Active Fault Diagnosis for Stochastic Systems Under Accuracy-Limited Sensors

This article aims to deal with the active fault diagnosis (AFD) problem for a class of stochastic systems with accuracy-limited sensors. The goal of the input design is to optimally discriminate different system modes. As there is no closed-form expression for the likelihood of misdiagnosis, the Young divergence family is defined based on the Young inequality to establish a new input design index for AFD. The maximization of the established design index results in the minimization of a reachable upper bound on the likelihood of misdiagnosis. This design index makes the input design problem solvable to global optimums. The probability density function (PDF) of the readout measurement output from accuracy-limited sensors is deduced and the functional relationship between the corresponding actual misdiagnosis probability and the sensor accuracy is derived. By virtue of the minimum actual misdiagnosis probability decision-making rule, we draw the AFD conclusion on the basis of the real-time readout measurement output from accuracy-limited sensors. Experimental results illustrate that the actual misdiagnosis probability increases with the decrease of the sensor accuracy.