A novel data-driven fault detection method inspired by parallel distributed compensation

In this paper, we propose a novel data-driven fault detection method for nonlinear system. The nonlinear system is denoted as Takagi-Sugeno fuzzy model. As the individual sub-system is presented as linear time-invariant model, we obtain every residual function in each operating point by means of fault detection used to linear systems. The construction of overall residual function is inspired by parallel distributed compensation whose initial application is to generate control strategy for Takagi-Sugeno fuzzy model. This fault detection method can be transformed into data-driven aided by implicit model approach which bridges input and output data and ultimate goals such as fault detection and so on. The specific implicit model approach in this work is based on a classical subspace identification method named past-output multivariable output-error state-space. The main merits of the implicit model approach resides in avoidance of the identification of cumbersome mechanism of systems and to some extent paving a shortcut to ultimate industrial application.