Wide-ranging Probabilistic Physics-guided Machine Learning Approach to Break Down the Limits of Current Fatigue Predictive Tools for Metals

It appears paradoxical how today's frontier & high-impact research seeks for designing new materials to delay structural failures – especially fatigue – while the same effort is not seen concerning the way materials can be efficiently and safely used in real structural applications.

BREAKDOWN aims to transform engineering products’ design methods by identifying and including a set of (sub)micro-scale material inhomogeneities characteristics in a novel probabilistic framework.

The time has come to exploit modern experimental techniques to probe material properties at a small scale, which are scarcely involved in current fatigue characterisation schemes.

To attain this very ambitious goal, the project will rely on a breakdown of different classes of inhomogeneities to advance the fundamental mechanical understanding of their contribution to fatigue, and then reunite them within an advanced Bayesian Physics-Guided Neural Network (B-PGNN) frame.

Over the past three years, I assiduously worked to prove the feasibility of BREAKDOWN and demonstrate its superior capabilities.

However, I have merely scratched the surface of what is potentially achievable with this approach, both in terms of knowledge advancement and real engineering applications.

An extensive multimodal experimental characterisation campaign will be conducted on different material inhomogeneity states to separate and identify their individual influence on fatigue in a systematic and detailed way.

Cutting-edge numerical & analytical models will be developed and exploited as the physics knowledge in the B-PGNN scheme to effectively tackle the small datasets issue when dealing with fatigue and to ensure soundness of results. The outstanding capabilities of the framework developed in BREAKDOWN will be confirmed through specific demonstrators.

BREAKDOWN will excellently contribute towards the development of a much more sustainable design procedure with unprecedented social, economic and environmental benefits.