Nonlinear methods for rolling bearing fault diagnosis

On the basis of the non-stationary and non-linear characteristics of rolling bearing vibration signals, two nonlinear methods, the correlation dimension and the symbolic entropy, are respectively used to extract characteristics factors of rolling bearing vibration signals. By means of the support vector machine, pattern recognition of extracted characteristics factors was executed. From the experimental results, some conclusions were obtained that two non-linear analysis methods were feasible and the classification results of symbolic entropy were better than the results of correlation dimension. The latter showed that the corresponding sign coding of deterministic signals in any vibration signals presented a big probability, while that of random noise possessed a small probability. Thus, the influence of random noise could be decreased by symbolic entropy. The faults in rolling bearing could be classified effectively and their diagnosis could be realized by using symbolic entropy's capability of capturing the characteristics of large-scale features in signals, as well as using vector machine's capability of recognizing small samples.