Research On Fault Diagnosis Of Simulated Marine Refrigeration System Based On SAE

The refrigeration unit is an important part of the marine auxiliary machinery.Modern ships are developing towards the direction of network and intelligence,and marine refrigeration systems are becoming increasingly complex.If the marine refrigeration system fails,it will reduce the quality of the cargo and the work efficiency of the staff,and result in serious economic losses.Therefore,ensuring the normal operation of the marine refrigeration system have a great significance in reducing operating costs and improving the economy of marine transportation.Deep learning has obvious advantages in dealing with complex problems,and combining deep learning with fault diagnosis has gradually become a research hotspot.This paper Aims at the problem that the traditional fault diagnosis method has low fault detection efficiency when the fault data of the refrigeration system is complicated and the fault degree is low,the Stacked Auto-Encoder in the deep learning is introduced into the fault diagnosis of refrigeration system.And this paper uses SAE network to study the fault diagnosis of refrigeration system by using ASHRAE 1043-RP project data and simulated fault data.The main work is as follows:Building a SAE network with tensorflow,and the ASHRAE 1043-RP project fault data is directly used as input of SAE network fault diagnosis model after simple preprocessing without feature selection,the unlabeled fault data is used to pre-train the network and select features automatically by the method of layer-by-layer greedy training,then fine-tuning the network by error back propagation using labeled fault data,the softmax classifier is used for fault diagnosis finally.Analysising the influence of network structure,epoch value,number ratio of training set to test set,batch_size value,learning rate,optimizer and noise figure on fault classification performance by comparative experiments,summarizing the rules and determining the SAE fault diagnosis network with the best performance.To prove the superiority of SAE network in fault diagnosis of refrigeration system by comparing SAE with other two conventional diagnosis methods.Building the simulation model of marine refrigeration system experimental bench based on Matlab/Simulink,the simulation results are compared with the measured results to verify the accuracy of the model.The faults of refrigerant leakage,lack of chilled water,lack of cooling water,refrigeration capacity mixed with non-condensable gas and condenser scaling are simulated under different fault degrees by adjusting the relevant parameters in the model,the fault analysis is carried out according to the working principle of refrigeration,the simulated fault data is obtained and it is used to be researched by SAE.The results show that SAE has high accuracy and great realtime performance in fault diagnosis,and it still maintains high diagnostic accuracy under low fault level,which can achieve fault diagnosis of minor faults and improve the reliability of refrigeration system.