Air Quality Implications of Using Ammonia as a Renewable Fuel: How Low Can NOx Emissions Go?

In addition to their lifecycle carbon emissions, another important issue with decarbonized energy pathways is their air quality, water, or land use implications. This paper considers the air quality issue for ammonia combustion. When directly combusting ammonia, reactions of its N atom with atmospheric oxygen lead to NOx emissions that are O(103) ppm, 2 orders of magnitude higher than EPA limits or the amount emitted by current natural-gas-fired technologies. In order to provide guidance to policymakers and technologists on what is fundamentally possible, this Perspective analyzes the fundamental minimum NOx emissions that can be produced from ammonia combustion. The analysis shows that it is possible to achieve quite low NOx emission levels of O(10) ppm, but these designs differ markedly from those used in today’s lean, premixed combustion systems.

Reactor Network Model Reduced order modeling by way of a chemical reaction network (CRN) was used in this study for the minimum NOx calculation (see Fig. 2).The reduced order modeling approach allows the main stage and second stage to be split and modeled as separate reaction networks.The reactor network modeling assumes a perfect premixed, 1-D flame that consumes the reactants in each stage.Adiabatic, constant-pressure batch reactors were used to mix various streams of gas coming into each stage.For the main stage, streams of fuel and oxidizer were mixed, while for the second stage, the products from the main stage were mixed with secondary air.The mass flow rates for each stream were calculated based on the global and main stage equivalence ratios (Φglobal and Φmain).The global equivalence ratio is specified, based upon the target combustor exit temperature.
The free flame model 26 outputs solutions in spatial coordinates, so numerical integration of axial velocities and distances between grid points was necessary to convert the solution to temporal coordinates.Although the flame zone generally starts at the same point in spatial coordinates, variations in flame speeds at different operating conditions cause variations in starting times for the flame.For this reason, peak NH2 concentration was used to define the start of the flame zone.
Therefore, time zero starts to count from where NH2 peaks to ensure a consistent definition of residence time.As one of the first species in the ammonia-air reaction pathway 27 , it is a useful baseline for referencing the rest of the system in temporal coordinates.
All calculations were done using Cantera 26 , which is an open-source library that can simulate chemical kinetics problems.The kinetic model used in this study was developed by Mei et al. 28 .
A few sample comparisons of the performance of different kinetic models and the selection of this kinetic model are provided in Fig. S1 and S2.
In the simulation, the fuel was pure ammonia, with a fixed temperature of 300 K. Oxidizer was synthetic air (79% N2 and 21% O2), with a preheat temperature of 650 K. Optimization was achieved at fixed values of Φglobal (overall equivalence ratio which controls combustor firing temperature), combustor pressure, and τglobal (total residence time) for a specific case.Each of these parameters was also individually varied to study their effects on minimum NO, which will be discussed in later sections.
Several kinetic models for ammonia combustion have been published, so a comparison was done to see which kinetic model would be most suitable for this study.Ignition delay times and laminar flame speeds were reviewed for agreement with experimental data, which is a standard format for validating kinetic models.Kinetic models by Glarborg from 2018 29 and 2022 30 , Mei 28 , Klippenstein 31 , and Okafor 32 were analyzed.For ignition delay times, comparisons were done with data from a study done by Mathieu and Petersen 33 .Shu et al. 34 also published ignition delay times for ammonia-air mixtures, for which a sample comparison can be seen in Fig. S1.Laminar flame speeds over a range of equivalence ratios were compared to datasets from Hayakawa et al. 35 , Takizawa et al. 36 , Pfahl et al. 37 , Zakaznov et al. 38 , and Ronney et al. 39 (Fig. S2).Since the Mei model showed overall good agreement with experimental data for both ignition delay and laminar flame speed, it was chosen as the suitable mechanism for this study.

Figure S1 .
Figure S1.Comparison of ignition delay times from various reaction mechanisms with

Figure S2 .
Figure S2.Comparison of laminar flame speeds from various experimental datasets and reaction