Lifetime Prediction Of Proton Exchange Membrane Fuel Cells

As one of the most promising and most popular new energy technologies,proton exchange membrane fuel cells(PEMFCs)have many advantages such as pollution-free,high energy con-version rate,low operating temperature and low noise.They can be widely used in transportation vehicles,combined heat and power,fixed base stations,portable devices and many other appli-cations.Although there are many actual industrial applications of PEMFCs are currently being deployed and demonstrated throughout the world,PEMFCs still face two major bottlenecks on the road to large-scale commercial applications,namely the limited lifetime under actual operat-ing constraints and the high cost.Therefore,most of the current research on PEMFCs focuses on these two issues:extending the service life and reducing the cost of PEMFCs.In addition to developing the high-performance and low-cost new materials and designing new PEMFC system structures,effective health management methods such as prognostics and fault diagnosis in parallel can also effectively increase the service life and reduce the maintenance costs of PEMFCs.Prognostics can predict the future degradation trend(FDT)of the target system as well as the failure time and risk.It is the key technology to transform the system maintenance from the traditional“failure and repair”to“prediction and prevention”.It is defined as an estimate of the future time to failure of the target system and predict the risk of one or more existing and future failure modes.The complete prognostics process includes the prediction of FDT and the estimation of remaining useful life(RUL).Studies have shown that effective prognostics methods can not only improve the life,reliability and safety of PEMFCs,but also reduce the cost and downtime.Therefore,prognostics is one of the effective solutions to improve the short service life and high cost of PEMFCs,and it is also the main research content of this paper.According to the contributions of different prognostics studies,the advances of prognostics of PEMFCs can be divided into the innovation of prognostics methods and innovation of degradation indexes.The main research contents and contributions of this paper are summarized as follows:·The background and significance of prognostics research for PEMFCs are introduced.The development and current status of PEMFCs prognostics research are reviewed systematically from two aspects:prognostics methods and degradation indexes.·The neural network based model-free prognostics method is proposed to realize the FDT prediction for PEMFCs.In this study,the short-term FDT prediction of PEMFCs is studied from the structural selection problem and the parameter adjustment problem of neural networks.Firstly,the neural network structure which is most suitable for the short-term FDT prediction of PEMFCs is determined through a comparative study.Based on the aging data of PEMFCs,three typical neural networks with different structures are selected for comparison,namely Elman neural net-work(ENN)with feedback structure,group method of data handling(GMDH)with multi-layer feed-forward structure and Adaptive Neural Fuzzy Inference System(ANFIS)with fuzzy neural network structure.The comparison results show that ANFIS combined with the advantages of fuzzy inference and neural network learning has the best short-term FDT prediction performance.At the same time,it is also found that the original sequence can be decomposed into multiple subse-quences by discrete wavelet decomposition method to obtain more accurate and stable short-term FDT prediction results.Then,based on the comparative study,the automatic parameter adjust-ment of ANFIS is realized by introducing the automatic machine learning(AutoML)algorithm The Particle Swarm Optimization(PSO)algorithm is introduced,and the membership function parameters of ANFIS are used as the variables of PSO algorithm,and the prediction accuracy of ANFIS in the training process is used as the fitness of PSO algorithm.The automatic parameter adjustment of ANFIS is realized by the optimization process of PSO algorithm.·The model-based method for RUL estimation of PEMFCs is proposed.A semi-mechanism voltage degradation model combined with the unscented Kalman filter(UKF)algorithm is first pro-posed to realize the RUL estimation of PEMFCs.At the same time,considering that the traditional UKF algorithm has a series of problems such as difficulty in initial parameter setting,inability to adaptively adjust the noise covariance parameters,and large error in estimation results,a covari-ance matching method is introduced to adaptively adjust the noise covariance parameters based on the estimation errors.Compared with the traditional UKF algorithm,the adaptive UKF(AUKF)algorithm not only can adaptively adjust the noise covariance parameters based on the estimation errors and improve the accuracy of the estimation results,but also greatly reduces the difficulty of the initial parameter setting and the effects of unreasonable initial parameters setting.The model-free method is good at learning and predicting the FDT and the model-based method is good at extracting the degradation indexes/features.Then,a hybrid prognostics method for PEMFCs is proposed by combining two different methods.Firstly,the accurate long-term FDT prediction of PEMFCs is realized by using the AutoML algorithm proposed in the previous study.Then,based on the long-term FDT prediction results,the accurate RUL estimation is realized by combining the semi-mechanism voltage degradation model and the AUKF algorithm.·The multi-scale hybrid degradation index(HDI)of PEMFCs is developed.Due to the limi-tations of the component degradation indexes of PEMFCs,that is,a component degradation index can only characterize the aging state of a component at a certain scale,and cannot characterize the overall aging state of PEMFCs.Therefore,this paper studies how to effectively fuse different com-ponent degradation indexes at different scales to develop a multi-scale HDI that can characterize the overall aging state of PEMFCs.First,the two key components that affect the aging of PEM-FCs,membrane and electrode are selected.Based on the aging data of PEMFCs and the different component degradation models,the degradation indexes of two components at different scales are estimated,namely Pt particle radius,electrochemical surface area,membrane thickness,fluorine release rate,and oxygen crossover rate.Then,how to fuse different degradation indexes is studied.The normalization method is first utilized to transform each degradation index into a unitless index with a range between[0,1].The effects of electrode and membrane degradation are quantified by analyzing the respective proportions of different voltage polarization losses(including activation loss,ohmic loss,concentration loss,and crossover loss)during the aging process of PEMFCs,thus determining the weighting coefficients of different normalized indexes in the HDI model.Finally,the estimated HDI is used to realize the prognostics for PEMFCs and its effectiveness is proved.·Finally,the contributions of this paper are summarized and the future research directions are discussed.