A Novel Fault-Prognostic Approach Based on Interacting Multiple Model Filters and Fuzzy Systems

An interacting multiple model (IMM) filter is a recognized method for adaptive estimation of states that is often necessary to characterize the behavior of dynamic systems with a multiple-mode operation. The traditional IMM filter adopts the measurement set to update the information about the active models. However, when this approach is used for fault-prognostic applications, it can lead to misleading state estimation. To overcome this problem, this paper proposes a novel fault-prognostic technique based on a modified IMM filter with fuzzy systems. The main methodological contribution of this paper is to build an approach based on an IMM filter able to make long-term fault predictions without measurements. This is done by incorporating a fuzzy system into the proposed IMM filter with the objective of modeling the system's dynamics and updating the probabilities of the observed modes. Furthermore, the proposed method and the standard IMM filter are compared in a numerical example and a real experimental platform PRONOSTIA for validation. The results analysis indicates a better prediction performance than the conventional IMM filter considering the failure time predictions and a measurement model not properly adjusted.