Proton Exchange Membrane Fuel Cell (PEMFC) systems require more than the theoretically needed hydrogen (H₂) fuel gas in the anode supply system to achieve high performance and extended operating life. The ejector-based H₂ supply and recirculation system (HSRS) is superior to its mechanical pump-based counterpart, with reduced operation and maintenance costs, low noise, and zero parasitic power consumption. However, the conventional ejector with fixed dimensions of a nozzle can only function within a narrow power range of PEMFC due to its restricted primary inlet pressure and mass flow rate. The ejector theoretical background and analytic models are reviewed to understand its working principles better. The ejector H₂ entrainment capabilities are affected by the key geometric parameters, including nozzle diameter (D_n), mixing chamber diameter (Dₘ) and its length (Lₘ), and the distance between the nozzle exit and the mixing chamber (NXP), as well as PEMFC system operating conditions such as anode pressure, temperature, and relative humidity. All these factors have been thoroughly investigated and analyzed using computational fluid dynamics (CFD) simulations.

This research presented an optimal design of a nested-nozzle ejector to satisfy the H₂ stoichiometric ratio (SR_H₂) for a wide power output range of the PEMFC stack. The nested-nozzle ejector consists of one large nozzle (BN) and one small nozzle (SN) with shared suction, mixing, and diffuser chamber. The BN mode was responsible for the stack high load conditions, while the SN mode performed at low load conditions. A bypass was adopted parallel to the nested-nozzle ejector in HSRS to extend the ejector operating range. The key geometric parameters, including nozzle diameters and the distance between two nozzles, were optimized using CFD simulations to maximize the ejector’s H₂ entrainment capability. The results demonstrated that the optimally designed ejector could provide adequate H₂ gas entrainment to satisfy the stack SR_H₂ from around 9% to 100% output power of a 150 kW PEMFC stack. The nested-nozzle ejector was produced, and a test bench was established to measure the ejector entrainment performance using air. Moreover, the nested-nozzle ejector was compared with the dual-ejector system using two conventional ejectors in terms of the operating range and anode inlet pressure fluctuation. The results showed that the nested-nozzle ejector could greatly reduce the system pressure fluctuation while fulfilling the requested anode SR_H₂.

Moreover, a supervised machine learning (ML) model is developed using a data-driven approach based on CFD simulation results to predict the H₂ entrainment capability of both conventional and nested-nozzle ejectors. The least-squares estimator is applied to obtain the optimal weight w* of the linear regression ML model. The predicted H₂ entrainment capability showed good consistency with the results from CFD simulations. The trained linear regression ML model can also be used to optimal design the ejector key geometric parameters by solving a formulated linear programming (LP) problem. Compared to traditional CFD simulation methods, this approach can greatly simplify the ejector design and simulation process. Based on the ML model, the entertainment performance of the optimized ejector was validated using CFD simulations on small, middle and large-size PEMFC stacks, showing less than 8 percent mean absolute percentage error.

Dynamic models of main components in HSRS and PEMFC system controller were developed using model-based design method in MATLAB/Simulink. Integrated with the existing PEMFC stack model and air supply system model, a closed-loop PEMFC system simulator was completed to validate the accuracy and transit behaviour of system models. The system performance is validated in a 150kW PEMFC system with a nested-nozzle ejector and a bypass. The simulation results demonstrated that the rule-based control strategies using feedforward together with PI control for air supply system and PI feedback control for HSRS can provide rapid response of system models during dynamic load inputs. The integrated system model can be beneficial to the actual product design and system development.