Highly efficient simulation of composites by determining failure initiation and fracture angle with artificial neural networks

The calculation efficiency of Puck and LaRC05 failure criterion is a significant limitation in the simulation of composite structures, due to the great iteration calculation. In this paper, a highly efficient method to determine the failure initiation and fracture angle of composites based on artificial neural networks (ANN) is proposed. Two ANN models for failure initiation and fracture angle are modeled and trained by data set generated by nondimensionalization formula through Monte-Carlo method, and proved valid by predicting of two typical composites. A highly efficient method is proposed based on ANN models and golden section search method, and realized in ABAQUS by subroutine. In the simulation of composites under shear load, the crack path predicted by proposed method is precise and the calculation time is only 4% of original method. The proposed method is accurate, highly efficient and easy to implement in the simulation of composites.