Fault diagnosis of underwater vehicle with FNN

Aiming at the character that the uncertainties of the complex system of underwater vehicle (UV) bring to model the system very difficult, a fuzzy neural network (FNN) with least adjustment is proposed to construct the motion model of UV. The adjustment of the dynamic learning rate and weights of FNN is studied. The FNN has the ability not only to approach the whole figure of a function but also to catch detail changes of the function, which makes the approaching effect preferably. Residuals are achieved by comparing the output of FNN with the sensor output. Fault detection rules are distilled from the residuals to execute thruster fault diagnosis. The feasibility of the method presented is validated by simulation experiment results.

I. Introduction

Underwater vehicle (UV) is an important tool for fulfilling many kinds of missions on the sea^[1], UV works in very tough environment and its surrounding condition changes very fast, so it is crucial to design an accurate and reliable control system to ensure its safety. It is the symbol of intelligence to realize the control system self fault diagnosis. A new way for fault diagnosis is provided by the neural network (NN) technology. Especially when refer to the complex or nonlinear system, because the method based on analysis model faces the problem that it is too difficult to construct the mathematical model, the method based on knowledge becomes an important and practical approach. And the input and output nonlinear mapping characteristics, the distribution of information and parallel processing of NN especially the abilities of high self-organization and self-learning make it a new method of fault diagnosis. The NN is applied in many real systcms ^[2]–[4].