Data-driven predicting the ignition of polymer-bonded explosives with heterogeneous microcracks

Abstract

ABSTRACT Ignition prediction of polymer-bonded explosives is difficult due to complex multiphysics coupling processes with heterogeneous microstructure, such as microcrack, microvoid, crystal size, and the interface property. Traditional simulation completely depends on the materials model and it is quite time-consuming. In this paper, considering heterogeneous microcracks, the data-driven ignition prediction method is proposed. A hybrid machine learning algorithm integrated with principal component analysis (PCA), binary gravitational search algorithm (BGSA) and backpropagation neural networks (BPNN) is developed. Based on the ignition database produced by finite element simulation, combining the developed prediction method, the results show better accuracy and efficiency on ignition prediction, compared with another four traditional machine learning algorithms.