A new fault diagnosis system for train BearingsBased on PCA and ACO

The objective of this paper is to propose a new system for fault diagnosis of train bearings using PCA and ACO. On the base of the analysis of time and frequency domain statistical features extracted from the vibration signals collected from the bearings, twenty features which were the most sensitive to different working states were chosen as the object of follow-on process. After zero-average and unit-variance pretreatment, PCA was carried out to compress the data dimension and eliminate the correlation among different statistical features. Then the eigenvectors consist of the front four principle components extracted by PCA were discretized to twenty domains as the input of the fault diagnosis system, in which the ACO was explored to generate classification rules. Heuristic function and probability function were suggested in this system to optimize the original ACO arithmetic. The experiments show that this new bearing fault diagnosis system can generate classification rules successfully and recognize different working states of bearings exactly.