Research On Fault Diagnosis Method Of Chiller Based On LSA And GRU Network

Chiller is one of the essential equipment in industry,agriculture and medical fields.It can meet the needs of various occasions of refrigeration and maintain a constant temperature and humidity.The operating principle and internal structure of chiller are more complex than other mechanical equipment,because the chiller involves the cooperation between several components and the process of refrigeration cycle.Once a fault occurs,it can be complicated to troubleshoot because it involves the interaction of multiple components.The problem of manual experience error exists in the process of fault diagnosis of chiller.How to deal with the data generated during the operation to make the diagnostic method more effective is also an important problem.Therefore,thesis mainly studies the chiller fault variable selection,fault diagnosis model construction and intelligent fault diagnosis system construction,and establishes a set of fault diagnosis framework.The main tasks are as follows:1)The FSLSA method is proposed for fault variable selection.The performance of chiller will gradually deteriorate during operation.By analyzing the data collected by the sensor,the cause of performance deterioration can be found in time.However,because there are many variables in the data,it is necessary to select some variables.Therefore,this paper proposes a method for fault variable selection in chiller.The FSLSA method combines the Fisher Score method with an improved Lightning search algorithm.By setting a threshold,the features are firstly selected,and then the improved Lightning search algorithm is used to make secondary selection on the basis of retaining the features.Finally,the variable set that is sensitive to faults is obtained.It is verified that the accuracy of the model is not affected when these variables are used for fault diagnosis.This shows that the FSLSA method can effectively reduce the number of feature subsets and maintain high fault diagnosis accuracy through feature selection.2)A chiller fault diagnosis method is proposed.In practical applications,due to various factors such as equipment conditions and data collection difficulties,it is often difficult to obtain chiller fault data.As a result,the data volume is usually limited,which affects the performance and reliability of fault diagnosis methods.This paper proposes the AGRN method,which can diagnose chiller faults with limited data samples.This method makes full use of the hidden layer cell information of the previous time step to improve the GRU gated cell and make the input it accepts more diversified.It is used as the feature extraction module of the relational network to improve the diagnostic accuracy of the model in the case of lack of data,and to provide an effective solution to solve the practical problems in the fault diagnosis of chiller.3)The proposed method is integrated into the system,which is convenient to monitor the operating condition of the chiller and discover the fault information in time,so as to realize the intelligent fault diagnosis of the chiller.The system can visualize the operation data of the chiller,which is convenient for the staff to analyze and check.It also has the ability to analyze operating variables,calculate the importance of variables and rank them using Fisher Score or FSLSA methods,reducing the scope of data that stuff need to see,so that they can quickly adjust the chiller.In addition,the system also provides the method of fault diagnosis,which can give early warning to the fault in time.