Sensor Fault Detection and Isolation Using System Dynamics Identification Techniques.

Abstract

A sensor, generally composed of a power supply, a sensing device, a transducer, and a signal processor, behaves like any other dynamic system. A damage in any of its components can cause unexpected deviations in the sensor measurements from the actual values. Due to its increasing importance in system diagnosis and controls, a faulty sensor may lead to a process shut down or even a fatal accident in safety-critical systems.

One of the the challenge is to detect and isolate a fault in the sensor from one in the monitored system once abnormal behaviors are observed in the measurements. This work first tackles such a challenge in a single-input-single-output system by tracking the dynamic response and the associated gain factor of the sensor and the monitored system. Inspired by the fact that sensor measurements depict the dynamics of the monitored plant and the sensor, a subspace identification approach is proposed to detect, isolate, and accommodate a sensor failure under regular operation conditions without additional hardware components.

In order to deal with the increased complexity in a multiple-input-multiple-output system, an approach is then proposed to identify the underlying relations in a nonlinear dynamic system with a set of linear models, each capturing the system dynamics in the representative operating regime. Evaluated based on the minimum description length principle, the proposed approach identifies the most correlated system inputs for the target output and the associated model structure using genetic algorithm.

An approach is finally developed to detect and isolate sensor faults and air leaks in a diesel engine air path system, a highly dynamic multiple-input-multiple-output system. The proposed approach utilizes analytical redundancies among the intake air mass flow rates and the pressures in the boost and intake manifolds. Without the need for a complete model of the target system, fault detectors are constructed in this work using the growing structure multiple model system identification algorithm. Given the addition information on operation regime from the identified model, the proposed approach evaluates both the global and local properties of the generated residuals to detect and isolate the potential sensor and system faults.