A Physics‐Informed Machine Learning Model for Predicting Fatigue Delamination Growth With Fiber Bridging‐Defined Confidence Intervals in Composite Laminates

Abstract

In fatigue delamination growth (FDG), for different fiber bridging intensity and load ratios ( R ‐ratios), the Paris‐type law on FDG rate ( da/dN ) is formulated using strain energy release rate ( SERR ) on the crack tip Δ√ G tip and effective SERR Δ√ G eff as similitude parameters. The resulting da/dN (Δ√ G tip ) and da/dN (Δ√ G eff ) data collapse into a band. Based on this finding, this paper proposes a physics‐informed machine learning (PIML) framework to characterize the FDG scatter. The framework is composed of three PIML models. PIML 1, with integrated modified beam theory (MBT), predicts the SERR . PIML 2‐a and 2‐b, with the modified Paris relation and using Δ√ G tip and Δ√ G eff as similitude parameters, predict the da/dN and its 95% confidence intervals under a single or multiple R ‐ratios using the SERR s from PIML 1. Bayesian optimization is adopted to optimize hyperparameters for PIML 1 and 2‐a/b. The developed PIML framework can be utilized for variability‐enabled prediction in composite laminates.