数据不平衡情况下的柴油机故障诊断方法  

Diesel Engine Fault Diagnosis Method for Imbalanced Data

作　　者：毕凤荣[1] 郭明智 毕晓阳 汤代杰 沈鹏飞 黄盟 Bi Fengrong;Guo Mingzhi;Bi Xiaoyang;Tang Daijie;Shen Pengfei;Huang Meng(State Key Laboratory of Engines(Tianjin University),Tianjin 300072,China;School of Mechanical Engineering,Hebei University of Technology,Tianjin 300401,China)

机构地区：[1]先进内燃动力全国重点实验室(天津大学),天津300072 [2]河北工业大学机械工程学院,天津300401

出　　处：《天津大学学报（自然科学与工程技术版）》2024年第8期810-820,共11页Journal of Tianjin University：Science and Technology

基　　金：河北省高等学校科学技术研究项目(QN2022159).

摘　　要：由于强调整体分类的准确率,机器学习方法在数据不平衡情况下的柴油机故障诊断效果不佳.因此,本文提出一种改进合成少数过采样技术(SMOTE)与机器学习技术相结合的故障诊断方法.首先对SMOTE算法进行改进,采用k近邻算法滤除多数类中的噪声样本,从而减少各种故障类别之间的重叠.同时,使用k-means算法确定少数类稀疏度和采样权重,减轻类内不平衡.然后,使用改进SMOTE算法平衡柴油机故障数据,并利用机器学习方法进行最终故障诊断.在二维数据集上的实验表明,改进SMOTE算法能有效减轻原始数据中存在的类重叠和类内不平衡问题.柴油机故障诊断实验表明,改进SMOTE算法生成的故障样本能更好地模拟原始故障样本,使用改进SMOTE算法能提高故障诊断方法的准确率.Due to the emphasis on the accuracy of overall classification,the machine learning method is ineffective in diesel engine fault diagnosis for imbalanced data.Therefore,in this study,a fault diagnosis method combining an improved synthetic minority oversampling technology(SMOTE)algorithm and machine learning technology was proposed.This method firstly improved the SMOTE algorithm.It employed the k-nearest neighbor algorithm to filter out noise samples from the majority class,thereby reducing overlap between various fault classes.Meanwhile,the kmeans algorithm was used to determine the minority class sparsity and sampling weight,which reduced intraclass imbalance.Then,the improved SMOTE algorithm was used to balance the diesel engine fault data,and machine learning methods were used for the final fault diagnosis.Experimental results on a two-dimensional dataset indicate that the improved SMOTE algorithm can effectively alleviate the class overlap and intraclass imbalance problems in the original data.Diesel engine fault diagnosis experiments show that the fault samples generated by the improved SMOTE algorithm can optimally simulate the original fault samples,and the improved SMOTE algorithm can improve the accuracy of fault diagnosis methods.

关 键 词：数据不平衡 故障诊断 合成少数过采样技术 柴油机 振动信号 

分 类 号：TK448[动力工程及工程热物理—动力机械及工程] TP181[自动化与计算机技术—控制理论与控制工程]