Breach diameter prediction of reinforced concrete wall under contact/near-contact explosives using deep neural network

Abstract

In recent years, the use of Machine Learning (ML) has been expanded to several engineering fields with applications in structural engineering. Deep Neural Network (DNN) models have been widely implemented to predict the response of structural systems under conventional loading. The accuracy of these DNN models relies on the size of the dataset used, which can vary from hundreds to thousands of data points, typically formed by images, test data, and/or finite element models built for a specific environment. Such dependency becomes a limitation when DNN models are intended to be used in blast analysis and/or design, as data are typically scarce or restricted. This paper introduces the implementation of an ML/DNN model for the prediction of breach diameter of a reinforced concrete target subjected to contact and near-contact explosions. The DNN model predicts the breach diameter using experimentally collected concrete breach data. The DNN model considers several parameters related to the charge (e.g., type, geometry, orientation, casement properties) and wall target (e.g., material properties, geometry, reinforcement, fiber content), as well as proximity between the target and explosive. It was found that charge geometry, charge placement type, target thickness, scaled distance, and fiber content in the mix were important features affecting the breach diameter prediction.