In-situ fatigue prognosis in laminated composites: A machine learning approach using self-temperature rise data

Abstract

This study presents a new methodology for fatigue prognosis in laminated composites at early stages of loading using in-situ thermal data. To this end, a Non-Dominated Sorting Genetic Algorithm II was coupled with an Artificial Neural Network model to optimize the model's hyperparameters. The model was then trained on carbon/epoxy laminate data, considering various stress concentration factors, loading levels, and stacking sequences, with a focus on self-temperature rise data as key input features. To enhance the training dataset, two different data augmentation methods were employed. Also, the performance of the developed model was compared to conventional regression-based machine learning algorithms, including Decision Tree and Gradient Boosting. Results showed that the model predicted the applied force on the composites with a mean absolute percentage error of 1.34 and achieved an R² score of 0.91 for fatigue life prediction on validation datasets.