Transfer Learning with Deep Neural Network Toward the Prediction of the Mass of the Charge in Underwater Explosion Events

In practical applications, the prediction of the explosive mass of an underwater explosion represents a crucial aspect for defining extreme scenarios and for assessing damage, implementing defensive and security strategies, and ensuring the structural integrity of marine structures. In this study, a deep neural network (DNN) was developed to predict the mass of an underwater explosive charge, by means of the transfer learning technique (TL). Both DNN and TL methods utilized data collected through coupled Eulerian–Lagrangian numerical simulations performed through the suite MSC Dytran. Different positions and masses of the charge, seabed typology, and distance between the structure and seabed have been considered within the dataset. All the features considered as input for the machine learning model are information that the crew is aware of through onboard sensors and instrumentations, making the framework extremely useful in real-world scenarios. TL involves reconfiguring and retraining a new DNN model, starting from a pre-trained network model developed in a past study by the authors, which predicted the spatial position of the explosive. This study serves as a proof of concept that using transfer learning to create a DNN model from a pre-trained network requires less computational effort compared to building and training a model from scratch, especially considering the vast amount of data typically present in real-world scenarios.