基于数据-机理联合驱动的制冷空调系统故障特征提取方法  

Data-driven and physical-driven combined fault feature extraction method for refrigeration and air-conditioning systems

作　　者：孙哲 金华强 李康 顾江萍[2] 黄跃进[1] 沈希[1,3] SUN Zhe;JIN Huaqiang;LI Kang;GU Jiangping;HUANG Yuejin;SHEN Xi(College of Mechanical Engineering,Zhejiang University of Technology,Hangzhou 310014;School of Engineering,Zhejiang Agriculture and Forestry University,Hangzhou 311300;College of Education,Zhejiang University of Technology,Hangzhou 310014)

机构地区：[1]浙江工业大学机械工程学院,杭州310014 [2]浙江工业大学教育科学与技术学院,杭州311300 [3]浙江农林大学光机电工程学院,杭州310014

出　　处：《高技术通讯》2023年第7期772-780,共9页Chinese High Technology Letters

基　　金：浙江省自然科学基金(LQ23E060006);浙江省重点研发计划(2020C04010)资助项目

摘　　要：制冷空调系统故障特征提取是系统故障诊断的基础。现有研究常利用端到端的黑箱模型实现自主特征提取,获得的特征不具备物理解析,无法保证其全局应用的可靠性。针对制冷空调系统展开具有明确物理意义的特征提取研究对实现可靠可信故障诊断具有重要意义。本文提出一种数据-机理联合驱动的故障特征提取方法,针对动态运行数据复杂多变特性,构建制冷系统基准模型和偏离特性表征策略,实现故障偏离特征的准确提取,并结合热力学机理分析对其进行理论解释。利用ASHRAE RP-1043数据集进行实验验证,获得6类典型故障特征并构建偏离特征矢量表,为故障诊断提供理论基础。The extraction of fault features from refrigeration and air-conditioning systems forms the foundation of fault diagnosis.Most of the existing research uses end-to-end model to achieve autonomous feature extraction.However,the resultant features lack physical analysis and thus,their reliability cannot be guaranteed.Research on interpretable feature extraction for refrigeration and air-conditioning systems is of great significance to achieve reliable fault diagnosis.A combined data-driven and physical-driven method for fault feature extraction is proposed.Focusing on the complex of dynamic operating data,a refrigeration system benchmark model and a deviation characterization strategy are proposed to achieve accurate extraction of fault deviation,and theoretical explanations are combined with thermodynamic mechanism analysis.Finally,the ASHRAE RP-1043 dataset is used for verification,and 6 types of typical fault features are obtained and the deviation feature vector table is constructed to provide a theoretical basis for fault diagnosis.

关 键 词：数据-机理联合驱动 故障特征提取 制冷空调系统 深度学习 

分 类 号：TB657.2[一般工业技术—制冷工程]