基于PCA-Clustering的压缩机回液故障诊断  

Fault Diagnosis for Compressor Liquid Floodback based on PCA-Clustering

作　　者：周镇新 李绍斌 谭泽汉 陈焕新[1] 王江宇[1] 刘江岩[1] 郭亚宾[1] 孙劭波 Zhou Zhenxin;Li Shaobin;Tan Zehan;Chen Huanxin;Wang Jiangyu;Liu Jiangyan;Guo Yabin;Sun Shaobo(School of Energy and Power Engineering,Huazhong University of Science and Technology,Wuhan,430074,China;Gree Electric Appliances,Inc.of Zhuhai,Zhuhai,517907,China)

机构地区：[1]华中科技大学能源与动力工程学院,武汉430074 [2]珠海格力电器股份有限公司,珠海5179070

出　　处：《制冷学报》2018年第4期111-118,共8页Journal of Refrigeration

基　　金：国家自然科学基金(51576074;51328602)资助项目;空调设备及系统运行节能国家重点实验室开放基金项目(SKLACKF201606)资助~~

摘　　要：在多联机(VRF)空调系统中,压缩机回液将导致能量损失。本文结合大数据提出了一种基于PCA-Clustering的压缩机回液故障诊断的方法。首先提取出故障相关变量,并通过数据预处理,剔除异常值与空值;然后将处理后的数据进行主成分分析(PCA),获取降维后的新主元变量数据;最后将新的主元变量进行聚类分析(Clustering analysis)得到回液故障数据分类标签。结果表明:该方法能够在数据标签未知的情况下,较好的区分不同类别的压缩机回液故障及正常数据,使压缩机回液故障诊断率达到94.29%。The liquid floodback in a compressor has an adverse impact on the variable refrigerant volume（VRF） air-conditioning system,which will cause energy loss. Nowadays,Big Data is being broadly utilized in fault detection and diagnostic（FDD）. Thus,the PCAClustering method,which is combined with Big Data,was proposed to diagnose compressor liquid refrigerant floodback fault. First,data quality was improved by data preprocessing; secondly,the principal component analysis（PCA） method was employed to obtain the new dimensional variable data; finally,the new principal variables were clustered by clustering analysis to get the classification label of liquid refrigerant floodback fault data. The results show that the model can preferably diagnose the liquid refrigerant floodback problem in the absence of real label data,with the diagnostic rate of the compressor liquid refrigerant floodback reaching 94. 29%.

关 键 词：多联机系统 压缩机回液 故障检测与诊断 聚类分析 主成分分析 

分 类 号：TB652[一般工业技术—制冷工程] TU831.4[建筑科学—供热、供燃气、通风及空调工程]