基于深度神经网络学习算法的压缩机回液故障诊断  

Fault Dagnosis for Compressor Liquid Floodback Based on Deep Neural NetworksLearning Algorithm

作　　者：张梦华 周镇新 陈焕新[1] 王江宇[1] 袁旭东 曹子涵 钟寒露 胡继孙 ZHANG Menghua;ZHOU Zhenxin;CHEN Huanxin;WANG Jiangyu;YUAN Xudong;CAO Zihan;ZHONG Hanlu;Hu Jisun(Department of Refrigeration and Cryogenics,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China;State Key Laboratory of Compressor Technology(Anhui Laboratory of Compressor Technology),Hefei 230031,Anhui,China;China-EU Institute for Clean and Renewable Energy,Huazhong University of Science and Technology,Wuhan 430074,Hubei,China)

机构地区：[1]华中科技大学能源与动力工程学院,湖北武汉430074 [2]压缩机技术国家重点实验室(压缩机技术安徽省实验室),安徽合肥230031 [3]华中科技大学中欧清洁与可再生能源学院,湖北武汉430074

出　　处：《制冷技术》2020年第6期17-23,共7页Chinese Journal of Refrigeration Technology

基　　金：国家自然科学基金(No.51876070,No.51576074);压缩机技术国家重点实验室(压缩机技术安徽省实验室)开放基金(No.SKL-YSJ201801)。

摘　　要：为了提高多联机系统压缩机回液故障检测率,本文首次提出了一种基于深度神经网络(Deep Neural Networks,DNN)学习算法的多联机压缩机回液故障诊断模型。故障诊断分为数据变量选取、建立初始模型、DNN模型训练和故障诊断分类预测四个主要步骤。实验共设置3种压缩机状态,选取了17个特征变量,建立了深度神经网络模型。结果表明,深度神经网络学习模型能更高效地检测出两种回液故障,准确率高达99.86%,而且相对于有监督算法的决策树模型无需相关性分析和剪枝过程,相对于无监督算法的聚类分析算法模型无需相关性和主成分分析过程,处理过程简便易操作,且效率相较于两者分别提高了3.48%和5.91%。In order to improve the detection rate of liquid return fault of compressor in multi-line system,a multi-line compressor liquid return fault diagnosis model based on deep learning algorithm(DNN)is proposed in this paper.Fault diagnosis is divided into four main steps:data variable selection,initial model establishment,DNN model training,and fault diagnosis classification prediction.A total of three compressor states are set in the experiment,and 17 characteristic variables are selected to establish a deep neural network model.The results show that the deep learning model can more efficiently detect two types of backflow faults with an accuracy rate of 99.86%.Compared with the supervised algorithm,the classification and regression tree(CART)model does not require correlation analysis and pruning.The class analysis algorithm model does not need the correlation and principal component analysis process,the processing process is simple and easy to operate,and the efficiency is 3.48%and 5.91%higher than the other two,respectively.

关 键 词：多联机系统 压缩机回液 故障检测与诊断 深度学习 

分 类 号：TP306.3[自动化与计算机技术—计算机系统结构] TB652[自动化与计算机技术—计算机科学与技术]