Artificial Intelligence-Aided Design for Unmanned Underwater Vehicles: A Multiple Activation Function Network-Based Hull Resistance Prediction

Abstract

Unmanned underwater vehicles (UUVs) require low-resistance hull designs to enhance their operational range and mission duration. However, the design process of UUV hulls is often multidisciplinary, sequential, and iterative, making it necessary to realize accurate and prompt resistance prediction. In order to address this challenge, this article presents a data-driven deep learning algorithm to provide a real-time prediction surrogate model for the hull resistance of UUVs. Specifically, we first collected UUV simulation data under different hull shapes with different hydrodynamic conditions. By introducing the multiple activation function network topology, we develop a deep learning algorithm that can predict resistance accurately in real-time while balancing training speed and prediction accuracy. We demonstrate that the developed deep learning algorithm can provide information about the UUV's performance by inputting hull shape and hydrodynamic condition without tedious meshing and calculation processes with an average error of less than 1.2% and a coefficient of determination of 0.9996. Finally, the UUV resistance prediction application scenarios of the constructed deep learning algorithm presented. We believe the algorithm construction and application process shown in this paper can make an invaluable contribution to artificial intelligence-aided design in underwater vehicles.