The Research On The Sensor Fault Detection And Diagnosis Of The Central Air-conditioning Water System Based On PCA

Fault detection and diagnosis in air-conditioning systems is important inimproving energy efficiency、maintaining a comfortable indoor thermal and humidityenvironment、improving the IAQ and extending the service time of the equipment.However, it has been beyond operators’ power to detect and diagnose the faults inair-conditioning systems rapidly and timely for a complicated air-conditioning system.Therefore, developing automatic fault detection and diagnosis (AFDD) systems arebecoming increasingly necessary. The sensor fault detection and diagnosis is the basisof an air-conditioning system’s fault detection and diagnosis. Nowadays, it doesn’thave too much valuable research in this field. Therefore, it has a theoriticalsignificance and application value to carry out the research on the sensor faultdetection and diagnosis of an air-conditioning system.This paper uses the method of principal component analysis (PCA) to detect anddiagnose the sensor fault in an air-conditioning system. PCA method is a typicalmultivariate statistical analysis method. It establishes the system’s model with itsmeasurement data in the normal operating condition. Then, the SPE test and T2testare adopted in the sensor fault detection and diagnosis based on PCA model. Thesensor fault detection and diagnosis based on PCA have four test results. Normally,the test result in which T2statistic is higher than its threshold and SPE statistic islower than its threshold is simply thought as being caused by the change of theoperating conditions or by noises. However, it also can be caused by the sensor fault.The fault with the above test result is named as T2-type sensor fault.An improved PCA method can be used to detect the T2-type sensor fault. Themethod divides SPE statistic into two new statistics PVR statistic composed ofprediction residuals of the variables significantly associated with the PrincipalComponents and CVR statistic composed of prediction residuals of the rest variables.Thus, the PVR test, CVR test and T2test are used in the sensor fault detection anddiagnosis. The improved PCA has8test results. So, the change of the system can bedescribed in detail by the improved PCA, which can effectively identify the reason ofthe above test result. In this way, the T2-type sensor fault is detected.The T2-type sensor fault diagnosis includes fault recognition and faultreconstruction. This paper defines one sensor vali dity index, namely SVIT, to recognise T2-type fault. If the SVIT index is close to0, T2-type sensor fault occurs inthe direction of fault reconstruction; If SVIT index close to1, it indicates that there isno sensor fault in the direction of fault recon struction. In essence, fault reconstructionis a process of seeking for a correct estimation of fault data. The paper adopts aniterative reconstruction method to reconstruct the T2-type fault.This paper also presents a self-Adaptive PCA (namely APCA) method. Themethod can remove the error samples of the modeling data improving the faultdetection efficiency. The detection results of the Normal PCA (namely NPCA) andAPCA method indicate that APCA can significantly enhance the sensor fault detection,especially at the low fault level. APCA also improves the fault detection efficiency ofthe negative fault (the fault’s value less than the true value).Finally, the paper compares two methods of determining the number of principalcomponent in detail, that is, cumulative percent variance (CPV) and bestreconstruction (BR). Two methods have their own advantages, but they also havetheir own limitations. BR has a better effect.