Physics-informed machine learning for loading history dependent fatigue delamination of composite laminates

Fiber bridging has retardation effect on Mode I fatigue delamination, making the damage loading history dependent. This research creates a physics-informed machine learning (ML) model for characterizing this fatigue delamination propagation. After a training, the model can predict fatigue crack growth rate for a given crack length, accounting for a certain amount of bridging fibers. Mode I fatigue experiments were first performed to obtain sufficient data for the ML. A semi-empirical Paris-type correlation determines fatigue damage evolution with bridging retardation. This correlation was integrated as a physical constraint into the physics-informed neural networks (PINNs). PINNs demonstrate excellent performance: the predictions of the delamination fall within a narrow scatter band of 1.5 times by crack growth rate, outperforming both the non-physics-informed ML model and the Paris-type correlation. The proposed ML model can be applied for the development, characterization and comparison of composite materials, and for composite structure design and life evaluation.