Residual tensile strength in composite laminates: a deep learning approach

To effectively predict residual tensile strength (RTS) of carbon fiber-reinforced plastics (CFRP) composite laminates after impact, an integrated framework is proposed. The framework incorporates a three-dimensional (3D) nonlinear progressive damage model and a backpropagation deep neural network (DNN) model with three hidden layers. The 3D model is developed to predict RTS and prepare dataset for the training of the DNN model. The model is validated by tensile tests on laminates that were damaged by impacts of various energies levels. The failure modes and the fracture morphology of the laminates are studied by simulation and scanning electron microscopy (SEM) results. Statistical analysis on the performance of the DNN demonstrates that a trained and constructed neural network can satisfactorily predict RTS of laminates pre-damaged by impacts.