Fault detection and data restoration based on PCA for sensors of autonomous underwater vehicle

To improve the reliability and security of running and operation for autonomous underwater vehicle (AUV), it is necessary to detect the faults of the navigation sensors fixed on it, whose outputs are fed back to the motion controller, and reconstruct the inaccurate running condition information when faults happened to any sensors. According to the advantage of principal component analysis (PCA) in processing large numbers of correlative variables, it introduces an improved PCA model to realize fault detection and data restoration of sensors for AUV. In view of the performance of AUV and the disadvantage of traditional PCA, the new method named cumulative percent variance based on average eigenvalue is proposed for choosing principal component scores in order to decrease the subjectivity of the cumulative percent variance method. The means of fault detection and identification based on PCA is described in detail by doing eigenvalue decomposition with covariance of AUV data. Because there may be estimation error in traditional PCA method, a fault sensor data reconstruction method based on the new PCA model is proposed to reduce estimation error. The results of corresponding experiment by using "Beaver" underwater vehicle show the feasibility and validity of the method.