基于SLLE算法和流形聚类分析的滚珠丝杠故障诊断

Fault Diagnosis of Ball Screw Based on SLLE Algorithm and Manifold Clustering Analysis

针对滚珠丝杠故障信号非线性的特点及故障特征集中冗余信息的干扰,提出了将SLLE降维方法与流行聚类分析相结合的故障诊断方法。采集滚珠丝杠不同故障状态的振动信号和噪声信号,构造原始信号特征向量;利用SLLE算法对特征向量进行降维处理,得到筛选后的特征向量,绘制出其三维分布图;计算每种故障的聚类中心和流形距离,根据"最短距离"原则进行故障识别诊断。并通过试验及与KPCA、LLE两种诊断方法的比较,验证了SLLE降维方法与流行聚类分析相结合的故障诊断方法的有效性和识别结果的准确性。

According to the nonlinear characteristics of ball screw fault signal and the interference of the fault characteristic, the fault diagnosis method based on SLLE dimension reduction method and the popular cluster analysis is put forward. Collect vibration signal and noise signal in different fault state of ball screw and structure original signal feature vector. Using SLLE algorithm to reduce the dimension of feature vectors and feature vectors are obtained. Draw its three-dimensional distribution map. Calculate the cluster centers and manifold distances for each fault and fault diagnosis based on the principle of ＆quot;shortest distance＆quot;. And through the experiment and comparison with KPCA and LLE two diagnostic methods, the validity and accu-racy of the fault diagnosis method combined with the SLLE dimension reduction method and the popular cluster analysis method was verified.