Selective-reinitialization multiple-model adaptive estimation for fault detection and diagnosis

Abstract

Copyright © 2014 by Peng Lu. The existing multiple-model adaptive estimation approachis able to detect faults quickly. However, there are three main problems when it is used for fault detection and diagnosis: false alarms, requirement of designing additional models to identify the faults, and slow response to detect the removal of the faults. In this paper, a novel selective-reinitialization multiple-model adaptive estimation approach is proposed. This approach introduces a state augmentation strategy that can identify the faults without designing additional models, as well as reduce false alarms. The major contribution of this approach is that three selective-reinitialization algorithms are proposed that can improve the performance of the multiple-model adaptive estimation significantly. The selective-reinitialization multiple-model adaptive estimation approach eliminates false alarms and is quick to detect the removal of the faults. The performance of the proposed approach is compared with the multiple-model adaptive estimation and the interacting multiple model withan example of the fault diagnosis of the inertial measurement unit and air data sensors for a Cessna Citation II aircraft. The simulation results suggest that the selective-reinitialization multiple-model adaptive estimation outperforms the multiple-model adaptive estimation and interacting multiple model in effectiveness and efficiency.