Adaptive Detection of Sensor Uncertainties and Failures

Abstract

This paper presents our recent new results on adaptive detection of sensor uncertainty and failures in control systems. The broad problem of model-based fault detection and isolation (FDI) has been widely studied in recent years. Fundamentally, the objective of all model-based FDI approaches is to quickly and reliably diagnose subtle incipient or abrupt system degradation by monitoring residuals constructed by comparing the measured system output to a predicted output synthesized using a model of the expected healthy system behavior. Model-based diagnostic and prognostic techniques depend upon mathematical models of the plant. In the presence of uncertainties, modeling errors can decrease system sensitivity to faults, increase the rate of occurrence of false alarms, and reduce the accuracy of failure prognoses. Building on our previous work addressing actuator FDI for uncertain MIMO linear systems and working in the general direction of improved diagnostic methods for uncertain linear MIMO systems (which remains largely unaddressed), this paper presents a new design for adaptive diagnosis of sensor uncertainties and failures. The new FDI methodology explicitly includes parametric representations of additive faults (sensor bias uncertainty), multiplicative faults (sensor scaling), and considers the special case of total sensor failure (which can result in a time-varying signal that is not inuenced by the true value of the system state). Moreover, the design can support diagnosis of arbitrary combinations of sensor faults. Two new adaptive schemes are presented: one for detecting sensor uncertainties and one for detecting sensor failures. Their design and analysis details are given in this paper. For the sensor uncertainty detection scheme, simulation results are presented to verify the desired adaptive system fault detection properties.