State Estimation of Underwater Vehicle Using Pressure Sensor Array and Artificial Neural Networks

Abstract

This study investigates a method for estimating the motion states of an unmanned underwater vehicle using pressure sensor array data attached to the vehicle's surface, based on an artificial neural network (ANN) algorithm. Inspired by the lateral line system of fish, the approach aims to infer motion states by measuring the spatial pressure distribution around the vehicle. Data were generated through computational fluid dynamics (CFD) simulations under various motion conditions, and the ANN was trained to estimate speed, gliding and drift angles, and yaw rate. The results demonstrated that the ANN could accurately estimate motion states across all considered motion scenarios using 45 sensors. Even when the number of sensors was reduced to four, the network maintained reasonable performance, although the RMSE increased by approximately three times compared to using all the sensors. Principal component analysis (PCA) revealed that the pressure data formed clearly distinguishable clusters according to motion states, even in a low-dimensional space, suggesting that meaningful information can be extracted with fewer sensors.