A Fault Detection Algorithm using Multiple Residual Generation Filters

Abstract

This paper proposes a fault detection algorithm based on multiple residual generation filters for discrete-time systems. Residuals are generated from estimation errors between the reference filter and multiple residual generation filters. These filters utilize only finite observation on the most recent window. The reference filter gives optimal state estimates based on all sensors. One the other hand, one of multiple residual generation filters can give the sub-optimal state estimates which can be independent of faulty sensor. Then, the fault detection rule is developed to indicate presence of fault by checking the agreement of multiple residuals. Multiple test variables for the detection rule are defined using the chi-squared distribution with one degree of freedom. Via numerical simulations for the aircraft engine system, the proposed algorithm is verified.