Estimation and Detection of Intermittent Faults for Nonlinear Systems Disturbed by Noises With Uncertain Covariances

In this paper, the problems of intermittent fault (IF) estimation and detection are investigated for stochastic nonlinear systems disturbed by noises with uncertain covariances. The research methodology begins with the transformation of conventional nonlinear systems into descriptor systems through state and fault augmentation. For IF estimation, an innovative moving horizon estimator is introduced by utilizing the variational Bayesian technique. This approach enables the simultaneous and iterative estimation of both IFs and noise covariance matrices through the strategic selection of appropriate conjugate prior distributions. Furthermore, a rigorous analysis is conducted on the statistical properties of the estimation outcomes, leading to the development of an IF detection framework based on Hotelling’s $T^{2}$ statistic. To validate the effectiveness of the proposed methodology, extensive experimental evaluations are performed on a rotary steerable drilling tool system, demonstrating the superior performance of our algorithm in practical applications. Note to Practitioners—In recent decades, intermittent faults (IFs) have emerged as a critical safety challenge across multiple industrial sectors, with particular prominence in aerospace systems and electronic infrastructure. The precise detection of appearance and disappearance time of IFs constitutes a crucial requirement for ensuring industrial safety. Conventional approaches dependent on truncated residuals frequently exhibit significant limitations in practical implementations, primarily attributable to their insufficient capacity in dealing with nonlinear dynamic behaviors and statistically ambiguous noise profiles. To address these challenges, an adaptive moving horizon estimation framework is designed for nonlinear systems with uncertain noise covariance matrices. Leveraging the estimation results, a novel detection methodology is developed to accurately identify the appearance and disappearance of IFs. Experimental validation through comparative case studies demonstrates the operational viability and detection efficacy of the proposed methodology, providing industry practitioners with a reliable tool for enhancing operational reliability in IF-affected systems.