Uncertain information fusion for gearbox fault diagnosis based on BP neural network and DS evidence theory

In order to realize a stable and effective gearbox fault diagnosis system, an uncertain information fusion method based on BP neural network and modified DS evidence theory is proposed. The common gearbox fault diagnosis system consists of two dominating parts: signal feature extraction and fusion decision. Actually, the main task of most papers is to extract the characteristics of gearbox fault with various algorithms, whereas the fusion decision part is still absent. To manage the integration of gearbox fault diagnosis in the fusion decision part, this paper puts forward a novel fusion method by the introduction of two-layer fusion. In this paper, BP neural network is used to establish the local-layer fusion, while an improved DS evidence theory is taken to build decision-layer fusion. It can be seen from the definition and derivation of the two-layer fusion, the original method settles the uncertainty of multi-sensor information fusion in gearbox fault diagnosis system. According to the experiment results and analyses, the innovative algorithm improves the accuracy of fault identification and diagnosis, in addition, reduces the uncertainty of fault diagnosis system.