interface or through the material, and which it does decides the toughness of the composite. That behavior lives at the scale of individual particles, and a simulation of a whole component cannot afford to resolve every one of them, least of all thousands of times over inside a design loop.

So the target is a component-scale description that remembers the interfaces. If effective properties, a Young’s modulus or a fracture toughness, could be written as functions of particle volume fraction, a component-scale model could use them and never resolve a particle. Those functions have to come from micro-scale simulations that do resolve interfaces: sparse or symbolic regression over simulation output, with the training range and the validation cases stated, rather than a curve picked because it happens to fit. That is the goal this project is working toward, and everything below is the groundwork for it.

First: which description of the interface can be trusted?

The image above is one of those runs: a phase-field calculation in which the crack is not tracked but emerges as a band of high damage, bright where the material is fully broken and dark where it is intact, deflecting as it goes rather than running straight.

Four formulations were implemented and run on the same problems, so differences could be attributed to the formulation rather than to the test case:

• a standard phase-field model, which smears the crack and finds it by minimizing total energy;
• a diffuse-interface phase-field model, which additionally represents the particle–matrix interface itself;
• a standard cohesive-zone model, with a traction–separation law throughout;
• an interface-only cohesive-zone model.

They are not interchangeable. In the cases run so far the standard phase-field model was the cheapest, but it does not acknowledge the interface: material properties change abruptly across it, so cracks tend to nucleate in the bulk rather than along the boundary where the physics puts them. The standard cohesive-zone model represents the interface explicitly, at a higher cost. The diffuse-interface phase-field model has lower computational cost than the full cohesive-zone calculation in the cases tested. It is therefore the current candidate for generating the microscale dataset, subject to further validation.

Then: calibrating it against something with an answer

A model that represents an interface needs a number for how much energy that interface costs to separate. Alison Reeves calibrated the critical energy release rate in the diffuse-interface model by simulating the peeling of lamellae, a configuration whose answer is known independently, so the value carried into the composite simulations is one the model has been made to reproduce rather than one assumed.

The double-cantilever-beam configuration served the same purpose for the cohesive-zone side, where modified beam theory gives an analytical energy release rate. In one verification case the computed value came back at 10.2 N/m against a 10 N/m reference.

It is worth being precise about what that establishes. Recovering the fracture energy you supplied confirms the implementation is doing its arithmetic correctly. It says nothing about whether that value describes a real interface; identifying that from experiment is a separate and harder problem.

From microscale simulations to effective properties

The next stage asks how effective elastic and fracture properties emerge from microstructure, using validated high-fidelity models to connect particle and interface behavior to the response observed at larger scales.

Alison Reeves carried out the model comparison and the calibration as an undergraduate researcher in the group; she has since moved on. The interface machinery she built is in use in failure and fatigue in magnetic soft materials, where the same question, what happens at a stiff particle’s boundary, is asked of a different composite.