Subspace-based Fault Detection - Multiplicative and Additive Fault

In this research, a technique is developed to separate an additive fault from a multiplicative fault of a system which is disturbed by integrated white noise. First, Predictor-based System Identification (PBSID) is formulated in a differenced form (DPBSID) in order to filter out the integrated white noise. Second, the differenced PBSID is reformulated in a recursive form (DRPBSID) which is for separating an additive fault from a multiplicative fault. If the fault type is multiplicative, the Frobenius norm of the difference between the identified system matrix and a changed one becomes greater than a threshold. By recursively updating the system matrix, the difference goes below the threshold hence, a multiplicative fault is separated from an additive one. This technique is applied to a complex system and its effectiveness is demonstrated using Matlab simulation.