Predicting bending after impact failure mode and strength of hybrid sandwich composites: A machine learning approach

This study investigates the use of machine learning to predict post-impact flexural failure modes and residual strength in hybrid foam-core sandwich composites. Experimental data were obtained from low-velocity impact and bending after impact (BAI) tests at various impact energy levels under room and low temperatures, and with inward/outward bending configurations. To enhance model performance and interpretability, domain-informed interaction features such as (impact energy × dent depth) and (dent depth × bending direction) are incorporated to improve classification and regression models. Three models named Linear Discriminant Analysis (LDA), Ridge Regression, and Random Forest are trained and evaluated. The improved LDA model achieved 97 % test accuracy in classifying failure modes with only one misclassification. For residual strength prediction, the interaction enhanced Ridge Regression model reached R² values of 0.90. Feature importance analysis reveals that combined damage indicators (e.g., dent depth under specific loading and temperature) are key drivers of model performance. Temperature as a feature, enabled the model to correctly predict core dominated failures for low temperature specimens and facesheet dominated failures for room temperature specimens, consistent with experimental observations. This work presents a predictive framework that incorporates physically meaningful interactions into interpretable machine learning models to predict BAI damage modes and failure strength using only impact conditions, without requiring complex failure criteria.