Information fusion method for fault diagnosis based on evidential reasoning rule

This paper presents an Evidential Reasoning (ER)-based method of fault diagnosis by combining uncertain information of various fault features collected from multiple sources for fault decision-making. A normalization approach is applied to acquire diagnosis evidence from the likelihood function of fault feature samples gathered from information sources (sensors).A novel method is proposed to calculate evidence reliability according to sensor accuracy specifications and the differences of capabilities in recognizing fault modes through different fault features. A bi-objective optimization model is presented to train evidence weights to reflect the relative importance of evidence. The ER rule is then applied to combine multiple pieces of diagnosis evidence, which are regulated by their weights and reliability factors, and fault decision-making can thus be conducted on the basis of the combined results. The proposed ER-based fault diagnosis method inherits the main features of Dempster-Shafer evidence theory in uncertainty modelling, while providing a systematic process for explicitly taking into account the reliability and importance of evidence, thereby enabling rigorous inference and robust decision making. Finally, a diagnosis experiment on a rotor test bed is conducted to show the effectiveness of the proposed ER-based fault diagnosis method.