Research On Online Fault Detection And Diagnosis Method Ofchiller

HVAC system has problems in operation,maintenance and management,it often fails after running for a period of time.And the energy loss accounted for 15%-30% of the total energy consumption in commercial buildings.Chiller is the main energy consuming component of HVAC system.According to statistics,the operating energy consumption of chiller is about 40%-50% of the total energy consumption of the entire air conditioning system,42% of maintenance services and 26% of the maintenance costs are caused by the fault of chiller.It is of great significance for the maintenance of the indoor environment comfort,reducing the loss of equipment and saving energy to apply the fault detection diagnosis to the chiller,find out the fault and eliminate it in time.In this paper,for the problem of online fault detection and diagnosis application and the leaking of fault sample date,support vector methods which need less training sample data are used to detect and diagnosis fault.In order to start-up fault detection with normal running data and detect undefined fault types which are not involved in the fault database,fault detection and diagnosis are considered as a kind of one class classification problem.In order to reduce the false alarm rate of detection,increase the diagnostic accuracy,density weight is introduced into the support vector data description(SVDD),to construct more accurate density weighted support vector data description(DW-SVDD)model.Fault detection and diagnosis are realized based on DW-SVDD which is a improved one class classification.To solve the confusion of diagnosis results in one class classification,which produces type ? error and higher misdiagnosis rate,support vector machine(SVM)is integrated in the diagnosis decision process.Through the method integration,an online fault detection and diagnosis method based on DW-SVDD-SVM is proposed.For the process,the method is divided into two parts: off-line model training and on-line fault detection and diagnosis.In the off-linemodel training phase,the data of normal operation and 7 kinds of typical faults are trained and established their own DW-SVDD models.At the same time,support vector machine(SVM)is used to train all kinds of fault data and build the full faults SVM model.In the process of online fault detection and diagnosis,the method of first detection and post diagnosis is adopted.Firstly,the fault detection is performed based on the trained fault free DW-SVDD model,and all the DW-SVDD fault models are used to diagnose the faults.If the diagnosis is confusing,the trained fully fault SVM model is used to diagnose again.The proposed online fault detection and diagnosis method of chiller based on DW-SVDD-SVM is verified with the experimental data of the ASHRAE RP-1043,and the results are compared with the traditional fault detection and diagnosis method based on SVDD.The results show: in the detection stage,the proposed method significantly reduces the detection false alarm rate(FAR),the FAR of test set decreased from 10.5%to 7%,down more than 30%.The new method maintains the same high detection accuracy for most faults,and even improves the detection accuracy of some faults.In the diagnosis stage,the fault diagnosis accuracy of the new method is higher than that of SVDD,and the confusion of diagnosis results is avoided,the misdiagnosis rate is decreased significantly.At the same time it can effectively detect the undefined types of faults.