WCCM-ECCOMAS 2026: solver-audited learned proposals

A presentation in MS220A on keeping the fracture solver authoritative.

At WCCM-ECCOMAS 2026 in Munich, I presented Solver-Audited Neural Proposals for Phase-Field Fracture in MS220A.

The presentation asked a practical question: where can learned models help without taking authority away from the mechanics? The workflow I presented lets a learned model propose a damage update, then projects and audits that proposal before either accepting it or returning to the exact fracture-solver route. The point is not to present a solver-free surrogate; it is to make the boundary of a learned contribution explicit and inspectable.

The week was also a valuable chance to learn from researchers across computational mechanics and applied machine learning, and to discuss the validation questions that matter when models are intended for engineering use. The underlying software and preprint are available through PhAST and arXiv:2606.23458.