A straightforward machine learning-based giga-cycle fatigue life regression approach for polymer matrix composites

Fatigue experiments, even during the design phase and material selection for composite structures, are time-consuming. Recently, machine learning (ML) models have shown a strong potential to predict fatigue life. However, obtaining very high-cycle fatigue (VHCF) and giga-cycle fatigue (GCF) data for composites remains a major challenge. A generalized ML model and procedure for estimating fatigue life across different regimes and loading conditions using minimal datasets are still lacking. Therefore, this work proposes a Gaussian noise-based bin augmentation approach combined with an extreme gradient-boosting regressor (XGBoost) to predict fatigue life. The performance of the model is compared to the other state-of-the-art ML regression models. The goal is to present a simple data augmentation strategy and regression procedure that can be applied to small datasets (at least 20 data points), even when the data exhibit scatter and only 50% of the dataset is available for training. This modeling procedure can support material selection during the design of composite structures in engineering applications. • Gaussian noise mimics scatter in composite fatigue data. • Bin augmentation enables training across multiple fatigue regimes. • Extreme gradient boosting gives accurate results for small datasets.