Research On Simultaneous Faults Diagnosis And Fault Intensity Identification Of Variable Refrigerant Flow Air-conditioning System

Due to incorrect installation or equipment aging after long-term operation of Variable Refrigerant Flow(VRF)system,there will be various unavoidable failures,therefore resulting in energy waste.In addition,operation under a fault state of the system will also accelerate the damage of equipment,significantly reducing the service life and causing large economic losses.Therefore,establishing a reliable VRF system fault diagnosis model,monitoring the system health status in real-time,and ensuring that the efficient and reliable operation of the system has an energy-saving and economic value.However,existing research on VRF system fault diagnosis seldom considers simultaneous faults and the influence of their intensity on the diagnosis results.Based on numerous experimental data and expert decoupling knowledge,this dissertation takes a general data balance method as the data guarantee module and uses the deep convolutional neural network algorithm as the core tool.Starting from improving the performance of the multiclass single VRF system fault diagnosis,the module evolves into a high-performance comprehensive diagnosis and fault intensity identification model integrating multiclass single faults and multiclass simultaneous faults.The main research contents of this dissertation are as follows:As for the lack of quantitative analysis of the simultaneous fault’s impact on the performance and thermophysical variables of VRF system in existing research,a large number of basic experimental studies on multiclass single faults and multiclass simultaneous faults have been carried out.By using trend analysis and rate-of-change value division methods,we can quantitatively analyze the performance of the VRF system while it has different fault types and intensity.Then we can summarize the influence law of simultaneous fault performance and demonstrate the difficulty of simultaneous fault diagnosis.By using expert knowledge,the typical thermophysical variables under different single and simultaneous faults are preliminarily decoupled.This experimental study provides the fault data basis for subsequent chapters and provides the basis for decoupling thermophysical features for simultaneous fault diagnosis.Under the unbalanced data samples,in order to solve the problem of reducing the accuracy of the fault diagnosis model of the VRF system,a solution method based on principal component analysis and synthetic minority oversampling technology is proposed.This method is validated using experimental data of different types of faults and a variety of commonly used data-driven algorithms.It provides a data balance processing method for the problem of unbalanced labels in subsequent VRF system simultaneous fault diagnosis and intensity identification.In order to solve the accuracy reduction of the multiclass single fault diagnosis model of VRF system under the process of defrosting,a multiclass single fault diagnosis method for VRF system based on a deep convolutional neural network is proposed.By combining experiments and data analysis,we can construct 2 new characteristic variables and introduce a deep convolutional neural network algorithm.By using its unique identification ability feature,the accurate fault diagnosis of the defrosting thermophysical process of the VRF system can be realized.The geometric mean accuracy of the proposed method is as high as 98.73%,which is 7.68% higher than other models on average.This method provides an algorithm basis for subsequent VRF system simultaneous fault diagnosis and provides a guarantee for improving the accuracy of simultaneous fault diagnosis.In view of the lack of comprehensive consideration of single fault and simultaneous fault synchronous diagnosis in existing VRF system diagnosis research,a simultaneous fault diagnosis method based on multi-label convolution neural networks is proposed,on the basis of the original high-performance multiclass single fault diagnosis method improved by ‘multi-label multi-classification’.Firstly,we construct 2 new decoupling variables by using expert knowledge and data analysis,then optimize the parameters of the multi-label deep convolutional neural network by grid search.In this way,the performance optimization of the multiclass single and simultaneous faults synchronize diagnosis can be realized.In the diagnosis of 5 types of typical single faults and 11 types of simultaneous faults,the absolute matching accuracy reaches 98.75%.Finally,in view of the requirement of gradual fault intensity identification of VRF system,on the basis of the original simultaneous fault diagnosis method,we introduce the multi-channel strategy and therefore propose the multi-channel convolutional neural network for the gradient simultaneous fault diagnosis and fault intensity identification method.Using the pre-data balance processing method,the problem of category imbalance caused by label sparseness under the multi-fault intensity is solved.The multi-channel strategy is innovatively used in the output layer of the convolutional neural network algorithm,which realizes the synchronization requirements for the diagnosis of multiclass single faults,multiclass simultaneous faults as well as the identification of their corresponding fault intensity.