Aimed at defects of the traditional BP neural network algorithm which had a slow convergencespeed and was easy to get in a local minimum in application tofault diagnosis algorithm of dual-redundancy system, this paper analyzed common faults of brushless DC motor(BLDCM), and mainly studied the character of five types of faults, and proposed a new approach for the fault diagnosis. According to the method of statistics, targeted data was picked up to be the symptom of faults. The front-network consisted of three layers of wavelet neural network. Aimed at defects of the traditional BP algorithm which was less of evidence in configuration of the parameter and topology of neural network, a genetic algorithm was adopted as the sample training algorithm. The chromosome coding was adopted to optimize the parameter of wavelet function and the parameter of neural network′. By means of simulation tests and applications to a microminiature autonomous underwater vehide(AUV), it is confirmed that the wavelet neural network method based on genetic algorithms has anexcellent fault-recognition ability.