Weak thruster fault detection for AUV based on Bayesian network and Hidden Markov model

Weak thruster fault detection problem is investigated for Autonomous Underwater Vehicle (AUV) with being less than 10% loss of effectiveness fault in a thruster. For the weak thruster fault, the fault features are small in general, meanwhile the fault features are also difficult to be distinguished from the external disturbances including measurement noise. In this paper, a weak thruster fault detection method is developed based on Bayesian network and hidden Markov model. In the proposed method, the fault features are enhanced at first and then the fault detection is achieved based on the occurrence probability. Specifically, Bayesian network is applicable to enhance weak signals, but the prior knowledge is difficult to obtain for an AUV. To solve this problem, this paper introduces sparse decomposition into Bayesian network to enhance the weak thruster fault features. And then, a hidden Markov model is established to provide the likelihood probability between the observation sequence and the predict one. The fault detection is achieved according to comparing the probability and the threshold. Finally, pool-experimental results on an AUV are presented to verify the effectiveness of the developed fault detection method.