A Novel Approach to Fault Detection and Isolation of Linear Systems using Modified Marginalized Likelihood Ratio Test and Uniform Priors

In this study a new methodology is proposed which is capable of accurately estimating the time of occurrence of the fault using the outcome of statistical fault detection and confirmation tests. The newly proposed approach is based on the modified marginalized likelihood ratio (MMLR) test developed using the assumption of uniform priors for fault magnitude. The isolation task in the proposed framework is achieved by means of the generalized likelihood ratio (GLR) test and fault signature matrices computed for different hypothesized faults. The efficacy of the proposed scheme has been validated on the benchmark CSTR problem using Monte Carlo simulations.