Moving Horizon Fault Estimation for Nonlinear Stochastic Systems With Unknown Noise Covariance Matrices
Li Sheng, Shiyang Liu, Ming Gao, Wuxiang Huai, Donghua Zhou
Abstract
In this article, the problem of fault estimation is studied for nonlinear stochastic systems with sensor faults. The system under investigation involves stochastic noise with unknown but time-varying covariance matrices, which bring in substantial difficulties in the fault estimator design. First, a nonlinear singular system is constructed by introducing an augmented vector containing system states and sensor faults, and there is no prior assumption on faults during this process. Subsequently, by combining the expectation maximum technique and the moving horizon estimation (MHE) method, an improved fault estimation algorithm is proposed for the nonlinear singular system. The augmented state and noise covariance matrices can be inferred by calculating the approximate posterior probability density function (pdf) iteratively without the prior distribution information. Finally, the effectiveness of the proposed fault estimation algorithm is verified by a numerical simulation and an experiment concerning the rotary steerable drilling tool system.