Bearing fault diagnosis based on multi-scale possibilistic clustering algorithm

In the on-line monitoring for the fault of rolling bearing, we have no information about the cluster number of the obtained data signal, which cause great challenges for on-line fault diagnosis when using clustering algorithms. In this paper, we extract three features of the vibration signals of rolling bearings as the parameters in time-domain, and then multi-scale possibilistic clustering (MPCM) algorithm is applied to unsupervised fault diagnosis. Because of the number of clusters is controlled by multi-scale factors automatically, the proposed method for fault detection can meet the actual demand of the machinery industry environment. The experimental results for bearing signals show that the method not only can automatically determine the number of clusters, but also has a better clustering performance than conventional possibilistic clustering algorithms.