基于数据挖掘与信息融合的制冷设备故障诊断  

Fault Diagnosis of Refrigeration Equipment Based on Data Mining and Information Fusion

作　　者：周旖鋆 武凯[1] 孙宇[1] 杨晓燕 楼晓华 ZHOU Yijun;WU Kai;SUN Yu;YANG Xiaoyan;LOU Xiaohua(School of Mechanical Engineering,Nanjing University of Science and Technology Nanjing,210094,China;Nantong Square Cold Chain Equipment Company Nantong,226300,China)

机构地区：[1]南京理工大学机械工程学院,南京210094 [2]南通四方科技集团股份有限公司,南通226300

出　　处：《振动．测试与诊断》2021年第2期392-398,418,共8页Journal of Vibration,Measurement & Diagnosis

基　　金：国家国际科技合作专项资助项目(2017DFR70090);江苏省高端装备研制赶超资助项目(JSTXZB201706)。

摘　　要：针对传统制冷设备监控系统对单一故障检测较为准确但难以对并发故障做出判断的局限性,提出一种基于指定元分析和支持向量机两种数据挖掘方式的信息融合方法对制冷设备并发故障进行诊断。首先,基于传统指定元分析不适用于非完全正交模式,对指定元分析算法进行改良,提出了一种非完全正交指定元分析方法;其次,通过实验证明非完全正交指定元分析与支持向量机模型均具有识别并发故障的能力,且各自在不同并发故障识别有一定优势;最后,采用加权证据理论对两种模型的诊断结果进行信息融合,融合后诊断效果得到进一步提升。结果表明:命中率提升至99.10%,虚警率降低至0.21%。Traditional monitoring system of refrigeration equipment is accurate in detection single fault,but difficult to judge concurrent faults. In light of this limitation,a data mining method is proposed based on the information fusion method combing the designated cell analysis and support vector machine with weighted evidence theory. First,a non-fully orthogonal designated cell analysis method is proposed to make up the limit of traditional designated cell analysis in the non-fully orthogonal mode. Then,experiments prove that both the non-fully orthogonal designated cell analysis and support vector machines models can identify concurrent faults,and each model has certain advantages in identification of different concurrent fault. Finally,the weighted evidence theory is used to synthesize the diagnostic results of the two models. The hit rate of the post-fusion diagnosis raised to 99.10%,and the false alarm rate is reduced to 0.21%.

关 键 词：制冷设备 并发故障 信息融合 指定元分析 支持向量机 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置] TH184[自动化与计算机技术—控制科学与工程]