Early Detection of Incipient Faults in Nonlinear Systems Based on State-Set Observation

Abstract

This paper deals with the early detection of incipient faults in nonlinear continuous systems. The model-based diagnosis algorithm is based on a state-set observation. A state-set observer is a set-theoretic approach for systems with unknown but bounded uncertainties in the input and output measurements. The calculated predicted state-sets from the state-set observer are used to check the consistency with the current measurement sets. In this paper, we present an incipient fault detection algorithm based on an extended consistency test. The new consistency test calculates a trust factor with the measurement and predicted state-sets. A small value of the trust factor indicates a small match between the measurement sets and the predicted state-sets. The time behavior of the trust factor is used to detect incipient faults early on in a technical system. An early detection and warning of incipient faults is valuable for safer process operations and preventive protection of process equipment. Finally, the proposed incipient fault detection method is demonstrated using measured data of fault-free and faulty operation of a nonlinear water tank system.