基于MSSMOTE-CNN模型的空调冷水机组故障诊断  

Fault Diagnosis of Water Chiller Based on MSSMOTE-CNN Model

作　　者：曹冉冉 田禾[1,2] 樊怀聪 冯明文 CAO Ranran;TIAN He;FAN Huaicong;FENG Mingwen(National Demonstration Center for Experimental Mechanical and Electrical Engineering Education,Tianjin University of Technology,Tianjin 300384,China;Tianjin Key Laboratory for Advanced Mechatronic System Design and Intelligent Control,Tianjin University of Technology,Tianjin 300384,China)

机构地区：[1]天津理工大学机电工程国家级实验教学示范中心,天津300384 [2]天津理工大学天津市先进机电系统设计与智能控制重点实验室,天津300384

出　　处：《机械制造与自动化》2024年第6期128-132,137,共6页Machine Building & Automation

基　　金：国网天津市电力公司科技项目(KJ22-2-02);国网天津市电力公司科技项目(KJ21-1-21);天津理工大学2022年天津市研究生科研创新项目(2022SKYZ070);天津理工大学2022年校级研究生科研创新实践项目(YJ2209);企业科技特派员项目(20YDTPJC01670)。

摘　　要：针对冷水机组运行过程中数据类别不平衡问题,提出基于马氏距离进行“三角”区域插值的MSSMOTE方法对故障数据进行扩充,将得到的数据输入CNN模型进行训练,实现对冷水机组中7种故障的诊断。在不同扩充比例下和同一种数据类型下分别进行仿真测试,结果显示:在扩充比例为4时,MSSMOTE-CNN模型对于正常样本测试的准确率和F 1-score分别达到0.961和0.971,能够较准确识别出冷水机组的故障类型。To deal with the unbalanced data types during the operation of water chillers,this paper proposes the MSSMOTE method based on Mahalanobis distance and"triangle"area interpolation to expand the fault data,and input the obtained data into the CNN model for training,so as to realize the diagnosis of seven kinds of faults in water chillers.Simulation tests were conducted under different expansion ratios and the same data type.The results showed that when the expansion ratio was 4,the MSSMOTE-CNN model achieved an accuracy of 0.961 and a F 1-score of 0.971 respectively for normal sample testing,which was capable of accurately identifying the fault type of the chiller.

关 键 词：MSSMOTE-CNN模型 数据不平衡 故障诊断 冷水机组 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]