Decision procedures for fault detection and isolation derived from knowledge and data

Abstract

This work presents a unified approach to derive decision procedures for model based fault detection and isolation (FDI) either from knowledge or from experiments. In the knowledge-based approach, fuzzy rule weights are defined directly from model structure. In the supervised learning approach, the decision procedure is derived from a data set. The symbolic to numeric integration provided by fuzzy sets in the proposed framework allows integrating symbolic symptoms into the decision procedure. The proposed method is applied to the FDI of a winding machine.