基于多重分形KPCA和LS-SVM列车车轮早期损伤的检测研究

针对车轮早期损伤难以检测的问题,利用JD-1轮轨模拟试验机进行了车轮早期损伤模拟振动试验,提出基于多重分形核主成分(KPCA)和最小二乘支持向量机(LS-SVM)的车轮早期损伤诊断方法。首先对信号进行共振稀疏分解预处理降噪,根据峭度和互相关系数最大原则,筛选子带分量重构信号,然后获得其多重分形参数,利用核主成分(KPCA)降维提取敏感特征向量矩阵,将该向量矩阵输入最小二乘支持向量机(LS-SVM)进行分类识别。研究结果表明,该方法可以有效地实现车轮早期损伤的识别,与未通过降噪信号的比较,其具有更高的诊断准确率,为深入研究损伤车轮振动特性提供理论依据。

Detection Research on Train Early Damaged Wheel Based on the Multifractal KPCA and LS-SVM

Aiming at the difficulty of detection of the early damaged wheel, the vibration experiments of the early damaged wheels were carried out using a JD-1 wheel/rail simulation facility. The early damaged wheel diagnosis method based on multifractal kernel principal component analysis(KPCA) and least squares support vector machine(LS-SVM) was proposed. Firstly, the signal was decreased noise by using resonance-based sparse signal decomposition, sub-band components were selected to reconstruct the signal according to the principle of maximum kurtosis and information entropy. Then, multi-fractal parameters of the reconstructed signal were obtained, and the sensitive eigenvector matrix was extracted by utilizing the reduction dimension of kernel principal component(KPCA), which is input into the least squares support vector machine(LS-SVM) for classification and identification. The results indicated that the method can be effectively achieved the identification of the early damaged wheel and has higher diagnostic accuracy in comparing with no noise reduction signal. It provide theoretical basis for further study of vibration characteristics of the damaged wheel.