Sensor fault detection and isolation based on nonlinear adaptive observers: A new approach for dealing with false alarms under partial faults

Abstract

This paper introduces a fault detection and isolation (FDI) system to diagnose faults in the sensors of an internal combustion engine (ICE). The FDI system relies on two nonlinear adaptive observers (NAOs) to execute analytical redundancy between the actual measurements and estimations. Two mathematical structures are proposed for developing the NAOs, which are used to construct a bank of observers. Such a bank produces estimations of the ratio between experimental and theoretical air–fuel ratios (lambda $\lambda$), temperature, and pressure, which are continuously compared with the sensors’ measurements, ensuring an uninterrupted diagnosis of the ICE sensors, even in case of total or partial fault. A threshold mechanism is incorporated into the FDI system, which is based on the variance analysis of residuals and the Euclidean norm of a sliding window vector to reduce false alarms during the partial faults diagnosis. The results of numerical tests with experimental data demonstrate the system’s ability to accurately detect multiple and simultaneous sensor faults while preventing false alarms.