AI-Assisted Statistical Analysis of Fragmentation Response of Heterogeneous Layered Media

Abstract

The fracture response of materials is highly dependent on their microstructural details, especially for fragmentation problems. The fragmentation response of a ring expanding with constant radial velocity is often analyzed using the method of characteristics. While most studies assume homogeneous material properties, we treat fracture strength as a random field to elucidate the interaction of key characters of the random field, correlation length and point-wise variation, with loading rate and other fracture and elastic parameters. The output parameters include dynamic strength and macroscopic energy dissipation capacity of the material. Due to the high dimensionality of the input random fields and other types of input data, e.g., loading rate, training machine learning models for predicting output parameters is challenging. For this purpose, we consider a few neural network architectures and feature extraction techniques to predict the response accurately with only a handful of influential input parameters.