Actuator fault diagnosis in autonomous underwater vehicle based on Neural Network

Abstract Actuator is critical component of the AUV, the actuator's fault diagnosis is of great significance to ensure the safe operation of the AUV. Therefore, it is of great important to investigate an intelligent method of AUV. In this research, A diagnostic network combining wide convolutional neural network (WDCNN) and extreme learning machine (ELM) is proposed. Firstly, the reliability of the diagnostic network when sailing downstream and upstream is considered, and a network with strong generalization ability is found structure. Secondly, the model does not rely on any domain adaptation algorithm or require information of the target domain. It is verified that WDCNN can effectively extract the deep features of data with high diagnostic accuracy and strong anti noise ability, and ELM can effectively increase the generalization ability of the model. The network can be applied to AUV actuator fault diagnosis.