Assessment of Fatigue Crack Growth Rate Based on Physics-Informed Neural Networks in Hydrogen Environments

Abstract

The safety of hydrogen storage and transportation systems is a fundamental component of clean energy, so the fatigue performance of metallic materials is very important. Nevertheless, the hydrogen embrittling effect caused by hydrogen environments mostly increases the rate at which fatigue cracks grow in the materials, and therefore, the problem of predicting material lifespan drastically increases. Conventional fatigue life prediction techniques make use of experimental tests and numerical modelling, but are hindered by high data collection costs and the complexity of modelling multi-physics coupled behaviour. Over the past several years, a new solution has been developed, Physics-Informed Neural Networks (PINNs), to serve fatigue crack prediction in hydrogen environments, capitalizing on their strengths of combining physical laws with data-driven methods. PINNs have more advanced capacities in cross-scale modelling and improved the reputation of model forecasts, which outperforms the traditional approach with high precision and efficiency in safe hydrogen energy infrastructure design.