Ignition Delay Times and Chemical Kinetic Model Validation for Hydrogen and Ammonia Blending With Natural Gas at Gas Turbine Relevant Conditions

For the past century, fossil fuel combustion has driven the global economy and led the energy sector as the main electricity provider. Although natural gas (NG) has replaced coal as a cleaner burning fuel, carbon dioxide (CO₂) is still a main product of combustion. As part of the worldwide initiative to decrease carbon emissions, studies into hydrogen (H₂) and ammonia (NH₃) combustion for uses in gas turbines have accelerated. Both H₂ and NH₃ are favorable as NG replacements due to their carbon-free emissions; however, both entail major challenges. The storage, transportation, and flame stability of these fuel alternatives have yet to be properly controlled. There are also concerns about the tangibility of these fuels as they both require methane steam reforming to be produced, but Haugen et al. have shown methods of capturing the CO₂ produced during the reforming process [1].

Hydrogen is a favorable fuel because it does not produce carbon nor fuel-nitrogen oxide (NO_x) emissions (unlike ammonia) when burned with pure oxygen (O₂) but draws major issues with its storage, transportation, and high reactivity. H₂ is also prone to flame flashback because of its high burning velocity [2,3], which causes the flame to propagate upstream of the designed location in the combustor. Conversely, blowout is a common issue encountered with NH₃ flame stability due to its slower chemical reactions [4,5], where NO_x emission levels are optimally low. When combusted, L-NH₃ has a higher volumetric energy density than L-H₂, but any ammonia combustor must also be able to confidently have near-complete combustion to prevent leftover toxic NH₃ molecules in the exhaust, or “ammonia slip.” On top of that, nitrous oxide (N₂O), another product from NH₃ combustion, has been said to potentially have a much larger impact on the climate than CO₂ itself [6] but can be effectively mitigated to some extent using selective catalytic reduction [7]. Ammonia can, however, be utilized to act as a hydrogen carrier. Through the reverse Haber–Bosch process, a heated catalytic surface (most likely ruthenium or iron) can be used to break up 2NH₃ ⇔ N₂ + 3H₂ [8]. Additionally, there are already existing L-NH₃ pipelines that have proven reliable for decades and could be used for ammonia transportation [9,10].

The combustion instabilities can be offset when premixed blends of NG/NH₃ or NG/H₂ are utilized, providing a stable and efficient flame with lowered carbonous and nitrogenous emissions while making minimal modifications to existing NG turbines [11,12]. While flame speeds and emissions are more optimal using blends solely of H₂/NH₃, the infrastructure required to swap existing NG pipelines and turbomachinery would be an immensely expensive and time-consuming operation. Several studies have been performed to measure the combustion characteristics of these blends to find an optimum equivalence and fuel ratio during turbine operation, as outlined by Chai et al. [13]. By use of a robust chemical kinetic model, these fuel blends can be optimized using computational fluid dynamics rather than experimental data. The scope of this project aims to experimentally study the ignition delay times (IDTs) of undiluted NG/NH₃/Air and NG/H₂/Air mixtures at the relevant pressures and temperatures (25 bar, 1000–1500 K) in order to validate and refine the current chemical kinetic mechanisms.

While much work has been performed over the past decade to study NG/H₂ mixtures, and few recent studies have looked into NG/NH₃ blends, little have tested undiluted mixtures at the temperatures and pressures experienced in gas turbine combustors. Previous ignition delay time measurements of H₂ addition into methane (CH₄) at elevated pressures by Gersen et al. demonstrated that mixtures containing less than 20% H₂ had a modest effect on ignition whereas 50% addition dramatically decreased IDT [14]. More shock tube studies by Herzler and Naumann [15] and Huang et al. [16] both found that chemical kinetic model performance decreased with increasing addition of H₂. De Vries and Petersen [17] studied undiluted lean (Φ = 0.5) mixtures of pure NG/H₂ mixtures near the same conditions presented in this study and found weak ignition to be prevalent. Weak ignition, in comparison to strong ignition, is a common issue with high-fuel-containing mixtures in the shock tube. As noted by De Vries and Petersen [17], there must be a distinction made between these two ignition events to properly build a chemical kinetic model. Similar work done by Laich et al. [18] also showed nonidealities for high-fuel-loaded mixtures of methane at lower temperatures near 15 bar.

Shock tube studies of methane addition into ammonia near 40 bar by Shu et al. revealed that increasing methane concentrations increased reactivity but conversely increased both CO and NO emissions [19]. Work by Baker et al. [20] has also already been performed at lower pressures and higher temperatures using diluted mixtures as an initiative to develop a robust NG/H₂/NH₃ model entitled UCF 2022. Similarly, mechanism development and reduction by Li et al. specifically targeted CH₄/H₂/NH₃ mixtures however does not encompass alkanes larger than C₂ [21]. This paper seeks to extend the pressure range of the chemical kinetic models as well as validate the ignition properties for undiluted mixtures of H₂ and NH₃ addition to NG.

Mixtures were shock-heated and auto-ignited using a high-pressure shock tube at the University of Central Florida's (UCF) HiPER-STAR facility [22]. A shock tube allows for consistent replication of combustion studies through velocity control of the generated shock. The tube is separated into two sections. The high-pressure section (referred to as the driver) is filled with a nonreactive gas, in this case, a tailored mixture of ultrahigh purity (Nexair, 99.999%) helium (He) and ultrahigh purity (Nexair, 99.999%) nitrogen (N₂) until the scored aluminum diaphragm between the two sections is ruptured at the predetermined break pressure. The large pressure difference following the diaphragm rupture forms a shock wave, which travels downstream of the shock tube. It should be noted that the driver gas tailoring is not used to extend test time (as commonly referred to in shock tube literature) but instead to target specific experimental pressure and temperature. The low-pressure section of the shock tube is referred to as the driven section, which contains the mixture of study and is heated initially by the incident shock wave, and then once again after the shock wave reflects off the end wall of the shock tube. The secondary heating from the reflected shock waves brings the volume behind the shock to the desired temperature and pressure of the study. Temperature and pressure measurements were calculated from an in-house Python ideal and real gas one-dimensional frozen shock equation solver (PyRGFROSH) [23] using the shock velocity from the time intervals between five piezo-electric sidewall dynamic pressure transducers downstream of the shock tube (shown in Fig. 1). For this study, the ideal gas solver was used. These were then validated by the State 5 pressure recording.

Mixtures were made using the partial pressures (assuming Dalton's law) calculated from mole fractions of balanced chemical equations and then premixed in a magnetically stirred high-pressure mixing tank. As noted earlier, the mixtures were undiluted to replicate the chemistry during realistic gas turbine conditions, meaning the only gas constituents were fuel and synthetically made air. In all cases, the O₂:N₂ ratio was kept constant at 1:3.76. The NG mixture was a premade blend of 97.37614% CH₄, 2.25% C₂H₆, 0.077% C₃H₈, 0.00506% C₄H₁₀, 0.242% N₂, and 0.0498% CO₂, which was primary standard certified by Nexair. It is important to note that larger quantities of higher-order hydrocarbons can accelerate the ignition process and change reaction pathways therefore, lower concentrations of these were used in the study as an analog to industrial applications [24,25]. Ultrahigh purity (99.999%) O₂ and N₂ were also supplied by Nexair. A list of the mixture constituents is formatted below in Table 1. Hereafter, mixtures will be referred to by their designated names from the table. NH₃ is also known to adsorb into the stainless-steel shock tube [26], which could disrupt the purity of the mixture; however, when high molar concentrations of NH₃ are used, it has been experimentally determined by Pochet et al. that any adsorbed ppm quantities are not enough to have an effect on IDT [27]. Furthermore, the facility mixing tank is Teflon coated to further reduce this phenomenon.

All measurement devices are located at the test section of the shock tube, which is located 1 cm in front of the driven section endwall. Ports containing sapphire windows are used to allow optical access. Strong ignition IDTs (τ_ign) were determined by the max peak in OH* chemiluminescence, starting from the point of max pressure rise in PCB 5 from the reflected shock wave (defined as “time-zero”), as shown in Fig. 2. It should be noted that a secondary peak in the OH* trace is seen to be higher than the reported τ_ign in the case of Fig. 2, which was present in a few ignition traces. This peak is not considered to be the “max peak” due to it coalescing with a postdetonation oscillation in the pressure trace. In all experiments, the OH* detector is radially located 180 deg from PCB 5, hence why oscillations in both traces are generally in phase with each other. The postdetonation environments in undiluted mixtures are very dynamic and can create oscillations in the diagnostics, such as these.

The reasoning for calculating time-zero at the max pressure rise in PCB 5 from the reflected shock wave is because a bifurcated shock wave can be observed in the pressure signal due to measurements collected from sidewall-mounted transducers as opposed to endwall-mounted. Although the pressure nearest the wall is mainly the boundary layer, it is assumed the core flow of the reflected shock wave is at the State 5 conditions, which is validated once the signal equilibrates several microseconds afterward. This concept is fully explained in detail by Petersen and Hanson [28]. Nonhomogenous IDTs were not reported in this study due to uncertainty in the IDT determination, which was observed to be mixture independent in the lower temperature region, as discussed in Sec. 2.5.

As reported by Petersen [29], reporting IDTs from sidewall-mounted diagnostics for undiluted shock tube mixtures will make them slightly faster than endwall based diagnostics due to strong ignition events volumetrically igniting the entire test section region. In these cases, the discrepancy comes from the time-zero measurement since the reflected shock wave arrives at the sidewall several microseconds after it departs from the endwall. In dilute mixtures cases, this phenomenon is not seen since strong ignition events generally do not occur, hence, sidewall-mounted diagnostics are not artificially faster because the ignition originates from the endwall rather than volumetric test section ignition. This reporting difference is considered to be both mixture and facility dependent. As said by Shu et al. [30], pure ammonia is slow to react and therefore strong ignition events are not seen. In addition, some facilities collect sidewall IDTs at different axial locations, affecting the time reporting difference when volumetric test section ignition occurs. In the case of this study, the shock tube was not capable of collecting endwall IDTs and therefore the variable endwall location was set to be 1 cm from the sidewall diagnostics to collect as close to the endwall as possible and reduce the time difference between endwall reported IDTs for strong ignition events. The time-zero difference was extrapolated using the reflected shock wave velocity from PyRGFROSH and calculated for each experiment to be between 25 and 27 μs, artificially making IDTs faster by this amount if volumetric ignition occurs in the test section region. For IDTs reported above 170 μs, this difference is within the uncertainty range however for higher temperature points with IDTs below this value, the uncertainty is increased to 16–38%.

Primarily a hydrocarbon-based chemical kinetic model, the NUIGMech 1.1 [31] contains 2845 species and 11,260 reactions with alkanes up to C₇. The model encompasses all of the species used for the NG mixture in this study (up to C₄) while also containing an NO submechanism by Glarborg et al. [32] for NH₃ chemistry. Comparisons to the mechanism were made by applying the State 5 conditions to a constant volume 0D batch reactor and simulated using a Cantera [33] Python code using the same IDT determination method as in the experimental calculations.

A common gas dynamic phenomenon experienced in shock tubes is a boundary layer forming while the reflected shock passes through the State 2 conditions. The boundary layer causes the reflected shock to bifurcate and form two oblique shocks on the leading and trailing edges nearest the walls. Fundamental research by Mark [36] and Hollyer [37] found that the bifurcation is directly dependent on the specific heat ratio (γ) of the gaseous mixture. Monatomic gases, typically used in diluted shock tube mixtures, do not exhibit a boundary layer large enough to affect the shock wave because of their larger γ. However, diatomic and polyatomic mixtures, which have lower γ values, bifurcate the shock wave and create a turbulent boundary layer. Yamashita et al. [38] performed advanced shock tube computational fluid dynamics and showed the triple point at which the oblique shocks meet the normal shock to be hotter than the core flow. The hot triple point then caused “hot spots” in the turbulent boundary layer, which are small pockets of high-temperature differences, as much as 50 K more for an 850 K State 5 temperature. The bifurcation also grows as the shock travels further, increasing the size of the localized hot spot in the triple point. Due to the low activation energy of the high-fuel mixture, the hot spot could locally ignite that region before the core flow. This is a nonideal pre-ignition phenomenon that onsets the rest of the ignition, thus the mixture is not igniting at the theoretical postshock temperature (T₅).

Since the mixtures studied in this work are being oxidized in air and undiluted, (all the molecules are diatomic or polyatomic) the resulting reflected shock wave experiences bifurcation, making it difficult to homogeneously ignite in the low-temperature region. Several attempts to mitigate nonideal ignition were made by use of the CRV stage filling technique, developed in the literature [39], but were concluded to not be sufficient for these high-pressure shocks in the earlier work by Pierro et al. [40]. An example of a nonideal ignition trace is shown in Fig. 3, which very clearly exhibits pressure and emissions signal rise before strong ignition (circled in red). Determination of the IDT cannot be well defined due to strong ignition occurring roughly 325 μs after pre-ignition onset. This phenomenon was seen in the low-temperature regions for all mixtures except for the rich conditions.

The IDT measurements presented in Figs. 4–6 were taken near 25 bar between a temperature range of 1000–1500 K. All collected points are listed in Table 2. Baseline mixtures of pure NG were taken to compare the addition of H₂ and NH₃ at each equivalence ratio. The NUIGMech 1.1 mechanism was used for simulations of each mixture at 25 bar (solid line) to show the logarithmic ignition trend across the entire temperature range. The experimental data are solid-filled points on the plot with a calculated 15% uncertainty, which is calculated from the uncertainty in the shock velocity and further explained by Urso et al. [22]. Hollow points represent the NUIGMech 1.1 simulated at the exact experimental pressure as the accompanying solid-filled experimental point, as some data points may be slightly above or below 25 bar. The mixtures are named based on their assigned “Mix” number in Table 1.

Figures 4–6 are fuel-independent IDT plots of baseline NG, NG/H₂, and NG/NH₃, respectively, comparing the changing equivalence ratio effects on ignition. Stoichiometric conditions are circles, rich conditions are squares, and lean conditions are triangles, which are consistent for this set of plots.

Figure 4 shows the IDTs for pure NG mixtures at different equivalence ratios. In general, experimental IDTs tend to decrease with increasing temperature. At 1437 K, the IDT was found to be 156 μs, though a decrease in ∼120 K increases the IDT to ∼650 μs at 1317 K and Φ = 1. For Φ = 0.5, IDT is found to be 138 μs at 1435 K, while at 1304 K, it is 640 μs. Previous studies [41] indicate that dilute hydrocarbon mixtures ignite faster at fuel lean conditions (Φ = 0.5) while slowing down ignition at fuel rich conditions (Φ = 2.0). However, at undiluted conditions, only negligible difference is observed in IDTs for lean and stoichiometric conditions. This is primarily due to the higher concentration of fuel in stoichiometric cases than in lean cases. At high temperatures (∼1400 K), the fuel rich mixture (Φ = 2.0) ignites faster than its Φ = 1 and Φ = 0.5 counterparts and shows similar IDTs (598 μs at 1317 K) at lower temperatures as evident from Fig. 4. Similar observations are made with NG/H₂ mixtures (Fig. 5), where lean case shows longest IDTs followed by stoichiometric and rich cases. Negative temperature coefficient (NTC) behavior was not seen in any case, unlike the results found by De Vries and Petersen [17] however, their study was in the lower temperature region where NTC is more commonly observed with CH₄ [42]. As mentioned in Sec. 2.2, concentrations of higher-order hydrocarbons were exceptionally low in the NG mixture and therefore the pure NG mix was expected to behave close to pure methane. This conclusion for NTC behavior is similarly drawn since NTC of large alkanes occurs near the same temperature region of CH₄.

Model agreement is excellent for stoichiometric conditions (apart from Mix 1), supporting that undiluted mixtures do not have a significant effect on ignition chemistry. De Vries and Petersen [17] similarly showed that ignition with methane does not change if an Ar bath gas is replaced with N₂, however, NO pathways for air-breathing engines using NH₃ can be affected if excess N₂ is present in the combustion, and ultimately affect emissions and ignition. For these cases, the N₂ chemistry from the NO submechanism predicts very accurate IDTs in all mixture conditions. Baseline NG ignition did not change with changing equivalence ratios (Fig. 4), whereas NG/H₂ mixtures auto-ignited faster in rich conditions and slowed in the lean conditions, which the model accurately predicts (Fig. 5). NG/NH₃ experimental data show that changing the equivalence ratio did not affect ignition, similar to pure NG. However, the model predicted lean conditions to be quicker and rich conditions to ignite slower (Fig. 6).

When increasing or decreasing the equivalence ratio, pure NG IDTs became faster than the model predictions at all temperatures while NG/H₂ were also slightly faster than the model at lower points in the temperature range. Additionally, NG/H₂ ignition shows a linear trend with changing temperature for stoichiometric and rich cases but trends nonlinearly for lean conditions, which the model fails to capture.

For this section, baseline NG mixtures are circles, mixtures with NH₃ addition are traingles, and H₂-containing mixtures are squares. Figure 7 shows stoichiometric conditions, Fig. 8 compares lean conditions, and Fig. 9 compares rich conditions.

Similar to the analysis made by Baker et al. [20], H₂ addition very much increases the reactivity of the mixture, thus experiencing much quicker IDTs. Hydrogen addition decreases the IDT by approximately 90% for stoichiometric conditions (Fig. 7), 86% for lean conditions (Fig. 8), and 89% for rich conditions (Fig. 9). Conversely, NH₃ addition shows insignificant effect on the ignition when compared to the baseline. These observations can be attributed to dominating H₂ chemistry for NG/H₂ mixtures whereas NG/NH₃ mixtures are dominated by hydrocarbon chemistry. Further analysis of the chemistry between these reactions is found in Sec. 3.3. Baker et al. [20] also noted that NH₃ greatly decreased the reactivity of the mixture in their study when compared to baseline NG. However, the NG mixture used contained larger amounts of higher order hydrocarbons, whereas this study only contains ppm quantities (>770 ppm C₃H₈ and >51 ppm C₄H₁₀), hence supporting that the higher order alkanes play little role in the ignition behavior when in small quantities. Baker et al. [20] similarly suggested that the larger hydrocarbons initially formed OH radicals faster than the rest of the reactants.

Sensitivity analyses between mixtures are shown in Figs. 10–13 to find shared reactions that are most prominent to ignition. As noted in Sec. 2.4, each sensitivity analysis was performed using OH formation rather than a fuel decomposition due to the blending from multiple fuels. Although this does not affect the sensitivity analysis, it does change the perspective of the sensitivity coefficients. Therefore, positive sensitivity coefficients are indicative of ignition-promoting reactions, while negative sensitivity coefficients are ignition-inhibiting reactions.

which is an elementary reaction in the hydrocarbon ignition process. (R10) is also very sensitive to NG/H₂ ignition because of the H-H bond separation in H₂. This reaction is then expedited due to two readily available H atoms rather than the one from C-H. The early onset of (R10) consumes available O₂ and inhibits methyl oxidation, leading to hydrocarbon pyrolysis rather than oxidation. With no free O₂ molecules left over H atoms bond together, leaving substantial amounts of H₂ in the products of the combustion. This pattern was common for all NG/H₂ mixtures, especially for fuel rich conditions.

Unlike the H₂ addition, NH₃ does not heavily affect the sensitive reactions to OH. Many of the commonly shared reactions of pure NG and NG/NH₃ in Fig. 13 have similar effects on ignition, indicating that hydrocarbon chemistry is dominant. (R10), (R114), and (R122) are still the most prominent reactions, however mainly from C-H breakdown in these cases. This similarly causes pyrolyzation of nitrogen chemistry and creates free H atoms from N-H separation, thus still leaving ample amounts of H₂ in the combustion products.

is initiated and promotes OH radical production. Figure 12 shows (R10517) to be an ignition promoting reaction for stoichiometric mixtures of NG/NH₃. Although this reaction sequence is also be driven by fuel NO from ammonia, energy is first required to break N-H bonds to get free NH molecules whereas an N-N bond separation readily frees two N atoms from N₂ in air. In the case of these mixtures, NH molecules from pyrolyzed NH₃ and separated N-N bonds are both contributing to NO in (R10467).

is shared for all baseline NG mixtures at high and low temperatures in the sensitivity analyses and would also equally shift each prediction closer toward the data since the normalized sensitivity coefficients are similar in magnitude at both 1250 and 1400 K. Upon investigation of (R118), the listed uncertainty in k is more than ±30% from work done by Srinivasan et al. [43], giving a large range of adjustability. (R118) is also only commonly shared among the pure NG mixture, which would not significantly affect the agreement of the model with mixtures containing H₂ and NH₃ addition. A brief investigation into solely decreasing the A-factor of (R118) by 30% decreased the IDTs by as much as 8%, depending on the mixture and temperature. More experimental data at alternative pressures and temperatures from proceeding studies and thorough chemical analyses are required to compare the model performance before any modifications can be made to the chemical kinetic mechanism.

may also bring the model closer in agreement and has been calculated with a ±20% in the work of Blitz et al. [44]. However, this reaction is not advised to be modified because it is commonly shared in the NG/H₂ and NG/NH₃ mixtures, which are already mostly in agreement with the model.

In this work, hydrogen and ammonia addition to natural gas mixtures ignition were probed under gas turbine operating conditions. Equal amounts of hydrogen and ammonia were added to undiluted, premixed mixtures of natural gas/air and combusted at the conditions relevant to current gas turbine operating conditions (1000–1500 K, 25 bar). The ignition delay times are reported here within and compared to the NUIGMech 1.1 for undiluted, high-fuel validation. Similar studies of undiluted NG mixtures have yet to be performed, but comparisons are made to fundamental work by De Vries and Petersen [17] as well as Baker et al. [20], who have performed relevant work. The work performed in this study is an initial step toward lowering carbon emissions for current NG turbines. The main conclusions of the study are as follows:

1. The NUIGMech 1.1 model shows good agreement for stoichiometric conditions but deviates during lean and rich conditions. Previously, undiluted chemistry effects on ignition at higher pressures were not explored but are now validated for chemical kinetic modeling. Pure NG measurements were moderately faster than the predictions, making a further analysis into relating reaction coefficients a necessary step to improve the model.

2. 50% H₂ fuel fraction greatly decreases IDT while NH₃ shows to have no effect at the tested conditions. An analysis of the reactions in the ignition process shows dominating hydrogen chemistry for mixtures containing H₂ and dominating hydrocarbon chemistry for mixtures containing NH₃. While validation of emissions formation of CO and NO_x is still imperative, from an ignition standpoint, NH₃ addition looks like a promising and favorable modification to current NG turbines without needing a major infrastructure change.

3. Sensitivity analyses show significant differences in the sensitive coefficient between stoichiometric pure NG and NG/H₂. Stoichiometric NG/NH₃ share similar sensitivity coefficients with pure NG, further supporting that hydrocarbon chemistry dominates ignition rather than inhibiting nitrogen chemistry. The reaction CH₃ + O₂ = CH₃O + O is suggested as a modifiable parameter to improve the model performance of pure NG, although not necessary for predicting H₂ and NH₃ ignition.

The authors thank the Department of Energy (DE-FE0032072), the NASA Florida Space Grant Consortium, and UCF for supporting this work. Feedback from Mitsubishi Power and General Electric is acknowledged.

• U.S. Department of Energy (DE-FE0032072) Funder ID: 10.13039/100000015.

The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.

This report was prepared as an account of work sponsored by an agency of the U.S. Government. Neither the U.S. Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the U.S. Government or any agency thereof. The views and opinions of the authors expressed herein do not necessarily state or reflect those of the U.S. Government or any agency thereof.