Backtracking search optimization algorithm and its application to roller bearing fault diagnosis

Abstract

It is clearly known that support vector machine (SVM) parameters have significant effects on the accurate rate of classification result. Adjusting the SVM parameters improves its effectiveness and accuracy, which is always a challenge. On the other font, the Backtracking Search Optimization Algorithm (BSOA), an evolutionary algorithm for solving optimization problems, is proposed and proven to be effective through various benchmark problems. This paper proposes an optimization method for the SVM parameters based on BSOA. For convenience, the proposed method has been named BSOA-SVM. This method is tested with some real-world benchmark data sets to verify its robustness and effectiveness. Then, BSOA-SVM is applied for diagnosing roller bearing fault, which is a real world problem. In this diagnosing process, the original acceleration vibration signals are first decomposed into product function (PFs) by using the local mean decomposition (LMD) method. Next, initial feature matrices are extracted from PFs by singular value decomposition (SVD) techniques to give single values. Finally, these values serve as input vectors for the BSOA-SVM classifier. The results from the problem show that the combination of the BSOA-SVM classifiers obtains higher classification accuracy with a lower cost time compared to other methods.