Development of robust machine learning models for predicting flexural strengths of fiber-reinforced polymeric composites

Abstract

Fiber-reinforced composites are widely used in engineering applications due to their excellent physical and chemical properties. However, evaluating their flexural properties using conventional experimental techniques is time-consuming, costly, and limited by material and fabrication variations. This study investigates the potential of machine learning (ML) techniques to predict the flexural properties of fiber-reinforced composites accurately and efficiently. Five ML algorithms—Light gradient boosting regressor (LGBR), Extra tree regressor (ETR), Decision tree regressor (DTR), Histogram-based gradient boosting regressor (HGBR), and Adaptive boosting regressor (ABR)—were employed to predict the flexural strengths using both experimental data generated in-house and data collected from open literature. Including heterogeneous data from both sources enhances the robustness and generalizability of the developed models. The results demonstrate that the extra trees regressor (ETR) achieves excellent accuracy when applied to the heterogeneous dataset, with a coefficient of determination (R²) value of 0.94, MAE of 31.97, and RMSE of 47.64, outperforming the other three models. Furthermore, the in-house experimental data yields even higher prediction accuracy, with the best-performing model achieving an impressive R² value of 0.99, MAE of 9.53, and RMSE of 13.15. The high prediction accuracy achieved, despite the slight variability in data obtained from the literature, highlights the potential use of ML techniques to streamline the development process and reduce the reliance on extensive experimental testing. These robust models take into consideration important composite production parameters to provide design engineers and research scientists with versatile and efficient tools for the prediction of flexural properties of fiber-reinforced composites and related materials for various industries, including aerospace, defense, energy, biomedical and automotive.