UC Irvine

A set of 16 aeroengine micromixing injectors were adapted from a lean direct injection of fuel concept to be operated on hydrogen, or hydrogen/natural gas blends. These come from a statistical design to study three of their main geometric factors (the air split, the fuel swirl and the air swirl), and were designed such that their effective areas were essentially the same. Collins Aerospace was in charge of the injector design and supply, targeting flows linked to a Solar Turbines industrial engine.

UCICL, with the support of the Department of Energy contract DE-FE0032073, was in charge of this first screening testing at atmospheric conditions, and the next demonstration stage with a laboratory scale model combustion at higher pressures.

A three-level Box-Behnken design was first employed to evaluate the lean blow-out limits of the test hardware, depending on the preheat temperature of the air flow, the air pressure drop across the airbox, and the fuel composition (in terms of the percentage of hydrogen in the hydrogen/natural gas blend). This resulted in some clear trends on the injector stability as a function of the combustion conditions, but also the injector geometry.

The fuel composition was the main factor on the predicted model, being the combustion conditions with a higher hydrogen content those with a wider range of stability, because of hydrogen's higher reactivity and diffusivity. Lower pressure drops (and therefore lower air flow velocity) also enhanced the stability of the combustion, and so did lower preheat temperature levels. On the injector geometry factors, both low levels of air split and air swirl expanded the lower flammability limits, because of the reduction on the mixing with the fuel flow and therefore yielding into diffusive flames (which showed longer stability ranges, specially for hydrogen fuels).

Once the stability was assessed, the so-called screening phase of the project was conducted. Again, a three-level Box-Behnken design was designed adding an extra factor to the previous three, this being the adiabatic flame temperature. Emissions and flame diagnostics were evaluated simultaneously, for the resulting testing matrix for each of the 16 injectors. The emissions data was recorded on a volumetric dry basis corrected to 15% oxygen, and then converted into a mass basis in ng/J, aligning with the most recently reported bias with the reported basis for hydrogen versus natural gas combustion.

Regarding the flame structure characterization, the main focus of the current work, three different cameras were operated in parallel. A Nikon D90 was used to capture the visible spectrum of the flame, fixing its operating mode to manual with a 2.5 shutter speed and F4 aperture. An FB-N9-U Dynacolor camera, mounting a Sony CMOS sensor, was used to record the OH* chemiluminescence (in the UV spectrum), fixing its exposure time to 0.9999 seconds and electronic gain to 6 dB. Finally, a Phantom v7.1 was used to record the high-speed dynamics of the flame, with 200 frames per second as the main factor to consider.

The study mainly focused on the responses extracted from the OH* chemiluminescence imaging. Because of its direct correlation with the rate of heat release, it is commonly used to characterize the flame front. These responses were, for the most part, the flame area and its average brightness, the heat release area and its center of gravity and leading edge. To assess the importance of the tested factors and their interactions, an analysis of variance was performed for each response, with significant indicators of p-values lower than 0.05. The obtained models were first evaluated against the variability of the responses, and then using the statistical parameters that the used software provided.

The effect of the confinement ratio of the flame by the combustor quartz tube was also studied, since its dimensions were arbitrarily defined during the design of the setup. A reduction on its length, and next on its inner diameter, concluded that no significant change occurred on the flame structure because of it.

Finally, the presented thesis concludes with the coupling of both the NOx emissions study and the imaging flame characterization. To do so, both the statistical software DesignExpert and a Random Forest machine learning algorithm were used. The latter performed significantly better than the forming, yielding into 80%-accurate predictions of the NOx emissions as a function of the injector geometry, the combustion conditions, and OH* chemiluminescence responses.

This coupled study was also utilized to reaffirm the conclusions that the emissions study extracted: the injector configuration with lower nitrogen oxides emissions is such with higher levels of air split and air swirl, and lower levels of fuel swirl.