A deep learning framework based on attention mechanism for predicting the mechanical properties and failure mode of embedded wrinkle fiber-reinforced composites

[Display omitted] •A deep learning framework is developed to predict the compressive mechanical behavior of embedded wrinkle composites.•The high-fidelity models considering the real paths of the fibers are established.•The three-dimensional FE simulations are considered to calculate the datasets and validated with experiments.•The influence laws of wrinkle severity parameters on mechanical properties and failure modes are given. To avoid the expensive computational costs process of high-fidelity simulation, a deep learning (DL) framework based on attention mechanism and three-dimensional stress state is proposed to predict the compressive mechanical properties and failure modes of embedded wrinkle thick-section composites in this paper. The deep learning framework includes strength and stiffness, stress–strain curves and failure mode prediction networks respectively using convolutional neural networks based on wrinkle angle distribution and material distribution. The attention-based loss function is considered in the failure mode network to accurately predict the local high damage areas. The high-fidelity three-dimensional finite element simulations based on progressive damage method are used to compute the datasets for training and validating. The results show that the deep learning framework can accurately predict the compressive mechanical properties and failure modes of embedded wrinkle composites. Meanwhile, the DL framework also reveals the influence rule of wrinkle parameters on mechanical properties and failure modes.