On Flame Morphologies and Stabilities of NH₃/Air Premixed Flames in a Dual-Swirl Gas Turbine Model Combustor: A Comparative Study of NH₃ and CH₄ as Pilot Fuel

Abstract

Substituting hydrocarbon fuels such as methane (CH₄) with ammonia (NH₃) reduces CO₂ emissions in gas turbines, but ammonia’s low reactivity challenges flame stability. Dual-swirl staged combustors using a low thermal power (P_pilot) pilot flame can stabilise the main flame. This work compares the morphologies and stabilities of NH₃/air premixed swirl flames using ammonia and methane pilot flames (APF and MPF). Flame imaging and simultaneous OH-NH planar laser-induced fluorescence (PLIF) are employed to analyse flame morphology. Main flame stability is assessed by measuring the lean blow-off equivalence ratio (ϕ_b,main). The results show that MPF significantly outperforms APF in main flame stabilisation. At P_pilot = 1.2–1.8 kW (14.2–21.3% of P_main), the dual-swirl flames exhibit a stratified structure, with OH concentrated in the pilot stage. Flames with MPF exhibit considerably lower ϕ_b,main than those with APF. For example, at P_pilot = 1.6 kW, ϕ_b,main is reduced to 0.42 with MPF, compared to 0.56 with APF, demonstrating MPF’s superior stabilisation capability. MPF can reduce CO₂ emissions by 82.4–87.6% compared to a CH₄ flame of equivalent thermal power. Two stabilisation modes are identified, namely primary recirculation zone-dominated and pilot-dominated modes. These findings demonstrate that a low-power MPF provides an effective strategy for enhancing ammonia flame stability and reducing CO₂ emissions in gas turbines.

1. Introduction

Gas turbines play a vital role in the energy and power sector, and have been widely used in electricity generation, oil and gas exploration, distributed energy systems, long-distance pipeline compression and marine propulsion. However, conventional gas turbines rely on natural gas (a major component is CH₄) and inevitably generate large amounts of CO₂ emissions. Driven by the global transition toward green development and increasing emphasis on CO₂ reduction, current research has been focusing on improving efficiency, reducing emissions and enhancing the fuel flexibility of gas turbines, especially through utilising carbon-free fuels, such as hydrogen (H₂) and ammonia (NH₃) [1,2,3]. Ammonia is recognised as a promising carbon-zero fuel for decarbonisation. As a raw material for agricultural fertilisers and explosives, ammonia has readily available production, storage and transport networks. Additionally, ammonia possesses a 17.6% mass fraction of hydrogen, and its combustion generates no carbon emissions. Therefore, ammonia is not only an excellent hydrogen carrier but also a promising zero-carbon fuel. Ammonia has been proposed as a potential fuel for decarbonisation in energy and transport sectors, such as gas turbines [4], marine engines [5], furnaces [6] and boilers [7]. However, the application of ammonia is constrained by its highly corrosive and toxic features. Ammonia leakage should be carefully prevented during its storage and transport to avoid health and environmental risks [1]. Most importantly, burning ammonia in gas turbines also faces dual constraints from its low combustion reactivity and easy formation of NOx pollution.

To improve ammonia combustion reactivity, reactive fuel co-firing is the most straightforward and effective strategy in swirl flames. Previous studies have widely adopted swirl model combustors to investigate ammonia flames co-fired with reactive fuels, such as hydrogen [8], methane [9,10], and DME [11]. Wei et al. [8] found that H₂ addition enables flame stabilisation in higher positive axial velocity regions and enhances H, NH, and NH₂ species at the flame root. In the work of Ji et al. [10], 50% CH₄ addition improves the blow-off limit of NH₃ by 16.6% through increasing the laminar flame speed and extinction strain rate. DME was also found to be effective in improving NH₃ flame stability [11]. However, previous studies have widely adopted single-swirl combustors. Swirl flow stabilisation is governed by vortex breakdown that creates a central recirculation zone, which acts as a continuous ignition source. Single-swirl combustors have only one nozzle; thus, the co-fired fuels are fully premixed. In this context, the reactive fuel is transported downstream and away from the combustor base. Although the combustion in the main reaction zone is enhanced [12], the combustion enhancement at the flame root remains limited. Although effective for flame stabilisation, this approach typically requires a large fraction of reactive fuel in the fuel mixture to achieve effective combustion reactivity enhancement. Okafor et al. [13] tested NH₃ and CH₄ co-firing in a micro gas turbine combustor, whilst the NH₃ accounted for a maximum heat fraction of 30% in the fuel mixture. This leads to a CO₂ emission reduction of 30%. In the work of Ji et al. [10], a 50% mole fraction of CH₄ is required in the NH₃/CH₄ fuel mixture to lower the ammonia flame lean blow-off (LBO) limits from 0.78 to 0.65 at a constant inlet bulk velocity of 4 m/s. Considering that the lower heat value (LHV) of methane (35.8 MJ/m³) is much higher than that of ammonia (14.14 MJ/m³) [14], the combustor is mainly powered by methane when it burns NH₃/CH₄ with 50% mole fraction of methane. Considering an identical thermal power as a pure CH₄ combustion, this only reduces CO₂ emissions by 28.3%. Therefore, to develop an ammonia-powered gas turbine combustor and reduce carbon emissions, further research is necessary to utilise the reactive fuel co-firing combustion enhancement strategy more effectively.

Centrally staged combustion is a key technology in aeroengine applications. Such combustors adopt a dual swirler or multiple swirler configuration to improve flame stability [15]. They offer independent control of multiple nozzles and can enhance spray combustion by improving droplet residence time, evaporation and mixing [16]. In the GE TAPS combustor designed for NOx reduction [17], a central pilot flame supports the main flame so that it can operate under very lean conditions. This concept is a promising way to improve gas fuel flame stability for gas turbine applications and has recently been adopted to enhance ammonia combustion. Some studies have tested co-firing ammonia with more reactive fuels in dual-swirl combustors. Elbaz et al. [18,19] studied NH₃/CH₄/air flames, where ammonia and methane were separately injected into the inner and outer nozzles of the combustor. Stable combustion can be achieved by burning CH₄ in the outer stage at the equivalence ratio (ϕ) of 0.6 and NH₃ in the inner stage. Such configurations can lower the NOx emissions. However, as the outer stage features a larger cross-sectional area than the inner one and the flame speed of CH₄ is high, a larger air and fuel flow rate is required at the outer stage to prevent flashback. This configuration inevitably requires a large amount of reactive fuel to enhance ammonia combustion. In the work of Li et al. [20], the primary stream, which consists of air, CH₄, and NH₃, is directly injected into the combustor, whilst the secondary stream consists of either swirling or straightly injected premixed CH₄/air. The NH₃ ratio in this work is also very low, accounting for less than 20% of the overall thermal power.

To conclude, current studies on ammonia/reactive fuel co-firing based on either single-swirl or dual-swirl combustors typically require a large amount of reactive fuel to effectively enhance ammonia combustion. Under such conditions, reactive fuels are not adopted as a pilot fuel; instead, the combustor in fact operates with the reactive fuels as they contribute most of the thermal power. To address the above-mentioned limitations, the present study designed a novel configuration to enhance the main ammonia flame with a low-power pilot flame. It aims to comparatively investigate the effects of low-power NH₃ and CH₄ pilot flames on enhancing ammonia/air premixed flames in a gas turbine model combustor, with special focus on flame morphologies and stability limits. Methane is selected as the representative reactive pilot fuel because it is the primary component of natural gas and is widely used in gas turbine combustors, allowing a well-defined baseline for evaluating the reduction in carbon emissions and the improvement in flame stability. To exclude the influence of extra thermal power brought by the pilot flame, ammonia pilot flames of the same thermal power are also tested to serve as a comparison. Planar laser-induced fluorescence (PLIF) is applied to simultaneously capture the spatial distribution of NH and OH radicals. Flame morphologies under stable and approaching blow-off conditions are analysed.

2. Methodologies

2.1. Experimental Setup and Operating Conditions

Figure 1 shows the schematic of the dual-swirl gas turbine model combustor, along with key dimensions. The model combustor features two coaxial counter-rotating swirlers, the pilot-stage one with a vane angle of 45° and the main-stage one with a vane angle of 30°. Swirl numbers (Sn) are calculated based on vane angle and the inner and outer diameter of the swirlers (Sn = $\frac{2}{3}[\frac{1-{(D_{\mathrm{i}}/D_{\mathrm{o}})}^3}{1-{(D_{\mathrm{i}}/D_{\mathrm{o}})}^2}]\mathrm{t}\mathrm{a}\mathrm{n}\alpha$) [21]. The swirl numbers of the pilot and main stages are 0.9 and 0.53, respectively. The dual-swirl nozzle configuration features two stages. The inlet diameters of the two stages are shown in Figure 1. The premixed pilot fuel/air is injected solely into the pilot stage to establish a pilot flame, whilst ammonia is burned in the main stage. The confinement ratio of the main stage is ~2. The pilot flame operates in low-power mode but serves as an ignition source and thermal stabiliser for the main flame. The main flame is a premixed NH₃/air flame, whilst the pilot flame is either a premixed NH₃/air flame (APF) or a premixed CH₄/air flame (MPF). Fuel and air are premixed in a mixing chamber before entering the combustor. The air flow rates in the two stages are independently controlled by two mass flow controllers (ALICAT MC series, ±1% accuracy), while the fuel flow rates are controlled by two other mass flow controllers (SevenStar D07-9E, ±2% accuracy). A quartz tube is used as the combustor wall for optical measurement.

Figure 1. Schematic of the gas turbine model combustor.

Figure 2 presents the flame diagnostic system. A Canon camera (EOS 80D) records the time-averaged flame images. The ISO, aperture, and exposure time are respectively 200, f/11, and 0.5 s. The laser system repetition rate is 10 Hz. The Nd:YAG laser sources (Q-smart 850, Quantel and Nimma 600, Beamtech) generate the raw laser beams with a wavelength of 532 nm. Then, the frequency is doubled with tuneable dye lasers (Cobra-Stretch, Sirah Lasertechnik) and the wavelength of the laser is transferred by each dye laser, respectively, to 282.9625 nm and 305.2798 nm. The average pulse energy of the 282.9625 nm laser is 13 mJ. It excites the Q1(8) line of A²Σ⁺–X²Π (0-0) transition of OH radical [22]. Meanwhile, the 305.2798 nm laser has an average pulse energy of 7.5 mJ. It activates the single-photon excitation of NH A³Π–X³Σ (1, 0) transition [23]. The laser beam is reshaped to obtain a laser sheet (120 mm in height and 0.5 mm in thickness) by a cylindrical concave plane lens and a spherical convex plane lens. The planar laser-induced fluorescence (PLIF) signals of OH and NH are simultaneously acquired with two intensified sCMOS cameras (pco.dicam C1), which provide a spatial resolution of ~11 pixel/mm. For OH measurement, a 310 nm bandpass filter (Edmund Optics, Barrington, NJ, USA, 310 ± 5 nm) is equipped. For NH measurement, a 340 nm bandpass filter (Edmund Optics, 340 ± 26 nm) is adopted. The PLIF images are calibrated with a calibration board (204-15, LaVision, Göttingen, Germany) to eliminate the distortion and justify the dimensions. The key experimental instruments in this work are listed in Table 1.

Figure 2. Schematic of the OH and NH-PLIF diagnostic system.

Table 1. Experimental instruments and key parameters.

Instrument	Model	Company	Accuracy	Key Parameter
Mass flow controller	ALICAT MC series	ALICAT Scientific, Tucson, AZ, USA	±1%
Mass flow controller	SevenStar D07-9E	SevenStar, Beijing, China	±2%
Camera	pco.dicam C1	Excelitas, Pittsburgh, PA, USA	/	exposure time: 250 ns, resolution: 2048 × 2048
Camera	EOS 80D	Canon, Tokyo, Japan	/	ISO200, f/11, 0.5 s; resolution: 6000 × 4000
Nd:YAG laser	Nimma 600	Beamtech, Shanghai, China	/	repetition rate: 10 Hz; wavelength: 532 nm
Nd:YAG laser	Q-smart 850	Quantel, Cournon d’Auvergne Cedex, France	/	repetition rate: 10 Hz; wavelength: 532 nm
Dye laser	Cobra-Stretch	Sirah Lasertechnik, Grevenbroich-Kapellen, Germany	±10−4 nm	wavelength: 305.2798 nm; pulse energy: 7.5 mJ
Dye laser	Cobra-Stretch	Sirah Lasertechnik, Germany	±10−4 nm	wavelength: 282.9625 nm; pulse energy: 13 mJ

The time phases of the lasers and cameras are synchronised by an analogue output device (NI PCIe-6738, Ha Noi City, Vietnam). It has a maximum update rate of 1 MS/s and produces a pulse signal with a time resolution of 1 ns. The signal output of the device is controlled by a LabVIEW programme, which can set the time delay between channels.

Experiments are conducted under atmospheric conditions. Table 2 lists the detailed experimental conditions. Across all tested conditions, the bulk velocity at the exit of the main-stage swirler is 2.3 m/s, and that at the exit of the pilot-stage swirler is 2.1 m/s. This corresponds to a pilot-stage Reynolds number of 1000 and a main-stage Reynolds number of 2800. Based on the empirical equation, TI = 0.153Re^−0.0779, proposed by Russo et al. [24]. The turbulence intensities of the pilot stage and main stages are 8.9% and 8.2%, respectively. For stable flames, the effect of the pilot flame on main flame stability is examined by progressively increasing the thermal power of the pilot flame (P_pilot) whilst maintaining the thermal power of the main flame (P_main). Thus, the equivalence ratio of the pilot flame (ϕ_pilot) varies while that of the main stage (ϕ_main) is fixed at 1.0. For flame stability measurement, the ϕ_main is decreased while keeping ϕ_pilot constant. The ϕ_main at which a complete flame extinction occurs is defined as the main flame stability limit (ϕ_b,main). It is measured by decreasing the outer stage fuel flow rate whilst keeping the inner stage unchanged. The outer stage fuel flow rate corresponding to complete extinction of flames in the combustor is used to calculate ϕ_b,main. The ϕ_b,main is repeated three times to ensure repeatability. The structures of the flames approaching ϕ_b,main (ϕ_main = ϕ_b,main + 0.02) are measured via PLIF.

Table 2. Experimental conditions.

Pmain (kW) (±2%)	ϕmain (±2.2%)	Ppilot (kW) (±2%)	ϕpilot (±2.2%)	Pilot Fuel	Measurement Objects
					Flame Image	PLIF	ϕb,main
8.45	1.0, ϕb,main + 0.02	0.8–2.0 (Δ = 0.4)	0.40–1.00	CH4	●	●
			0.38–0.95	NH3	●	●
/	/	0.8–2.0 (Δ = 0.1)	/	CH4	●		●
				NH3	●		●

2.2. Data Postprocessing and Analysis Methods

The captured PLIF images are calibrated with the DaVis 8.0 program (LaVision). The NH-PLIF image resolution is 11.7015 pixel/mm, while the OH-PLIF image resolution is 10.8952 pixel/mm. The calibrated .im7 files are imported to MATLAB R2024b software to process simultaneously acquired OH-PLIF and NH-PLIF images to generate composite visualisations of radical distributions in the tested flames. For each pair of images, the NH-PLIF frame is first scaled to match the resolution of the OH-PLIF image using the ratio of the calibrated resolutions. Both images are then median-filtered to reduce noise and cropped to a common region of interest (0–60 mm in x, 0–120 mm in y), where x is the radial position of the combustor and y is the axial position, as illustrated in Figure 1. The OH signal is assigned to the red channel and the NH signal to the blue channel of an RGB image; each channel is normalised by its maximum intensity, and low-intensity pixels (<0.01 after normalisation) are set to zero to suppress background noise. A total of 300 PLIF images of OH and NH are used to obtain the time-averaged distribution. The resulting overlay is displayed and saved as a PNG file for analysis of the flame structure and the spatial correlation between OH and NH radicals.

Under the assumption of rotational symmetry for typical swirl flames, the total OH/NH signal intensity (I_OH/I_NH) within the ROI is obtained by integrating the intensity over the radial direction with a weighting factor of $2\pi x$ where x is the radial coordinate (Equation (1)). The OH/NH density (ρ_OH/ρ_NH) is then defined as the ratio of the total integrated intensity to the volume of the flame region, where the volume is calculated from the binarized flame mask (threshold = 0.1).

$$I={\displaystyle \frac{{\int }_o^{x_{max}}{\int }_0^{y_{max}}i(x,y)dy·2\pi xdx}{{y_{max}\pi x_{max}}^2}}$$

(1)

NH-PLIF visualises the flame front of NH₃ flames [25], and the flame surface density (FSD) of the NH₃ main flame is calculated based on two-dimensional NH-PLIF results. The FSD is defined as [26]

$$\mathsf{\Sigma }=\underset{\mathsf{\Delta }x\to 0}{\lim }{\displaystyle \frac{{\overline{A}}_f}{{\mathsf{\Delta }x}^3}}$$

(2)

where ${\overline{A}}_f$ is the time-averaged surface area of flamelets within a cubic interrogation box of size $\mathsf{\Delta }x$. In this work, only two-dimensional information is available from NH-PLIF results. Thus, an approximation is made [27]:

$$\mathsf{\Sigma }=\underset{\mathsf{\Delta }x\to 0}{\lim }{\displaystyle \frac{{\overline{A}}_f}{{\mathsf{\Delta }x}^3}}\cong \underset{\mathsf{\Delta }x\to 0}{\lim }{\displaystyle \frac{{\overline{L}}_f}{{\mathsf{\Delta }x}^2}}$$

(3)

where ${\overline{L}}_f$ is the time-averaged length of flamelets within the control area of size $\mathsf{\Delta }x$.

For each image, the flame front is first identified using a custom function. In this function, each image is first median-filtered to reduce noise, converted to greyscale, and low-intensity background pixels (<0.2) are set to zero. Otsu’s method [28] is then applied to determine a binarization threshold, and the resulting binary image is cleaned by removing objects smaller than a specified number of pixels. For each row (i.e., each y-position), the code locates the transitions from unburned (0) to burnt (1) and vice versa, thereby identifying individual flame segments. For each such segment, the column index at the midpoint of the segment is recorded, and a binary matrix “flame_front” is generated with ones at these midline positions. This yields a thin, continuous representation of the flame front.

After extracting the flame front for all images, a 7 × 7 pixel sliding window is then applied to count the number of flame front pixels within each local neighbourhood. This count is then divided by the physical area of the window (derived from the imaging resolution of 11.7015 pixels/mm) to obtain the local FSD. The procedure is parallelised using a “parfor” loop for efficient batch processing. After processing all images, the ensemble-averaged FSD field is saved in a Tecplot-compatible data file, containing the spatial coordinates (x, y) and the corresponding FSD values. The FSD contours are generated with Tecplot 9.0 software.

3. Results and Discussions

3.1. Stable Flame Morphologies

The pilot fuel type and thermal power have a large impact on the flame macrostructure of stable ammonia swirl flames. Figure 3 presents the time-averaged flame images of stoichiometric ammonia swirl flames with either APF or MPF. When NH₃ is used as the pilot fuel, the flame chemiluminescence increases with increasing P_pilot. This can be attributed to elevated total thermal power and increased ammonia content. The former brings more heat to the combustor while maintaining a constant air flow rate, and the latter produces more NH₂ radicals. Both factors enhance NH₂* chemiluminescence from the A²A₁ state of NH₂ [29,30], resulting in a brighter ammonia flame. Despite the enhanced chemiluminescence, the flame macrostructure remains almost unchanged. Across all P_pilot values, the flame anchors exclusively at the pilot-stage exit, which can be attributed to the insufficient combustion reactivity of the APF and the dominance of the primary recirculation zone in flame stabilisation.

Figure 3. Time-averaged flame image of stoichiometric ammonia swirl flames with (a) APF and (b) MPF at varying P_pilot (The red arrows indicate the pilot-stage mixture whilst the blue arrows indicate the main-stage mixture).

In contrast, ammonia main flames with MPF exhibit distinct flame chemiluminescence and macrostructures as P_pilot varies. At P_pilot = 0.8–1.1 kW, the flames appear to detach from the combustor base. However, a methane flame does exist in the dark area near the combustor base, although its CH* and C₂* chemiluminescence [29] are too weak to be detected due to the low methane content. OH-PLIF results confirm the presence of this flame, as discussed later. Increasing P_pilot to 1.2 kW results in a stratified flame structure comprising the methane pilot flame and the ammonia main flame. At P_pilot = 1.2–2.0 kW, the chemiluminescence of the methane pilot flame grows progressively. In contrast to the blurred flame edges observed under other conditions, flames under these conditions exhibit sharp boundaries, indicating a stable main ammonia flame. Owing to the significantly higher combustion reactivity of methane than ammonia [31], methane/air mixtures can sustain an independent pilot flame at the pilot-stage exit. Despite operating at low power, this pilot flame acts as an ignition source and thermal stabiliser for the main flame, enabling it to anchor at the main-stage exit and maintain a stable structure.

Detailed analysis of OH and NH distributions provides further insight into how the pilot flame influences the flame macrostructure. Figure 4 presents the normalised PLIF images of selected flames in Figure 3. Since the time-averaged PLIF images are symmetric, only the right halves of the images are shown for clarity. Figure 5 shows the total OH/NH PLIF signal intensities (denoted I_OH and I_NH) and their corresponding signal densities (ρ_OH and ρ_NH).

Figure 4. Normalised time-averaged NH/OH distribution of stoichiometric ammonia swirl flames with (a) APF and (b) MPF at varying P_pilot, along with (c) normalised radial profile of NH distribution at 60 mm above the combustor base in ammonia swirl flames with APF. Each image is normalised with its own maximum NH/OH signal intensity and thus only reflects the distribution of species. Each line in (c) is normalised with the radial maximum value.

Figure 5. Total signal intensity of (a) OH and (c) NH at various P_pilot, together with their corresponding (b) OH and (d) NH signal density.

The normalised distribution of OH/NH radicals within ammonia swirl flames with APF and MPF are compared in Figure 4. For ammonia swirl flames with an APF, increasing P_pilot results in a more distributed OH radical field located farther from the central axis (Figure 4a). Moreover, at higher P_pilot, OH radicals appear at larger radial positions (Figure 4c), indicating a spatial shift in the main reaction zone. Whilst for ammonia swirl flames with an MPF, the variation in P_pilot leads to a distinct distribution of OH and NH. At P_pilot = 0.8 kW, the OH/NH distribution resembles that observed with the NH₃ pilot flame (Figure 4). The OH signal can be observed near the combustor base, confirming the presence of a methane flame in this region. In addition, the total signal intensities and the signal densities of both radicals are comparable between the two pilot flame types. However, at P_pilot = 1.2–2.0 kW, the flame transitions to a stratified flame structure, with OH concentrated in the pilot stage whilst NH in the main stage.

The OH/NH signal intensity values are quantitatively compared in Figure 5. In ammonia swirl flames with an APF, as shown in Figure 5a,b, both the total intensity and signal density of OH decrease with increasing P_pilot. This trend is attributed to the variation in total equivalence ratio (ϕ_total). As P_pilot rises from 0.8 kW to 2.0 kW, ϕ_total rises from 0.88 to 0.99 as more NH₃ is introduced to the combustor at a constant air flow rate. In typical swirl flames, reactants are primarily consumed in the shear layer [32] before being recirculated into the primary and corner recirculation zones. Within the equivalence ratio (ϕ) range of 0.8–1.1, the laminar burning velocity of NH₃/air mixtures increases with ϕ [33], thereby enhancing the flame. It should be noted that increasing the ϕ to a higher value can cause a quenching effect and excess NH₃ emissions. The decrease in I_OH with increasing ϕ_total is consistent with observations by Zhang et al. [34] and Wei et al. [8] for ammonia swirl flames across ϕ = 0.8–1.0. In contrast to the intense variation in OH fluorescence, NH fluorescence exhibits only minor changes with P_pilot. As P_pilot increases, the total NH signal intensity rises slightly (Figure 5c) owing to higher NH₃ content. However, the signal density and spatial distribution of NH remain largely unchanged.

In ammonia swirl flames with an MPF, although the total OH signal intensity varies with P_pilot due to increased ϕ_total at higher P_pilot, the OH signal density is significantly improved (Figure 5b), indicating a strong and compact reaction zone within the pilot stage. This establishes the MPF as an effective ignition source, providing a high-temperature environment and abundant reactive radicals, both of which significantly enhance the ignition of the main ammonia flame. As shown in Figure 5c,d, although the total NH intensity of flames with an MPF is lower than that with an APF due to the reduced ammonia content available for NH production, the NH signal density is higher. This indicates a more compact and intense reaction zone in the ammonia flame when stabilised by an MPF.

3.2. Main Flame Stability

The variation in the type and power of the pilot flame leads to distinct stability limits of the flame, ϕ_b,main. As illustrated in Figure 6, with the increase in P_pilot, the flames with either CH₄ or NH₃ pilot flames exhibit higher stability compared to the single-stage ammonia main flame without pilot flames. Across all P_pilot values, flames stabilised with an MPF have a lower ϕ_b,main than those with an APF. For the APF, ϕ_b,main decreases almost linearly with P_pilot, demonstrating progressively enhanced flame stability. In contrast, the CH₄-piloted flame displays a three-stage response. ϕ_b,main decreases slowly with P_pilot at low P_pilot values (the blue region in Figure 6), followed by a rapid decrease at intermediate P_pilot (the pink region in Figure 6). It should be noted that for P_pilot = 1.2–1.8 kW, P_pilot is only 14.2–21.3% of P_main; such a low-power MPF can bring significantly improved flame stability to the main flame. At sufficiently high P_pilot, ϕ_b,main of the CH₄-piloted flame reaches zero (the yellow region in Figure 6), indicating that the methane pilot flame can sustain stable combustion independently in the pilot stage without support from the main flame. By definition, ϕ_b,main refers to the main flame only, and it does not represent total system stability. However, the reactivity of ammonia is much lower than that of methane, so that the ammonia pilot flame cannot burn independently in the range of the current study.

Figure 6. Measured ϕ_b,main of ammonia swirl flames with NH₃ (red rhombuses) and CH₄ (blue dots) pilot flames at different P_pilot. The dash-dot line marks the lean blow-off limit of the NH₃ flame at the same total air flow rate but without pilot flames.

Corresponding to the variation in ϕ_b,main with P_pilot, flames approaching blow-off exhibit distinct behaviours. Figure 7 presents the time-averaged images of ammonia swirl flames approaching blow-off. For flames with an APF, and for those with an MPF at low P_pilot (<1.2 kW), the behaviour resembles that of atmospheric swirl flames approaching lean blow-off as reported by Skiba et al. [35]. In these cases, the ammonia flame mainly resides within the primary recirculation zone, which stabilises the flame by recirculating hot products back to the reactant mixture inlet [32]. For flames with an MPF at intermediate P_pilot (1.2–1.8 kW), a stratified structure develops, comprising a methane pilot flame and a main ammonia flame. The pilot flame supplies reactive radicals and a high-temperature environment to the main flame. Under these conditions, the main ammonia flame is stabilised predominantly by the pilot flame rather than by the primary recirculation zone. However, because the pilot flame is not yet strong enough to sustain itself independently in the pilot stage, blow-off still occurs when ϕ_main is reduced to a critical threshold. At higher P_pilot (>1.7 kW), the equivalence ratio of the MPF exceeds 0.85, enabling the pilot flame to burn independently in the pilot stage. Therefore, even when the ϕ_main is reduced to zero, the flame does not experience blow-off.

Figure 7. Time-averaged flame images of ammonia swirl flames approaching blow-off (ϕ = ϕ_b,main + 0.02) with (a) APF and (b) MPF at varying P_pilot.

3.3. Enhancement of Pilot Flame on Main Flame Stability

To understand the mechanism behind the enhanced main flame stability, this work performs a detailed analysis of the stable flame and the flames approaching blow-off. The flame surface density (FSD) within the stable flames is calculated to quantitatively compare the combustion enhancement effects of APF and MPF. Moreover, the instantaneous distribution of OH/NH radicals within flames approaching blow-off is analysed to characterise the effect of the pilot flame on main flame stabilisation.

FSD is an important parameter in analysing turbulent flames. In turbulent premixed combustion, the mean reaction rate can be expressed as [36]

$$\dot{\omega }={\rho }_{\mathrm{u}}{S}_{\mathrm{L}}\mathsf{\Sigma }$$

(4)

where ${\rho }_{\mathrm{u}}$ is the unburned gas density, $S_{\mathrm{L}}$ is the stretched laminar burning velocity, and $\mathsf{\Sigma }$ is the FSD. Moreover, the turbulent burning velocity ($S_{\mathrm{T}}$) and flame wrinkling ratio ($A_{\mathrm{T}}/A_{\mathrm{L}}$) is also correlated with FSD [26]:

$${\displaystyle \frac{S_{\mathrm{T}}}{S_{\mathrm{L}}}}={\displaystyle \frac{A_{\mathrm{T}}}{A_{\mathrm{L}}}}={\int }_{-\infty }^{\infty }\mathsf{\Sigma }d\eta$$

(5)

where $\eta$ is the coordinate normal to the flame brush. According to Pope [37] and Halter et al. [38], the local FSD is determined as

$$\mathsf{\Sigma }(c^{*})=(\overline{|\nabla c||c=c^{*}})\overline{P}(c^{*})$$

(6)

where $c$ is the progress variable, describing the reaction progress, with 0 corresponding to the fresh mixture and one the burnt mixture.$\overline{\left|\nabla c\right|\left|c=c^{*}\right.}$ represents the conditional average modulus of progress variable gradient for $c=c^{*}$. $\overline{P}(c^{*})$ is the probability of $c=c^{*}$ at the specific location.

Since the present study focuses on the flame stability of the main flame, NH, rather than OH, is used to indicate the flame front and define the flame brush. Compared with the APF, the MPF demonstrates superior performance in enhancing ammonia flame surface density, particularly at the flame root. Figure 8 presents the FSD contours of ammonia swirl flames with different pilot flames and at various P_pilot values. At P_pilot = 0.8 kW, flames with either pilot flame exhibit low FSD and a thick flame brush, indicating that the ammonia flame is highly vulnerable to perturbation under these conditions [39]. At higher P_pilot, flames stabilised by the MPF exhibit markedly elevated FSD at the flame root. As illustrated in Equation (6), elevated FSD reflects a higher probability of flame front in this region, which means a stable flame root is formed. Therefore, the MPF establishes a stable flame front at the root of the main ammonia flame. In contrast, although higher FSD is also observed at flame root in NH₃-piloted flames at P_pilot = 1.2–2.0 kW, the enhancement remains relatively limited.

Figure 8. Flame surface density contour of ammonia swirl flames with APF (left) and MPF (right) at P_pilot = (a) 0.8 kW, (b) 1.2 kW, (c) 1.6 kW, and (d) 2.0 kW.

The above-mentioned variations in ϕ_b,main and flame structures can be attributed to different flame stabilisation modes as the flame approaches blow-off. Figure 9 presents instantaneous NH/OH distributions in ammonia swirl flames approaching blow-off. For flames with an APF, burnt islands and unburnt pockets are observed, which is attributed to turbulence–flame interactions [40]. This indicates that the flame experiences local extinction under these conditions [32]. Under these conditions, the APF forms a V-shape flame at the pilot-stage exit, with the combustion region located in the inner shear layer. The main flame reactants are recirculated to the primary recirculation zone before being ignited, and the flame is mainly stabilised by the primary recirculation zone.

Figure 9. Instantaneous normalised NH/OH distribution in ammonia swirl flames approaching blow-off with (a) APF and (b) MPF at varying P_pilot.

In contrast, the MPF results in distinct flame characteristics under approaching blow-off conditions. At P_pilot = 0.8 kW, strong OH signals can be observed at the flame root, while NH is located further downstream, resulting in a lifted NH region. Similar to the APF, a V-shape CH₄ flame anchors at the pilot-stage exit and the main flame is ignited in the inner shear layer. This suggests that ammonia swirl flames with a low-power MPF are also stabilised by the primary recirculation zone. At higher P_pilot, however, a robust M-shape CH₄ flame is established at the pilot stage, with OH distributed in both the inner and outer shear layers of the pilot stage. The transition of the pilot flame from V-shape to M-shape can be attributed to its increased resistance to strain, which allows the flame to propagate and stabilise in the highly strained outer shear layer region [41]. This forms a stable ignition source for the main flame at the main-stage exit and in the primary recirculation zone, enabling efficient NH₃ consumption and NH formation in these two regions. As P_pilot is further increased to 2.0 kW, the pilot flame can burn independently at the pilot stage while no ammonia is supplied to the main stage, and the figure enclosed by the dashed box in Figure 9 is the OH-PLIF from a pure methane flame.

To conclude, two flame stabilisation mechanisms within ammonia swirl flames can be revealed. For flames with APFs and MPFs of low P_pilot, the flame is mainly stabilised by the primary recirculation zone, where pilot flames provide an ignition source. In contrast, for flames with MPFs of high P_pilot, the M-shape pilot flame provides an ignition source at the main-stage exit and in the primary recirculation zone, which is more effective in stabilising the main flame. At a P_pilot that is 14.2–21.3% of P_main, the main flame stability is significantly enhanced. Compared to a methane flame of an identical thermal power, this approach is estimated to reduce carbon emissions by 82.4–87.6%. At P_pilot = 1.8 kW, the thermal power of the pilot flame only counts for 21.3% of the main flame under stable operating conditions. However, the pilot flame guarantees the flame against blow-off even if no fuel is supplied to the main stage.

4. Conclusions

This study investigates the flame morphologies, flame surface density, and stability limits of premixed ammonia/air main flames stabilised by a pilot flame in a dual-swirl gas turbine model combustor. The effects of pilot fuel type (NH₃ vs. CH₄) and thermal power are systematically examined. The main conclusions are as follows:

• For flames with APF, increasing P_pilot shifts the OH radicals radially outward and reduces both total OH intensity and density, while NH distribution remains largely unchanged. For flames with MPF, a stratified structure emerges at P_pilot ≥ 1.2 kW, with OH concentrated in the pilot stage and NH in the main stage. Although the total NH intensity is lower for flames with MPF, the NH signal density is significantly higher, indicating a more compact and intense reaction zone in the main ammonia flame.
• Flames with MPF exhibit lower ϕ_b,main than those with APF across all P_pilot values. The MPF markedly enhances the FSD of the ammonia main flame, particularly at the flame root, where elevated FSD reflects a higher probability of flame front occurrence. At P_pilot = 1.2–2.0 kW, flames stabilised by MPF exhibit narrower radial FSD profiles and higher peak values compared to those with APF, indicating a thinner flame brush and improved flame stability. The effective flame stabilisation results from the abundance of reactive radicals and the high-temperature environment provided to the main flame root.
• Two stabilisation modes are identified for ammonia main flames as P_pilot increases, namely primary recirculation zone-dominated stabilisation with APF and with MPF at low P_pilot (<1.2 kW) and pilot-dominated stabilisation with MPF at intermediate P_pilot (1.2–1.8 kW), where the main flame is sustained by the ignition source formed by the pilot flame at the main-stage exit and in the primary recirculation zone. Compared to a methane flame of an identical thermal power, an MPF at P_pilot = 1.2–1.8 kW can reduce carbon emissions by 82.4–87.6%. These findings demonstrate that a low-power MPF offers an effective strategy for enhancing ammonia flame stability in gas turbine combustors and reducing carbon emissions.