A machine learning model for predicting the ballistic impact resistance of unidirectional fiber-reinforced composite plate

It has been a vital issue to ensure both the accuracy and efficiency of computational models for analyzing the ballistic impact response of fiber-reinforced composite plates (FRCP). In this paper, a machine learning (ML) model is established in an effort to bridge the ballistic impact protective performance and the characteristics of microstructure for unidirectional FRCP (UD-FRCP), where the microstructure of the UD-FRCP is characterized by the two-point correlation function. The results showed that the ML model, after trained by 175 cases, could reasonably predict the ballistic impact energy absorption of the UD-FRCP with a maximum error of 13%, indicating that the model can ensure both computational accuracy and efficiency. Besides, the model's critical parameter sensitivities are investigated, and three typical ML algorithms are analyzed, showing that the gradient boosting regression algorithm has the highest accuracy among these algorithms for the ballistic impact problem of UD-FRCP. The study proposes an effective solution for the traditional difficulty of the ballistic impact simulation of composites with both high efficiency and accuracy.