Prediction of low-velocity impact responses for bio-inspired helicoidal laminates based on machine learning

The inherent capacity of natural protective systems to withstand impact loadings, attributed to their microscale helicoidal architectures, has garnered significant interest. Drawing inspiration from this mechanically robust design, this study aims to introduce the composite laminates with a helicoidal distribution and to accurately and efficiently predict their Low-Velocity Impact (LVI) responses. Initially, the Latin hypercube design (LHS) was employed to generate 500 samples representing various pitch angles. An experimentally verified finite element model was then established to capture the load-displacement curves and energy-time curves for these 500 samples. Subsequently, the Convolutional Neural Network (CNN) model was utilized to accurately predict the load-displacement curves and energy-time curves for bio-inspired helicoidal laminates across different pitch angles. The principal component analysis (PCA) was used to enhance the efficiency of learning the load-displacement and energy-time curves in a reduced dimensional space, and the SHapley Additive exPlanations (SHAP) method was employed to investigate the feature importance of pitch angle. Finally, the helicoidal laminate with the highest energy absorption for a given volume was obtained by the genetic algorithm (GA) combined with the CNN model. This optimized laminate demonstrates a remarkable 9.5 % improvement in energy absorption compared to the best-performing sample within the original data set. Furthermore, the "spiraling" delamination damage of the helicoidal laminates was studied, which indicates that the delamination with small pitch angle is more pronounced for that with large pitch angle. The proposed method offers significant advantages in terms of cost reduction and efficiency enhancement for predicting the LVI responses of helicoidal laminates, holding immense potential in structural design and optimization of composite materials.