关于轴承故障位置判别的研究

Study on Bearing Fault Location Identification

摘要: 本文基于轴承四种故障状态数据与轴承正常状态数据进行分析，从而形成对无标签轴承是否故障以及故障位置的类型划分。首先，利用Python清洗数据后，结合MATLAB将原始数据按照所选取的故障状态类型分别进行特征提取，对所提取数据进行训练集和测试集的划分，并使用一种新的基于欧氏距离和中位数的判别方法进行判别。同时，利用SPSS使用最邻近算法训练划分出的训练集，用测试集进行验证。其次，将两种算法结合聚类算法迁移到未知标签的数据，最后借鉴KNN算法的投票机制进行改进，并对数据进行具体划分，改进分类结论。

Abstract: This study is based on data from four bearing fault states and normal bearing state data, aiming to classify unlabeled bearings in terms of fault presence and fault location type. First, after data cleaning using Python, feature extraction is performed on the raw data in MATLAB according to the selected fault state types. The extracted data are divided into training and test sets, and a novel discrimination method based on Euclidean distance and median is applied for classification. Meanwhile, the k-nearest neighbors (KNN) algorithm is implemented via SPSS to train the divided training set, with the test set used for validation. Subsequently, the two algorithms are integrated with a clustering algorithm and transferred to unlabeled data. Finally, inspired by the voting mechanism of KNN, improvements are made to refine the data partitioning and enhance the classification results.