Hydrogen Recirculation Ejector


Practical fuel cells cannot consume all the hydrogen or oxygen supplied to them, as this would allow liquid water and inert gases to accumulate in the areas of the cell that are close to the end of the gas channel, resulting in severe performance loss due to low reactant concentration. Instead, they must be provided with some excess gas (the ratio of total gas delivered to gas consumed is known as the stoichiometry). The excess can be simply vented in research fuel cells or small systems where simplicity takes priority over efficiency, but in a large system like that used to power a vehicle, fuel efficiency requires that the gas be recirculated. Virtually all vehicles (except submarines and spacecraft) use ambient air rather than bottled oxygen, which is vented so that the inert nitrogen does not accumulate in the stack, but the hydrogen must still be recirculated by some means.

Many fuel cell systems use mechanical pumps to circulate the hydrogen. Unfortunately, this presents serious challenges: many common engineering materials are corroded by the moist hydrogen found in a fuel cell system, and the pump cannot use oil to lubricate any parts in contact with the gas, lest it enter the fuel cell and coat catalyst sites or plug pores. In addition, pumps necessarily consume power and thus decrease net system efficiency; the pump used on UD's Phase 1 fuel cell bus consumed approximately 250 watts to serve a stack producing a maximum of 17 kW.

Fortunately, an alternative exists: a variable flow ejector can use the mechanical potential energy of the compressed hydrogen that supplies the fuel cell to suck the hydrogen from the stack outlet and reinject it at the inlet. Fixed geometry ejectors are common devices, used in a variety of industrial applications; however, they can only perform well at one specific gas flow rate. Above this flow rate, choked flow occurs in the secondary nozzle throat, and below this flow rate, gas velocity from the primary nozzle is greatly decreased; both of these conditions decrease the pressure rise from secondary inlet to outlet. A variable flow ejector refines this concept by adding a needle that can change the size of the primary nozzle opening so that a consistent gas velocity is maintained across a range of flow conditions, so that performance loss with decreasing flow rate is minimized.

UD's Ejector

Basic Design

The design started with a 2D axisymmetric CFD (computational fluid dynamics) model of the ejector. Geometry was generated using a script with ANSYS GAMBIT 2.4.xx, and the flow equations were then solved with ANSYS FLUENT 6.3.26 using the shear stress transport k-omega turbulent flow model.

The variable flow ejector constructed at UD.
  • High pressure hydrogen enters through the primary inlet,
  • accelerates to Mach 1 in the gap between the needle and the primary nozzle opening,
  • further accelerates to supersonic speeds around the needle,
  • entrains low pressure hydrogen from the secondary inlet,
  • passes through the secondary nozzle throat, where the fast moving jet accelerates the slow moving entrained hydrogen,
  • and passes through the diffuser, slowing down and converting kinetic energy to pressure potential energy.

Because of hydrogen's low density, achieving the desired pressure increase of 4 psi across the stack requires that the flow from the primary nozzle be supersonic. To form a coherent jet when the primary nozzle orifice is occluded by the needle, the needle is made so that it extends through the primary nozzle and into the converging portion of the secondary nozzle. The flow then remains attached to the needle and accelerates, much as in an aerospike rocket motor, reaching up to Mach 2 in UD's ejector design. The needle also has the effect of occluding part of the secondary nozzle's throat when fully extended; since the greatest performance can be achieved by sizing the throat just slightly larger than what would be necessary to achieve sonic flow within, this effect improves the ejector's performance.

Model Validation

A first prototype was built as a combination of machined aluminum and rapid-prototyped plastic. To reduce the cost and safety risks of testing, air was used as a working fluid rather than hydrogen, using the principle of flow similarity by matching dimensionless numbers, specifically Reynolds and Mach numbers. The Reynolds number expresses the ratio of inertial force to viscous force, and the Mach number defines the degree to which compressibility affects the flow. By reducing absolute pressure of air to 0.475 times that of hydrogen, and reducing velocity to 0.247 times that of hydrogen, room temperature air can simulate the behavior of 60°C hydrogen in the same ejector.

The similarity is not perfect; viscosity changes with temperature at a different rate in air as compared to hydrogen, and the lower velocity in the same size channels means that unsteady flows will not be modeled correctly. However, the CFD simulation was run with air as the working fluid, and the results matched closely, indicating that the viscosity difference is not significant (probably because of the high Reynolds numbers seen in the parts of the flow where temperature changes significantly from stagnation conditions). Furthermore, the flow in the ejector appears to be essentially steady, and the volume of the rest of the test apparatus can be adjusted to make the rate of pressure change similar between air and hydrogen.

A test bench was constructed to simulate the conditions inside the fuel cell system. One throttling valve simulated the pressure drop of the fuel cell stack, while another allowed air to escape the system, simulating consumption of hydrogen by the stack. Pressure gauges and flow meters allowed the performance of the ejector to be characterized; the pressure rise was found to be quite close to the results obtained from CFD, despite the relatively crude construction of the first prototype. The first version of the bench included a water tank through which air was bubbled as it recirculated within the system; this allowed the air to be humidified and mixed with small amounts of liquid water, to determine how much water droplets would impede the ejector's operation. Testing showed that even relatively large slugs of liquid water (much larger than the droplets that would be seen in the actual fuel cell system) caused only momentary drops in performance, and plugging of the secondary nozzle throat did not occur.

Control Systems and Design Improvements

To use the ejector in a real fuel cell system, automatic control of the needle position would be needed to maintain the correct system pressure. A mechanical control system using a diaphragm that moved the needle using the difference between stack hydrogen and air pressure was initially considered. However, this option was rejected because the friction between the needle and its seal would have caused unacceptably large pressure errors, and might have also caused the system pressure to oscillate. An electronic control system was therefore designed, with a linear actuator based on a stepper motor and leadscrew selected to allow open-loop position control, driven by a custom circuit board carrying a PIC 18F1220 microcontroller and stepper driver. A PID control algorithm was selected because of the limited processing power available, and initial modeling to determine stability and error of the algorithm was carried out in MATLAB/Simulink. The algorithm was then implemented in PIC-C18 and downloaded to the microcontroller; its inputs were commanded pressure and sensed pressure, and its output was a desired position for the stepper motor.

A new version of the ejector was built, using a custom tool to allow the entire secondary nozzle to be made from aluminum. All parts were then anodized to prevent corrosion by the wet hydrogen. It was also found that attaching the needle directly to the linear actuator shaft caused alignment problems that prevented the needle from closing fully, because of loose tolerances in the actuator. To avoid this problem, the high pressure chamber was redesigned to include two sleeve bearings to fully control the needle's position, and the needle was fitted with an "anvil" on which the actuator presses to move the needle inwards. Outwards force is provided by the pressure inside the chamber, and is sufficient to overcome seal friction even with the lower pressures used in air testing.

Testing with the new control board and ejector components led to improvements in the control algorithm:

  • The derivative term was found to have no substantial effect on overshoot or rise time during step changes in pressure command, and was eliminated.
  • A first-order digital filter was added to the pressure inputs, to reduce the effects of electrical noise.
  • A pressure error deadband was added before the PI algorithm, to reduce hunting caused by hysteresis of the pressure sensor.

The ejector is currently being prepared for bench testing with hydrogen to experimentally validate the similarity assumptions made for air. Once this testing is complete and any necessary changes are made, it will be installed in the Phase 1 fuel cell bus.