Takagi-Sugeno fuzzy inference based cascaded hybrid modeling and fault diagnosis

A cascaded hybrid modeling strategy is adopted based on the combination of a prior model and a nonparametric model. The prior model is built according to process mechanism; while the non-parametric model is obtained from process data. Takagi-Sugeno fuzzy inference is used for estimation of the time-varying parameters in the non-parametric model. Based on the developed cascaded hybrid model, a “double granularities” fault diagnosis method is proposed. At the coarse granularity level, the parameters of the prior model can be used to isolate the location of the occurred fault. At the fine granularity level, more precise fault information can be obtained based on process data and the parameters estimated by T-S fuzzy inference. The experimental results on a three-tank system show the effectiveness and feasibility of the proposed method.