Improved Principal Component Analysis-based Method Of Fault Detection And Diagnosis For Refrigeration And Air-conditioning Systems

With the continuous development of the society,in the field of refrigeration and air conditioning,the corresponding index such as Indoor Environment Quality and Energysaving issues have been received far more attention.In order to achieve both improving the quality of indoor environment and energy conservation goals,inside of an air-conditioning system installs various sensors to measure some thermodynamic parameters of main state points.If the sensors broke down,fault will not only affect the normal operational conditions of the system,but also will lead to the increasing energy consumption.As a result,sensor fault detection and diagnosis(FDD)research for refrigeration and air conditioning system is of much theoretical significance and engineering application value.In recent years,with the rise of data-driven methods,more and more attention has been paid to the study of its applying to air conditioning systems.As a common sensor FDD method,principal component analysis(PCA)is one of the basic data-driven methods.The thesis chooses actual typical refrigeration and air conditioning systems as experimental objects.Firstly,it is established that sensor fault analysis models of air conditioning systems,and based on the traditional PCA-based sensor FDD strategy,two FDD strategy based on improved PCA is proposed: one is an optimized data-driven fault detection strategy based on PCA and preprocessed by clustering analysis,and the other is an improved neural network fault diagnosis strategy with PCA pretreatment.According to the characteristics of different experimental objects,different experimental means and FDD strategies are employed so that different data analysis models based on improved PCA are established.Then,under the assigned experimental conditions,data analysis models are tested and analyzed.The results show that: the combined clustering method of fusion subtraction clustering and k-means clustering realizes the classification and preprocessing process of training and testing data,and the established multiple PCA models improve the performance in multi-condition sensor fault detection compared to traditional PCA;Under a premise of the same accuracy,the reduction of data dimension through process of feature variable extraction of PCA can reduce the number of nodes that input into neural network and thus reduce training time of neural network.