Dimensionality reduction method based on similarity balance discriminant projection for bearing fault diagnosis

For data-driven intelligent diagnosis, comprehensive mining of the information in the data is a key issue and a great challenge. Knowledge of how to discover potentially useful information in the data is particularly critical to increase the accuracy of fault pattern recognition. The structural and discriminative information in the data present a potential direction from which to solve the issue and overcome the challenge. In this study, a similarity balance discriminant projection (SBDP) algorithm is proposed, by incorporating an optimized support vector machine (SVM) and then developing a fault diagnosis model for rolling bearing fault diagnosis. To address the defects of unbalanced adjacency relations and non-strictly monotonic between-class weight functions, SBDP reconstructs new within-class and between-class adjacency graphs by fusing structural and category information, thus effectively preserving the intrinsic manifold structure information of high-dimensional feature data. The effectiveness and suitability of the proposed model is demonstrated for two kinds of bearing diagnosis applications. The application results indicate that SBDP is better able to extract features representing the intrinsic information of faults and the optimized SVM is effective in identifying the fault types with high accuracy.

Automated summary

This study proposes a similarity balance discriminant projection (SBDP) algorithm, which incorporates an optimized support vector machine (SVM) to develop a fault diagnosis model for rolling bearing fault diagnosis. The application results demonstrate that SBDP is effective in extracting features representing the intrinsic information of faults, and the optimized SVM is successful in identifying fault types with high accuracy.