Online fault detection for temperature sensors based on orthogonal-complement-structure observer and dynamic threshold algorithm

Abstract

In this paper, a novel fault detection observer structure for aero-engine temperature sensor faults is proposed. The observer structure overcomes the difficulty that the measured input cannot be obtained and has the advantages of the reduced-order observers. The performance index in the finite frequency domain is used to solve a set of observer gain matrices offline, which greatly reduces the computational burden of online applications, ensures the availability of the algorithm on board, and makes the observer robust to noises and sensitive to faults. A dynamic threshold (DT) method based on the isolated forest (IF) algorithm and a box-plot sliding window mechanism is proposed to realise early fault detection, which updates the threshold depending on the current state and reduces false and missed alarm rates. Finally, two traditional fault detection methods are presented as comparative examples to verify the effectiveness of the proposed method.