Physical And Data-driven Modeling For Water And Thermal Management Of Proton Exchange Membrane Fuel Cells

Proton exchange membrane(PEM)fuel cell is a high-efficiency and zero-emission energy conversion device that directly converts chemical energy of fuels and oxidants into electricity.Water and thermal management infers to the physical processes and states inside the fuel cell that directly determines the cell performance and durability.Physical model and data-driven model are the two mains methods for the water and thermal management studies and achieve the optimization,and each of methods has its unique characteristics and applications.Therefore,this dissertation focuses on the development of physical and data-driven models for the water and thermal management of PEM fuel cells.Three models,quasi-two-dimensional(quasi-2D)transient model,three-dimensional(3D)computational fluid dynamic(CFD)model,and datadriven surrogate model are employed to achieve high-efficiency prediction and optimization on the cell-level,flow field-level and electrode-level,respectively.On the cell-level,the quasi-2D transient model of the PEM fuel cell is developed.In order to obtain the physical field distributions and ensure the high computational efficiency simultaneously,the transport mechanisms are simplified as the quasi-2D approach.Diffusion along the through-plane direction is the main transport in the membrane electrode assembly(MEA),and the transport in the flow field is simplified as the one-dimensional(1D)transport along the straight channel.Heat and mass transport,water phase change,nitrogen crossover and electrochemical kinetics are taken into account.The developed model has high computational efficiency compared to the 3D CFD model,and is suitable for the simulation of the PEM fuel cell transient process.Then,the quasi-2D transient model is employed to investigate the transient behaviors of the PEM fuel cell with anode recirculation.The simulation results show that for a PEM fuel cell initially operated with dry hydrogen,the selfhumidification effect is obvious to improve the fuel cell performance in the beginning,and then the performance constantly decreases caused by nitrogen crossover and accumulation in the anode.Increasing anode stoichiometry can enhance the self-humidification effect,reduce the voltage decline rate,and improve the species distribution along the channel and accordingly avoid the local fuel starvation.The purge strategy optimization with anode recirculation is further investigated.For voltage-based purge,purge interval is defined by the voltage drop rate of the voltage peak,and therefore the optimal purge duration is defined as the purge stops when the voltage starts falling.Energy efficiency and fuel loss rate both increase with decreasing purge interval for the simulated operating conditions.On the flow field-level,the 3D two-phase full cell model is employed for the flow field design,and a novel dot matrix and sloping baffle flow field plate of the cathode is originally proposed.The plate consists of dispersive and arrayed blocks with sloping angles as the shoulder,and the top surface of each block is diamond shape.The 3D full cell model and volume of fluid(VOF)model are conducted to predict cell output performance,internal transport process and liquid water removal process of PEM fuel cell with the designed matrix flow field.The simulation results show that the matrix flow field can effectively improve cell output performance,especially in high current density region.The advantages of the matrix flow field can be summarized as the enhancement of oxygen supply to the gas diffusion layer(GDL)and improvement of oxygen uniform distribution.On the electrode-level,this dissertation proposes the data-driven surrogate model framework which combines the 3D physical model and data-driven model.Simulation results of a state-ofthe-art 3D physical model of PEM fuel cells construct a dataset and are applied to train the surrogate model.The developed surrogate model has a comparable accuracy with the physical model,but much higher computation efficiency.It can replace the physical model to achieve the high-efficiency and-accuracy simulation.Firstly,the surrogate model is employed to achieve the multi-physical field prediction of PEM fuel cells,namely the digital twin.The results demonstrate that the predicted multi-physics fields well mirror the main distribution characteristics of the simulation results.Based on the digital twin,two model-based designs,the PEM fuel cell healthy operation envelope and PEM fuel cell state map,are demonstrated which could be used to understand the healthy operation ranges in advance and run as the digital twin embedded into the PEM fuel cell control system.Secondly,the catalyst layer(CL)composition optimization to increasing the maximum power density of PEM fuel cells is conducted by combining the surrogate model and stochastic optimization algorithm.According to the optimized CL composition,the percentage error between surrogate model predicted and physical model simulated maximum power densities is only 1.3950% which demonstrates the optimization validity of the proposed framework.