Thruster fault identification using improved peak region energy and multiple model least square support vector data description for autonomous underwater vehicle

Abstract

This article investigates a novel fault identification approach to determine the percentage of the thrust loss for autonomous underwater vehicle thrusters. The novel approach is developed from a combination of the peak region energy (PRE) and support vector data description (SVDD) by considering that PRE is able to acquire a primary feature in low dimensions from signals without any secondary process and that SVDD can establish a hypersphere boundary for a class of fault samples even in the case of a small number of training samples. Three improvements, namely removing the fusion, an energy leakage and a homomorphic transform are applied to the PRE. It forms an improved PRE to increase the area under the curve. Furthermore, another three new contents, namely the least square, a multiple model fusion and a dead zone are added to the SVDD. It constructs a multiple model least square SVDD to increase the overall identification accuracy. Experiments are performed on an experimental prototype autonomous underwater vehicle in a pool. The experimental results indicate the effectiveness of the proposed method.