Abstract: A fault detection method based on hierarchical stochastic gradient algorithm is presented for sensor fault. Compared with the leastsquare identification algorithms, stochastic gradient(SG) identification algorithm has less computational load. The fault detection method is proposed based on stochastic gradient algorithm for the system with limited computing power. The analysis reveals that the higher estimation accuracy of identification algorithm, the higher estimation accuracy of detection method. To improve the accuracy, the hierarchical identification principle is used firstly, and then the unknown true parameters of residual are replaced by the latest estimation. The calculation loads analysis of the proposed algorithm is given. In addition, a simulation example is employed to show the advantage of the proposed approach in sensor fault detection.

Key words: hierarchical stochastic gradient, sensor fault, fault detection