Research On Sensor Fault Diagnosis Of Ground Source Heat Pump System Based On PCA

With the development of today's social economy,and the requirements for the building environment are constantly improving,buildings are gradually becoming larger,more multifunctional and more intelligent.Nowadays,the automatic control system is becoming more and more popular in various industries.As an important part of the intelligent building system,the air-conditioning fault diagnosis system has gradually increased its proportion and more attention has been paid to this aspect.Because the sensor is an important component of the air conditioning monitoring system,its reliability will affect the real-time regulation and operation control;furthermore,will affect the energy utilization,indoor thermal and humid environment,and air quality.Therefore,it is necessary to conduct fault diagnosis research on the sensor of air conditioning system,which has strong application value.This study first briefly describes a complete fault diagnosis process including the basic idea of the Principle Component Analysis(PCA),mathematical proof method,modelling method,the process of fault diagnosis,the number of principal elements,fault detection method,fault separation method based on contribution graph,etc.Secondly,the ground source heat pump system and its faults was introduced,including the development and application of ground source heat pump system,the ground source heat pump system used in this experiment,the characteristics of four kinds of sensor faults commonly found in air conditioning systems and their mathematical models.Finally,system variables are filtered based on the conditions of the existing system.The principal component analysis method was used to detect and diagnose the sensor fault of the ground source heat pump system.The operation data was collected from the actual ground source heat pump system.In order to verify the diagnostic effect,the Principal Component Analysis method was used to diagnose the three faults of the air conditioning system sensor: Bias fault,Drifting fault,and Precision degradation fault.The results show that sensor fault detection based on the PCA method was effective.However,when a minor fault occurs,the effect of diagnosis was not satisfactory.Two improved methods,the Kernel Principal Component Analysis(KPCA)and the Exponentially Weighted Dynamic Kernel Principal Component Analysis(EWDKPCA),were proposed for the shortcomings of the principal component analysis.The core theory of the kernel principal component analysis can be analyzed that the kernel principal component analysis does not need the accurate mathematical model of the system,onlyuses the correlation between the variables to analyze,then uses the kernel function to perform nonlinear mapping.Therefore,it is possible to describe the nonlinear relationship between variables in a more accurate way while extracting the nonlinear characteristics of the system.Obviously,the kernel principal component analysis is applicable to the fault diagnosis of nonlinear systems such as air conditioning systems.By studying the exponential weighted dynamic autoregressive statistical model and data update with equal step length,an exponentially weighted dynamic kernel principal component analysis is proposed for the process data with dynamic time-varying characteristics.According to use the weighting factort to combine the new and old data,to establish a kernel principal adaptive model with dynamic characteristic.After the verification of both methods to the ground source heat pump air conditioning system,The results show that both methods can improve the diagnosis effect.At last,the diagnostic results of the three diagnostic methods was compared in detail im this study.These three methods are feasible methods,but each of them have their own characteristics and limitations.For example,the characteristics of PCA method is low sensitive and no misdiagnosis,KPCA method has high sensitive and low precision,the indicators of EWDKPCA method are well balanced.From a comprehensive comparison,exponentially weighted dynamic kernel principal component analysis is the method better than others in most air conditioning applications.