A Physics-Informed Machine Learning Model for Mode I Fatigue Delamination with Fiber Bridging Retardation of Composite Laminates

Fatigue delamination growth (FDG) is one of the most important causes of failure in composite structures. The presence of fiber bridging can have significant retardation effects on FDG behavior. Models based on machine learning (ML) have been developed to determine fatigue properties (i.e. fatigue life and crack propagation) of metals and composites in the last decade [1-2]. There is sufficient evidence that the use of ML can have great capacity in materials’ fatigue life prediction and crack evaluation. However, the use of ML without physical constraint takes disadvantages of black-box and lack of physical interpretation, which can result in fatigue determination violating physics. Physics-informed neural networks (PINNs) incorporates physical principles recently therefore have been proposed and employed to improve ML model interpretability and prediction accuracy [3]. The objective of this research is to propose a physics-informed ML strategy for developing a reliable model that accurately represents FDG with fiber bridging retardation. The study involved the following steps: (1) conducting fatigue experiments under mode I delamination loading with varying fiber bridging, according to the test procedure initially developed by the first author [4]; (2) analyzing the fatigue experimental data via the Paris law, and performing regression analysis on the curve-fitting parameters related to crack propagation, contributing to a semi-empirical Paris-type relation for fiber-bridged fatigue delamination; (3) constructing a PINNs-based ML algorithm with dynamic weight parameters and a dynamic learning rate; (4) conducting a comparison on fiber-bridged FDG characterization between the PINNs-based ML model and non-physics-informed ML model, as well as the Paris-type relation. The results provide sufficient evidence that the use of PINNs-based ML algorithm can have more accurate predictions, as compared to non-physics-informed ML model as well as the theoretical model. Furthermore, this research demonstrates the importance of incorporating physical information (constraint) in ML to have an accurate model to represent fatigue delamination of composite laminates.