关键词: 煤矿机械轴承, 故障诊断, 特征选择, 嵌入学习, 稀疏回归, K-最近邻

Abstract: Aiming at the problems that the traditional feature selection algorithm used for fault diagnosis of coal mine machinery bearing separates embedding learning and feature sorting, it cannot accurately select subsets that can represent high-dimensional data sets and the accuracy of fault diagnosis is not high. In this paper, a fault diagnosis method for coal mine machinery bearing based on joint embedding learning and sparse regression is proposed. The method first constructs an embedded learning model to learn the manifold structure of high-dimensional data. And then the ι_(2,1)-norm with group sparsity is introduced into the regression model to effectively eliminate the influence of redundant features; Furthermore, a feature selection framework is constructed jointly with embedding learning and sparse regression, and the essential features that can accurately characterize the original high-dimensional data are selected; Finally, the fault diagnosis of coal mine mechanical bearing is carried out by combining with K-nearest neighbor algorithm. The experimental results show that the proposed model significantly improves the fault diagnosis accuracy of coal mine mechanical bearings.