Research On The Fault Diagnosis And Remaining Life Prediction For Fuel Cells Based On Deep Learning

In recent years,with the aggravation of energy shortage and air pollution,proton exchange membrane fuel cell has been widely used in vehicles represented by cars and ships because of its advantages of high energy efficiency,zero emission and strong temperature adaptability.However,due to the short development history,the current fuel cell has the problems of low reliability and poor durability,which seriously restricts the large-scale commercial application of fuel cell.Developing new materials with high reliability and long life,establishing highprecision processing production lines and implementing accurate battery control strategies are effective measures to improve reliability and durability.The most important cell control strategies are health management strategies(PHM)such as residual service life prediction(RUL)and fault diagnosis(FD).These strategies can timely find abnormal conditions or fault factors affecting the performance and life of fuel cells,and improve the service life and reliability of fuel cells.Aiming at the health management strategy of fuel cell fault diagnosis and remaining service life prediction,this paper is based on long-term fuel cell life test data and real vehicle fault simulation test data under multiple operating conditions.in this paper,a construction method of diagnosis model and prediction framework based on deep learning is proposed,and the measured data are used to train and verify the effectiveness.The main research contents and innovations of this paper are summarized as follows:1.The failure mode and attenuation mechanism of fuel cell are analyzed and summarized.According to the above research results,a fault simulation experiment is designed to get the fault simulation data set.The preprocessing method of the data set,the selection of input parameters,the fault index and aging index of the fuel cell system are determined.2.The residual life prediction method of fuel cell based on long-term and short-term memory(LSTM)neural network is studied.The effects of different neural networks and different model optimization methods on residual life prediction are compared.Finally,a fuel cell life prediction method based on PCA-BA-ATTENTION-LSTM model is proposed.This method can realize the off-line prediction function of long-term life and short-term life,help to reduce the cost of evaluating the life of fuel cells,timely maintain fuel cells,adjust the design scheme and operation strategy of fuel cells,and prolong the service life.3.A fuel cell on-line fault diagnosis and recovery strategy based on LSTM neural network is proposed,which plays an important auxiliary role in the water state management of fuel cell system.The advantage is that it can realize on-line fault diagnosis of fuel cell system,use the existing measurement parameters to realize the diagnosis function and recover the fault problems in time.On the one hand,it avoids adding complex sensors in the fuel cell system to reduce the total cost of the system;on the other hand,it helps to improve the reliability and durability of the fuel cell.The above research results show that the deep learning method has a good effect on the life prediction and fault diagnosis of fuel cells.To a certain extent,it solves the pain points such as the traditional diagnosis and prediction methods rely too much on expert experience,the diagnosis and prediction process is not universal,and the need for more or more expensive measuring instruments and equipment.The test results of the fuel cell life prediction method proposed in this paper on the French fuel cell laboratory(FClab)data set and the proposed fuel cell fault diagnosis method on the simulated fault data set have achieved high prediction and diagnosis accuracy.