An efficient surrogate model for damage forecasting of composite laminates based on deep learning

In this paper, full-field damage forecasting of a laminated composite structure under different low velocity impact (LVI) conditions is realized through the proposed surrogate model, named VQ-SM. First, an efficient surrogate modelling method is proposed based on the advanced Vector Quantised-Variational AutoEncoder (VQ-VAE) proposed by DeepMind. Second, numerical simulation based on the progressive damage model of composite laminates is performed to obtain the training dataset. After training, the performance of VQ-SM is evaluated compared to the surrogate model without a representation learning process. The results show that VQ-SM has better performance with high-precise and good robustness, trained on the small dataset. Finally, the impact damage field of composite laminates is analyzed based on the surrogate model. The proposed surrogate modelling method provides not only the full-field damage forecast model for composite structures, but also an efficient method of improving the performance of the "generative" surrogate model.