An improved decentralized model for sensor fault detection and isolation demonstrated on an airplane system

Abstract

With various modeling technologies applied, the sensor fault detection and isolation scheme based on the decentralized model (also referred to as dedicated observer scheme) becomes a popular approach for sophisticated systems. However, the commonly used modeling approach in many literatures that directly takes measurement values as model inputs may result in residual crosstalks and even false alarms. In this paper, the traditional decentralized model scheme is analyzed and a novel scheme based on the time window interactive prediction structure is proposed. Then, the Elman neural network is applied to model identification due to its nonlinear approximation and online learning properties. Finally, Simulations for comparison using the decoupled longitudinal motion model of some airplane are performed, and the results show that the proposed scheme has higher detection speed, lower false alarm rate and less undetected faults.