Selective reinitialisation multiple model adaptive estimation for fault detection and diagnosis

Abstract

A novel multiple model adaptive estimation approach called Selective Reinitialisation Multiple Model Adaptive Estimation (SRMMAE) is proposed for input and output Fault Detection and Diagnosis (FDD). The existingMultipleModel Adaptive Estimation (MMAE) approach is able to detect the faults quickly. However, there are four main problems when the MMAE is used for FDD: false alarms, ambiguities, requirement of designing additional models to identify the faults and slow response to detect the removal of the faults. In this paper, the MMAE approach is improved in three ways. Firstly, the Kalman Filter (KF) or Extended Kalman Filter (EKF) is replaced by the Unscented Kalman Filter (UKF) which improves the accuracy as well as the convergence of the estimation when dealing with nonlinear systems. Secondly, a state augmentation strategy is introduced to reduce the ambiguities when applying the MMAE approach. This approach is able to estimate the faults without the need to design additional models. In terms of the false alarms and the inability to detect the removal of the faults quickly, a SRMMAE approach is proposed which consists of three different adaptive reinitialisation algorithms, which is the major contribution of this paper. The adaptive reinitialisation algorithms are designed for input and output FDD respectively and are able to detect the removal of the faults quickly without the requirement of doing maneuvers or input excitation. The SRMMAE approach for output FDD is able to estimate the output faults correctly while the estimation of the augmented MMAE diverges. The performance of the proposed algorithms to detect, isolate and identify the faults in the Inertial Measurement Unit (IMU) and Air Data Sensors (ADS) faults is compared with the MMAE and the Interacting Multiple Model (IMM) approach with an example of a Cessna Citation II CE-550 model. The simulation results suggest that the SRMMAE outperforms both the MMAE and IMM approach in effectiveness and efficiency.