Fault Diagnosis Method for the Train Axle Box Bearing Based on KPCA and GA-SVM

Abstract:

Online fault diagnosis for the train axle box bearings is a wide and important study topic since it plays a critical role in train safety. Due to the vibration signals nonlinear and non-stationary characteristics, accuracies of the methods such as neural network and hierarchical clustering are less than 90% which are not satisfying. In this paper, kernel principal component analysis (KPCA), a nonlinear process technique, was to tackle each signals 18 feature parameters for extracting the main features to reflect the signal characteristics. Then, in fault pattern recognition, support vector machine (SVM) based on genetic algorithm (GA) was used to identify the current fault type of the bearings, including normal, outer ring fault, inner ring fault and rolling element fault. The results show that the prediction accuracy of GA-SVM method reaches to 96.33%, which is quite effective.

• Periodical: Applied Mechanics and Materials (Volume 441)
• Pages: 376-379
• DOI: https://doi.org/10.4028/www.scientific.net/AMM.441.376
• Online since: December 2013
• Authors: Ci Wang, Li Min Jia, Xiao Feng Li
• Keywords: Axle Box Bearings, Fault Diagnosis, Feature Parameters, GA, KPCA, Support Vector Machine (SVM)

References

[1] Y. Liang, L. M. Jia, G. Q. Cai, et al: China Railway Science Vol. 34 (2013), pp.95-100.

[2] Yoon. Seongkyu and J. F. MacGregor: AIChE J Vol. 46 (2000), pp.1813-1824.

[3] B. Schölkopf, S. Mika, A. Smola, et al: ICANN 98 Vol. 1 (1998), pp.147-152.

[4] V. N. Vapnik: The Nature of Statistical Learning Theory (Springer Verlag, New York 1995).

[5] B. Schölkopf, C. Burges and V. Vapnik: KDD-95 Proceedings 1995, pp.252-257.

[6] E. Rahimpour, V. Rashtchi and M. Pesaran: Electr. Eng. Vol. 89 (2007), pp.37-42.

[7] X. Zhang, W. W. Yan, Z. Y. Liu, et al: Proc. CSEE Vol. 27 (2007), pp.56-61.

[8] K. S. Tang, K. F. Man, S. Kwong, et al: IEEE Signal Process. Mag. Vol. 13 (1996), pp.22-37.