Research And Implementation Of Fault Diagnosis For Proton Exchange Membrane Fuel Cell

Proton exchange membrane fuel cell(PEMFC),as a new type of clean,high-efficiency energy,is being closely watched by researchers and industry all over the world.In order to improve the value of fuel cells in practical production applications,research on the detection of fuel cell operating failures has also received increasing attention.With the development of storage devices nowadays,the application of industrial big data,how to use these massive operating data for fault research has become a popular direction.The fault data generated by the proton exchange membrane fuel cell during operation has the characteristics of strong correlation timing and unbalanced data samples.Therefore,the Long Short Term Memory Network(LSTM)is used as the main fault diagnosis method;There are 8 kinds of fault conditions and normal conditions in the exchange membrane fuel cell test system.During the training process,considering that some conditions of the data are limited due to the collection conditions,the sample imbalance problem leads to the problem of low efficiency of partial fault detection.Based on the LSTM method,it is improved and the DAE-LSTM method is constructed for fault diagnosis research.The main research contents are as follows:(1)In view of the problem that the data of proton exchange membrane fuel cells in different operating conditions have uneven distribution and are susceptible to noise interference,this paper combines the denoising autoencoder(DAE)pair to restore the original input and be able to restore the original input in a complex noise environment.After extracting the characteristics of more robust features,a DAE-based data denoising process is constructed.Compared with the sparse autoencoders(SAE)method,the results show that the reconstructed original sample data features are more robust.(2)Aiming at the strong correlation timing characteristics between the fault data collected by the proton exchange membrane fuel cell,this paper combines the long dependency relationship between the current data value and the previous data value of the LSTM network to process the time series data,and constructs the DAE-LSTM diagnostic model.The diagnosis model has a three-layer structure of noise reduction coding layer,feature extraction layer and fault diagnosis layer.Through comparison experiments with traditional machine learning methods and simple LSTM diagnostic methods,the DAE-LSTM method designed in this paper has a fault diagnosis accuracy rate of 94.360%,49.129%higher than PCA,24.857%higher than SVM,and 19.104%higher than RNN method.Compared with the LSTM method,it has increased by 14.397%,especially in the case 3(exhaust pipe clogging failure)where the sample data volume is seriously missing compared to other cases,the accuracy of diagnosis is significantly improved.(3)The possible failures of the proton exchange membrane fuel cell test system are experimentally summarized.On this basis,the functional requirements and performance requirements of the diagnostic system are analyzed,and the overall scheme of the fault diagnosis system is designed and determined.The designed DAE-LSTM algorithm model completes the realization of the diagnostic system.The diagnosis system provides online fault diagnosis function,takes real-time voltage data as input,conducts fault diagnosis,and displays fault diagnosis results,battery voltage trends and key parameters of the platform in real time;the system has a diagnostic model training function,which can carry out algorithm models on the system The library is updated,managed and retrained;meanwhile,the system also provides basic functions such as user authority login,experiment record management and personnel management.After the system function test and performance test,the effectiveness of the system fault diagnosis is verified,and a variety of working conditions are successfully tested.Tests show that the success rate of system fault diagnosis is over 94%,and the diagnostic error is less than 0.1%.The feasibility of DAE-LSTM method is verified from the system level.