Study On Comprehensive Method Of Fault Detection And Diagnosis For Chillers

The chiller is one of the core components of the HVAC system.The faulty operation of the chiller directly affects the system energy consumption and the indoor environment.Therefore,it is of great significance to carry out research on the detection and diagnosis of chiller faults.The research of scholars at domestic and abroad mostly focuses on the development and optimization of single diagnostic models.Due to the complexity of chiller faults and the limitations of diagnostic model algorithms,the diagnostic results of a single model show different performance in different types of faults and low-level faults.Based on this,a comprehensive diagnosis model based on random forest algorithm isproposed in this paper,and the result of a single diagnostic model is combined to further improve the diagnostic accuracy.Firstly the five factors: input parameters,regression models,characteristic parameters,fault thresholds and diagnostic rules are analyzed,which affect the accuracy of the chiller fault diagnosis when the model method is used.By analyzing the working principle and thermal characteristics of the chiller,three parameters that can comprehensively reflect the operating conditions of the chiller are selected as the input parameters of the model.By analyzing the performance characteristics of the fault,the characteristic parameters and fault diagnosis rules are determined.The EWMA method is selected to analyze the residual of the characteristic parameters,to determine the fault threshold of the characteristic parameters,and to optimize the fault threshold by controlling the false alarm rate <0.1.Through literature analysis,considering the complexity of modeling,the accuracy of the model and the frequency of use,three regression models are selected.The polynomial regression diagnosis model,the BP neural network diagnosis model and the support vector regression diagnosis model are established as three single fault diagnosis models.Using the ASHRAE RP1043 project data,three single diagnostic models are used to detect and diagnose five faults at four different severity levels.The result shows that the polynomial regression diagnostic model has the highest diagnostic accuracy for cooling water reduction fault,chilled water reduction fault,excessive refrigerant fault and condenser fouling fault.For refrigerant deficiency fault,the BP neural network diagnostic model has the highest diagnostic accuracy.Different diagnostic models present a large difference for the same type of fault.Based on the single diagnostic model,the comprehensive diagnosis method for chiller faults based on random forest algorithm is proposed.The diagnosis matrix of the comprehensive diagnosis model is established by the diagnosis result of the single fault diagnosis models,and the random forest algorithm is trained.the diagnosis result of the comprehensive diagnostic model is obtained through the 5-fold cross-validation,and the result of comprehensive diagnosis model is analyzed for the different types of faults and the different severity levels of the same fault.The result shows that for five fault,the diagnostic accuracy of the comprehensive diagnostic model is 100%,100%,88.89%,99.07%,and 87.96%,respectively.For the different types of faults,the diagnostic accuracy of the comprehensive diagnostic model is better than the single diagnostic model.In addition,the comprehensive diagnostic model is also superior to the single diagnostic model at low levels of fault.Comprehensive diagnostic models can significantly improve the accuracy of fault detection and diagnosis.The research results of this paper can provide theoretical guidance for fault diagnosis of chillers and operation and maintenance of building electromechanical systems.