A Chiller Fault Detection And Diagnosis Method Based On Developed Principal Component Analysis

Faults exist in HVAC system widely,particularly occurring in chiller system.Faults not only incur waste of energy and reduce indoor thremal quality,but also shorten life of equipment leading to high maintenance cost.Although many researchs have been devoted to chiller fault detection and diagnosis(FDD),the models con-structed have the characteristics of fixed with non-time-varying,single method and can't be updated.Thus,a developed principal component analysis method,namely moving window multiscale principle component analysis(MW-MSPCA),is proposed to detect and diagnose faults in chiller.Ultilizing the advantages of principal component analysis(PCA),wavelet packet analysis and moving window technology,this paper presents three chiller FDD strate-gies base on PCA,multi-scale PCA(MSPCA)and MW-MSPCA and builds corre-sponding models of three strategies.Verified and compared by ASHRAE 1043-RP and actual engineering test data for three modes,the conclusions are as follows:MW-MSPCA method can realize chiller FDD of the dynamic non-stationary process except the start-up and shutdown process,but not only the steady or quasi-steady state process.MW-MSPCA method can use part of the operation condition data as the training data to construct model without affecting performance of the FDD model.Analyzing the data from the view of multi-scale,MW-MSPCA method can eliminates the influence of data noise for FDD results.The dynamic self-updating model constructed by MW-MSPCA method has higher accuracy and sensitivity in chiller FDD,and it can also achieve detection and diagnosis of low grade faults.Compared with traditional 2-di-mensional contribution diagram,the 3-dimensional bar cahrat using by MW-MSPCA method is more intuitive and accurate in faults diagnosis.The verification results from experimental data and actual engineering test data indicate good generality of the pro-posed MW-MSPCA method.Compared with the experimental results from the proposed three models,the dy-namic self-updating model constructed by MW-MSPCA method is superior to the other two models in terms of the correctness of chiller FDD and the sensitivity of the model to faults or fault grade.