基于PCA-KNN聚类的通用在线故障诊断算法设计

Design of General On-Line Fault Diagnosis Algorithm Based on PCA-KNN Clustering

摘要:

为了克服以往故障诊断算法所具有的难以诊断效率低、诊断精度不高和模型通用性不强的缺点,提出了一种基于PCA主元分析特征优化和KNN聚类的故障诊断法算法；首先,给出了故障诊断的总体模型和诊断原理,然后在故障征兆原始样本数据的基础上,通过PCA主元分析法进行特征优化,获得维数约简的样本数据,从而提高故障诊断的效率；在此基础上,采用训练样本数据对模糊K均值分类器进行训练,并计算每个聚类的距离和阈值；最后,将在线获取的测试样本数据或离线样本数据输入到模糊K均值分类器,获得其所属分类,并采用KNN最近邻算法来获取其K个近邻,根据其与近邻的距离平方和与所属聚类距离平方阈值来判断其是否为故障样本,从而实现故障诊断；以滚动轴承故障诊断试验和模拟电路故障诊断试验为例,实验结果证明了文中方法较其它方法具有诊断效率高和诊断精度高的优点,是一种通用的和可行的在线故障诊断方法。

Abstract:

For conquering the problems of traditional fault diagnosis algorithm such as low diagnosis efficiency, low accuracy and the generalization of model is not strong enough, a fault diagnosis algorithm based on Principal Component Analysis feature PCA optimizing, fuzzy k-means and K-Nearest Neighbor clustering is proposed. Firstly, the main model is introduced and diagnosis principle is designed. Then the PCA method is used to realize the feature optimizing and the data after dimension reducing is obtained, so as to improve the diagnosis efficiency. The training sample data is used to train the fuzzy k-means classifier and get the cluster distance and threshold is computed.to get which the classifier it owns to, then the KNN algorithm is used to obtain its K nearest neighbors to compute the sum of distance square between it and them, by comparing the relationship of value of sum of distance square with distance sum threshold it can conclude whether the sample is fault sample or normal sample. The fault diagnosis experiment of antifriction bearing experiment is operated, and the result shows it has the high convergence rapid, diagnosis efficiency and diagnosis accuracy, so the method in this paper can be accepted as a universal fault diagnosis method and has large application foresight.