Study On Data-driven Chiller Multi-fault Detection And Diagnosis Method And Application

Chillers are an important part of HVAC systems and are often used for air conditioning.In actual industrial applications,chillers have complex structures and variable use environments.Chiller faults can be caused by uncertainties such as improper installation,improper manual operation,improper control strategies,aging of parts and components,and equipment faults,thus bringing a series of problems such as energy waste,user dissatisfaction,economic loss,and safety hazards.Therefore,timely and accurate chiller fault detection and diagnosis are important for the application of chillers in real industrial environments.Based on the data-driven method,ASHRAE RP-1043 experimental data,and the idea of multi-classification,this study used the LSSVM algorithm and XGBoost algorithm to build comprehensive chiller fault detection and diagnosis models covering seven typical chiller faults.The LSSVM and XGBoost models were compared and analyzed from the multi-classification effect and fault diagnosis effect respectively.The results showed that the macro-accuracy of the LSSVM model was 89.32% and the macro-F1 score was 90.61%,the macro-accuracy of the XGBoost model was 97.58% and the macro-F1 score was 97.88%.The multi-classification effect of the XGBoost model was better than that of the LSSVM model,and its ability to solve classification problems with large sample sizes and multiple categories was stronger.The FDR of the LSSVM model for RL,CF,RO and EO was only 64.38%,64.25%,75.61% and 86.06%,and the FDR at Level＿1 and Level＿2 was only 69.82% and 76.67%,the FDR of XGBoost model for FWC,FWE,EO and NC were all over 98%,and the FDR for RL,CF and RO were above 91%,and the FDR at Level＿1 and Level＿2 were 93.28% and 95%.The fault diagnosis effect of the XGBoost model was better than that of the LSSVM model,and its ability to detect and diagnose chiller faults with high data similarity and coupling as well as minor faults was stronger.In addition,the XGBoost fault detection and diagnosis model applicable to the target chiller was created based on actual chiller measurement data in real projects.The macro-accuracy was 99.72% and the average FDR was 99.2%,which exhibited good multi-classification performance and fault diagnosis performance.Finally,a fault detection and diagnosis platform applicable to the target chiller was created based on the model,which could realize autonomous modeling,and update,save and recall the model periodically,so as to achieve fault detection and diagnosis of the target chiller.In practical engineering applications,the fault detection and diagnosis model and the fault detection and diagnosis platform created in this study can help maintenance personnel detect chiller faults and determine the types of faults in a timely manner,so that appropriate maintenance measures can be taken.This can reduce equipment wear and tear,extend equipment life,improve chiller efficiency,and enhance the energy efficiency of HVAC systems.