Abstract: To solve the problem of rotating machinery fault diagnosis, a method based on correlation vector machine（RVM） is proposed, which is a Bayesian sparse kernel method for regression and classification. The prominent advantage of RVM is the sparseness of the model and the probability of prediction. In order to improve the robustness of RVM and reduce the influence of outliers on the predicted value, and solve the problem that the redundant feature can not be removed when Lasso method is used for feature selection, the RFE recursive feature elimination method using Lasso as the bottom algorithm is proposed to remove the irrelevant and redundant features of the sample data-set. Finally, the data collected in industrial environment is used as a sample set for testing and compared with the traditional algorithm, the results show that the method can improve the detection efficiency and maintain high detection rate.

Key words: rotating machinery, correlation vector machine, fault diagnosis, feature elimination

摘要： 针对旋转机械故障诊断问题，提出一种基于相关向量机（RVM）的故障检测方法，RVM是一种用于回归和分类问题的贝叶斯稀疏核方法，其突出的优势是模型的稀疏性和预测的概率性。为进一步提高RVM模型的鲁棒性，减小样本数据中异常值对预测值的影响，针对Lasso方法进行特征选择时无法去除冗余特征的问题，提出以Lasso为底层算法的RFE递归特征消除方法去除样本数据集中无关特征和冗余特征。最后以工业环境下采集的数据作为样本集进行实验，同传统算法进行了比较，结果表明该方法在保持较高检测率的同时，提高了故障预测的时效性和稳定性。

关键词: 旋转机械, 相关向量机, 故障诊断, 特征消除