System Construction Of Fault Diagnosis And Early Warning In Refrigeration Equipment Based On Multi-dimensional Time Series Characteristics

Refrigeration equipment is one of the main equipment in modern large machine room,which plays an important role in the field of industrial production,people’s life,national defense and security.However,the structure of refrigeration equipment is complex and the components are highly related.It is of great significance to carry on online fault diagnosis and early warning of the refrigeration equipment.It can ensure the safe operation of large machine room.Based on the comprehensive analysis of the principle of the refrigeration equipment,we focus on the fault diagnosis and early warning technology about the key components of the refrigeration equipment.And we design and realize the real-time monitoring and abnormal condition early warning system of the refrigeration equipment.The main work of this paper includes:1)We study the basic composition and working principle of refrigeration equipment in detail,summary the failure mode of refrigeration equipment.And the important characteristics and threshold range of key components are analyzed to provide the basis for subsequent fault diagnosis and early warning.2)A fault diagnosis method of refrigeration equipment based on multidimensional time series characteristics is proposed.Aiming at the defect of traditional statistical analysis method in the face of high-dimensional and small sample data,we firstly analyze the fault data of centrifugal water chiller based on the widely used open data set(RP-1043).The open data set covers a wide range of fault types.Through increasing dimension,data enhancement preprocessing,adding learning rate reduction strategy and dropout layer,we construct the BP neural network fault diagnosis method.Compared with traditional SVM,PCA and BP neural network,the model improves the accuracy of fault diagnosis by 2%9%.Furthermore,in view of the imbalance of sample number in practical application,on the basis of the proposed model,the down-sampling method is integrated to realize the fault diagnosis of water chiller based on multidimensional time series characteristics,which further verifies the practicability of the proposed model.3)An improved long short term memory(LSTM)multi-dimensional time series feature fault early warning method is proposed.Firstly,the LSTM model is used to predict the characteristics of each dimension of the chiller in different periods.Secondly,aiming at the limited selection of traditional LSTM memory module and the problem that the length of input sequence should not be too large,an improved multidimensional time series feature fault early warning model of refrigeration equipment is established through data preprocessing,correlation analysis,model training and model evaluation.The experimental results show that the proposed algorithm has higher prediction accuracy than linear regression,SVR,GRU,RNN,LSTM.4)A fault diagnosis and early warning system based on multi-dimensional time series characteristics is designed and constructed.The functions of personnel information management,equipment information management and data query,fault diagnosis and early warning are realized.The reliability of each functional module is verified by system test,which provides technical support for intelligent monitoring of large machine room.