Neural network based propulsion system fault diagnostics for the NPS AUV II

Abstract

The use of artificial neural networks to provide a method of detecting and isolating impending failures in an autonomous underwater vehivle propulsion system has been studied. Two types of fault diagnostic systems, each capable of detecting different kinds of faults, were designed. The first system addresses the fault identification proces by looking at the raw data available from system sensors. The second design processes sensor data with a Kalman filter before it is input to a neural network. The Kalman filter was designed to identify system parameters that characterize its dynamic response. These parameters serve as input to the network. This system is capable of fault detection, isolation and severity level determination.