Bearing Fault Diagnosis using Multi-sensor Fusion based on weighted D-S Evidence Theory

This paper has presented a novel method for bearing fault diagnosis using a multi-sensor fusion approach based on an improved weighted Dempster-Shafer (D-S) evidence theory combined with Genetic Algorithm (GA). Vibration measurements are collected from an industrial multi-stage centrifugal air compressor using three wireless acceleration sensors. Fine-to-Coarse Multiscale Permutation Entropy (F2CMPE) is applied to extract the complexity changes of vibration data sets. Then, the extracted feature vectors produced by F2CMPE via multiple scales are fed into Back Propagation Neural Network (BPNN) for fault classification. The normalized probability outputs of BPNN are considered now as inputs of the proposed weighted D-S evidence theory for multi-sensor information fusion. The measurements collected from real industrial equipment are analyzed using the proposed diagnosis method, and the experimental validation has demonstrated its efficiency to identify rolling bearing conditions, the results of which have also shown higher accuracy compared to those using individual sensor signal analysis.