Bearing condition monitoring and fault diagnosis based on mathematical morphology and fractal theory and PNN

Abstract:

Aiming at the problems about poor real-time monitoring of rolling bearing condition and low accuracy of fault diagnosis, a feature extraction algorithm based on improved local mean decomposition (ILMD) and mathematical morphology fractal theory is proposed, and making it combine with probabilistic neural network (PNN) to complete the intelligent recognition and classification of bearing status. Firstly, the algorithm decomposes the original signal of the bearing through ILMD, selects the two-order component with the largest correlation coefficient and finds its fractal dimension as the feature vector. Secondly, the “morphological coverage area” is used as the third-dimensional feature vector. At the same time, a three-dimensional feature matrix is constructed. Finally, the feature matrix is input into PNN to complete the state recognition and classification. Using the actual bearing data of CWRU，the experiment results show that the proposed algorithm can not only accurately identify bearings in different states, but also effectively classify bearing states with different damage levels under the same failure. The average recognition rate exceeds 99.6%, and the average recognition time is 0.21 s.