An evidential reasoning-based fault detection method for multi-mode system

Abstract

A multi-mode system (MMS) is often switched among different modes which can experience great state changes. Compared with the single-mode system, fault detection for MMS is more challenging due to the diversity of states. The existing fault detection methods for MMS are mainly the data-driven based which cannot effectively combine expert knowledge with observation data. In this paper, a new multi-mode identification framework is constructed based on the evidential reasoning approach with interval reference values (ER-IRV), which can integrate thresholds of different modes. In this framework, a two-stage fault detection method is proposed. In the off-line stage, a multi-threshold optimization algorithm is developed by aggregating historical data recursively using the ER-IRV. Based on the optimized thresholds, transition processes are modeled using the membership degree of standard transitional data to steady modes. In the on-line stage, steady modes are identified by aggregating multiple attributes, and transition processes can be identified by analyzing the membership degree variation characteristics. Finally, fault detection can be realized by univariate threshold detection method and transitional slope detection method respectively. The implementation procedure of the methodology is given by a numerical example. The practical application effect is verified by a fault detection experiment on satellite turntable system.