Machine Learning Algorithms for Identification of Cohesive Zone Parameters for Mixed-Mode Fracture in Composite Sandwich Structure

Abstract

We investigate three machine-learning algorithms to identify the cohesive zone parameters used in the fracture analysis of a honeycomb/carbon-epoxy sandwich structure.  Determining the cohesive parameters from experimental results can be complicated because multiple parameters can fit the experimental results equally well, resulting in non-unique solutions that do not capture the correct values.  Numerically, determining the cohesive zone parameters can be time-consuming due to modeling challenges, such as fine meshes needed near the crack tip.  In this study, we apply three machine learning algorithms, namely, Support Vector Regression (SVR), Random Forest (RF), and Artificial Neural Network (ANN), to identify the interfacial cohesive parameters between the honeycomb core and carbon-epoxy facesheets. Our study represents the interfacial fracture using a cohesive zone model in a finite element (FE) simulation of the Asymmetric Double Cantilever Beam (ADCB) specimen configuration.  The input variables include four cohesive zone parameters: the maximum normal contact stress, critical fracture energy for normal separation, maximum equivalent tangential contact stress, and critical fracture energy for tangential slip. FE simulations with ranges of cohesive model parameters produce a database of load-displacement responses that serve as training data for the machine learning algorithms.  The results show that the suggested methods demonstrate remarkable accuracy for situations where the interfacial characteristics lie within the training dataset or beyond and reveal the importance of specific parameters. Additionally, we perform hyperparameter tuning and optimization techniques to enhance the model's performance.