Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Enhancing deep learning of ammonia/natural gas combustion kinetics via physics-aware data augmentation and scale separation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A network trained on one laminar flame, with physics-aware augmentation and a scale-split target, replaces stiff chemistry integration in ammonia/methane flames with up to a 20x speedup.

desk verdict A solid, practical ML-surrogate combustion paper: the interpolation augmentation and direct increment power transform are genuinely useful, and the 2D HIT validation deserves peer review despite the unmeasured coverage and mass-conservation gaps. read the letter →

arxiv 2507.08277 v2 pith:SGLOIMYS submitted 2025-07-11 physics.flu-dyn

Ke Xiao , Yangchen Xu , Han Li , Zhi X. Chen This is my paper · ORCID
classification physics.flu-dyn PACS 47.70.Pq
keywords deeplearningchemicalkineticsammonia-methanecombustiondataaugmentationflamestructureinterpolationscaleseparationstiffODEsurrogateturbulentpremixed
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to show that a deep neural network can replace the stiff chemical ODE solver inside ammonia/natural-gas flame simulations without losing fidelity, using training data drawn from a single one-dimensional laminar flame. Two obstacles stand in the way: reactive flame-front states are badly under-sampled, and the species increments the network must predict span many orders of magnitude. The paper's remedies are a physics-aware augmentation pipeline — linear interpolation of neighboring flame states at even temperature intervals, plus constrained random perturbations — and a scale-separated target transformation applied directly to the concentration change. Validated in a two-dimensional flame under homogeneous isotropic turbulence, the surrogate reproduces the reference solver's flame evolution, heat release, and low-temperature species fields while delivering up to a 20x speedup in end-to-end simulation time.

What carries the argument

Two mechanisms carry the argument. The first is physics-aware data augmentation: linear interpolation between neighboring thermochemical states at evenly spaced temperature values densifies the flame-front region that the raw 1D flame under-samples, and constrained random perturbation — $T' = T + 100X$, $p' = p + 0.15(p_{\max}-p_{\min})X$, $Y'_\alpha = Y_\alpha + 0.15X$ with $X$ uniform on $[-1,1]$, then filtering on temperature range, N$_2$ mass-fraction bounds, and negative heat release — broadens the composition manifold so the network tolerates states displaced from the laminar trajectory. The second is the scale-separated target: learning $F[\Delta Y] = \operatorname{sign}(\Delta Y)|\Delta Y|^{0.1}/0.1$ instead of a difference of transformed concentrations decompresses the low-magnitude regime, keeping tiny predicted increments tiny. The hybrid deployment rule — Box-Cox targets above 1000 K, power-transformed targets from 305 K to 1000 K, zero increments below 305 K — is what lets one surrogate stay accurate at high temperature and stable in low-temperature mixing regions.

What would settle it

Run the same 2D HIT flame kernel with noticeably stronger turbulence than the reference case, or with an equivalence-ratio-stratified NH$_3$/CH$_4$ mixture, and compare the low-temperature species fields (NH$_3$, NO$_2$, C$_3$H$_7$) against the CVODE reference: the central claim fails if the surrogate's fields drift from the reference in the mixing regions, because that would mean the augmented 1D manifold did not cover the turbulent states.

Watch

Extended reading notes

Core claim

The paper's central claim is that DFODE — a multilayer perceptron mapping local thermochemical states $\{T, p, \mathbf{Y}\}$ to species increments $\Delta\mathbf{Y}$ over a fixed $10^{-6}$ s step — can replace stiff ODE integration inside a reacting-flow solver for a 60\% NH$_3$/40\% CH$_4$ premixed flame, provided the training set and the target encoding are handled correctly. The authors show that interpolating between adjacent flame states at evenly spaced temperatures fills the under-represented reaction zone, and that random perturbations of $T$, $p$, and the mass fractions, filtered to exclude nonphysical states, widen the covered composition manifold to roughly 8 million training states. They further show that the established target formulation $\hat{Y} = \mathrm{BoxCox}(Y(t+\Delta t)) - \mathrm{BoxCox}(Y(t))$ loses distinguishability for tiny increments when the initial species mass fraction is not near zero, whereas $\hat{Y} = \operatorname{sign}(\Delta Y)|\Delta Y|^{0.1}/0.1$ keeps those increments small and reliable. A hybrid model using the Box-Cox formulation above 1000 K and the power-transformed target below reproduces the reference solver's flame-area evolution, conditional heat release below 1000 K, and spatial fields of NH$_3$, NO$_2$, and C$_3$H$_7$ in the 2D HIT flame, with roughly a 526x speedup in chemistry source-term evaluation and up to a 20x total simulation speedup.

Load-bearing premise

The load-bearing premise is that the 1D laminar flame states, after interpolation and constrained perturbation, cover the thermochemical states the 2D turbulent flame actually visits, so the network is never asked to extrapolate beyond its training manifold.

Editorial extensions

If this is right

  • A single canonical 1D flame, after augmentation, is enough training data for a surrogate that generalizes to a 2D turbulent flame under the same operating conditions.
  • The temperature-split hybrid target (Box-Cox above 1000 K, power-transformed below) captures low-temperature species fields that a Box-Cox-only model misses.
  • Chemistry source-term evaluation becomes roughly 526x faster than CVODE, and the end-to-end flame simulation about 20x faster.
  • The data-generation and augmentation pipeline can be re-run for other ammonia blending ratios or mixture compositions by regenerating the canonical flame data, since the authors state the trained model itself is limited to the same operating conditions as the sampled flame.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • An implicit stress test the paper does not run: pushing the 2D HIT case to higher turbulence intensity or near-extinction conditions, where mixing drives states farther off the augmented 1D manifold; the perturbation magnitudes in Equation (5) set an upper bound on how far the surrogate can be trusted to wander.
  • The Small-Scale Prediction Index is a template for the field: reporting small-target fidelity separately from $R^2$ would make surrogate comparisons in stiff-chemistry problems more meaningful, since $R^2$ is insensitive to the tiny-increment regime.
  • The hand-set 1000 K switch between target formulations hints at an adaptive alternative — a learned or locally weighted blend of target encodings could remove the threshold and its per-case tuning.
  • The augmentation logic is mechanism-agnostic and should transfer to other stiff fuels, such as hydrogen or dual-fuel blends, whose canonical flame data under-sample the reactive zone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This paper proposes a deep learning surrogate (DFODE) for ammonia/natural gas combustion kinetics with the 59-species Okafor mechanism. Training data are generated from a single 1D freely propagating premixed laminar flame, then augmented by linear interpolation between neighboring flame states and by constrained random perturbations. The paper compares two target formulations—a Box-Cox transformation of absolute concentrations and a power transformation of the concentration increment—and proposes a hybrid BC/PT model with a 1000 K switching temperature and a 305 K frozen-chemistry threshold. The models are validated in the same 1D laminar flame and then tested a posteriori in a 2D HIT premixed flame, where global flame area and conditional species statistics are compared with CVODE. The central claims are that the augmentation and scale-separation improve fidelity and that the surrogate delivers up to a 20x end-to-end speedup.

Significance. The paper addresses a practically relevant problem: accelerating stiff-chemistry evaluation for NH3/CH4 combustion with a comprehensive mechanism. The interpolation-based augmentation is a simple but useful idea for mitigating data imbalance near the flame front, and the PT target formulation demonstrates a clear improvement in small-concentration predictions (SSPI increases from 0.69 to 0.99 on the test set). The a posteriori 2D HIT validation is a positive step: the hybrid model reproduces the CVODE flame-area evolution and conditional HRR and C3H7 profiles with close agreement. The speedup, if confirmed, would make DNS/LES of ammonia flames more tractable. However, the generalization claim is currently over-stated relative to the evidence, and several load-bearing details (mass conservation, threshold sensitivity, timing breakdown) need to be supplied before the paper can be recommended for publication.

major comments (4)
  1. [Section 2.1 and Section 2.4] The DFODE predicts species increments ΔY = Y(t+Δt) − Y(t) for the 58 reactive species, with Ar excluded from the outputs because it is assumed constant. The manuscript nowhere describes a constraint that the predicted increments sum to zero, nor any renormalization of the updated mass fractions. In a reacting-flow simulation, if Σ ΔY_α ≠ 0, total mass is not conserved and the composition will drift over the 4 ms of the 2D HIT runs. The authors should either predict N−1 species and obtain the last species (e.g., Ar or N2) by difference, or enforce Σ ΔY_α = 0 in the loss, and they should report the residual sum of increments in the validation. This is a load-bearing physical-consistency issue because the claimed 'high fidelity' of the flame statistics depends on accurate species budgets.
  2. [Section 3.3 and Section 2.3] The abstract states that the trained models 'generalize to unseen conditions,' but the only out-of-sample test is the 2D HIT flame, which uses the same fuel blend (60% NH3/40% CH4), the same equivalence ratio, the same unburnt temperature (300 K), and the same pressure as the 1D training flame. What is 'unseen' is the turbulent flow structure, not the thermochemical operating point. The paper does not quantify how far the 2D HIT thermochemical states lie from the augmented training manifold (e.g., by PCA projection, nearest-neighbor distance, or kernel density). Without such a coverage measure, the success of the 2D test cannot be attributed to the augmentation strategy, and the boundary of the model's applicability remains unknown. I recommend adding a simple coverage analysis for the HIT states relative to the training set.
  3. [Section 3.3] The hybrid BC/PT model introduces two thresholds: the frozen-chemistry limit of 305 K (below which ΔY is set to zero) and the switching temperature of 1000 K (above which the BC model is used, below which the PT model is used). No sensitivity analysis is presented for either threshold. The 1000 K choice is justified only by a qualitative inspection of Figure 7, and the 305 K value is stated without justification. Since these thresholds are part of the proposed method, their robustness should be demonstrated, e.g., by rerunning the 2D HIT case with thresholds of 900 K, 1100 K and 290 K, 320 K, or by reporting the sensitivity of the SSPI and flame-area error to these parameters.
  4. [Section 3.3, final paragraph] The paper reports a 526x speedup in chemistry source-term calculations and a 20x end-to-end speedup for the DFODE simulations, but it does not provide wall-clock times for the CVODE, BC-only, and BC/PT hybrid runs, nor a breakdown of time spent in inference, data transfer, and flow solving. The hybrid model requires two neural networks and a threshold switch, so its overhead cannot be assessed from the aggregate speedup. Please report the measurement procedure (warm-up, number of repetitions, hardware) and the per-configuration timings.
minor comments (5)
  1. [Equation (5)] The formula for Y'_α appears garbled: 'Y 1+0.15·X α' is dimensionally inconsistent. It should likely be Y'_α = Y_α + 0.15·span(Y_α)·X_α (or similar). Please correct the typesetting and clarify whether X_α denotes an independent random draw per species.
  2. [Figures 5–10] Several figures contain '/uni...' artifact tokens that obscure the plotted curves; the production figures need to be cleaned before publication.
  3. [Section 3.3] The description of the data-batching optimization ('simplified the process of invoking PyTorch') is vague; a few sentences on the batching strategy and the resulting GPU utilization would help readers understand the transfer overhead.
  4. [Section 2.4] The inverse of the power transformation F(y) = sign(y)·|y|^λ/λ is never written; stating the inversion (and how it is applied to the predicted values) would improve reproducibility.
  5. [Table 1] The definition of 'Num. pred.' could be clarified: 8,000,000 states × 58 species = 464,000,000 predictions, but the number of species is not obvious from the table alone.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central generalization claim is checked against an independent CVODE reference in a 2D HIT flame, not against the training data.

full rationale

The paper's derivation chain is an empirical regression: DNNs are trained on CVODE-generated species increments from a 1D laminar flame, and the claimed accuracy is assessed by comparing DFODE-coupled CFD against CVODE-coupled CFD in the same 2D HIT configuration. The training target ΔY=Y(t+Δt)-Y(t) is obtained from the external CVODE solver, and the 2D a posteriori test is an independent benchmark with the same chemical mechanism but a different flow configuration; no equation in the paper defines the validation result in terms of the fitted parameters. The interpolation and perturbation augmentation (Algorithm 1, Eqs. (3)-(5)) alter the training distribution but do not encode the validation metrics. The BC/PT transformation comparison (Eqs. (6)-(7)) is a target-formulation choice and is evaluated on a held-out HIT snapshot; while the 1000 K switching threshold is selected with reference to that snapshot (Section 3.2), this is a mild test-set-informed model selection rather than a construction that forces the subsequent flame-area agreement. The 1D laminar validations use the same flame configuration that generated the training data (Section 3.1), so they are in-sample ablations, not evidence of generalization; the paper does not rely on them for the generalization claim. Citations to prior work by the same group (Refs. [32,33,37]) supply solver infrastructure and training methodology but are not load-bearing 'uniqueness' arguments, and the core validation is self-contained against CVODE. No circular step was found.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The central claims rest on several hand-set parameters (perturbation amplitudes, filter bounds, transform exponent, and switching thresholds) and on the domain assumption that the 1D flame manifold plus augmented points spans the 2D HIT state space. No new physical particles, forces, or conserved quantities are proposed. The SSPI metric is a new evaluation statistic, not a physical entity. The species mass conservation assumption is left implicit.

free parameters (8)
  • Exponent lambda in power and Box-Cox transforms = 0.1
    Lambda = 0.1 in Box-Cox(Y(t+dt))-Box-Cox(Y(t)) and in sign(y)|y|^0.1/0.1 is a hand-chosen hyperparameter, carried from prior practice, controlling the scale compression.
  • Random perturbation amplitudes = T +/- 100 K, p +/- 15% of range, Y +/- 0.15
    Equations (3)-(5); these hand-picked amplitudes define how far augmented states move from the flame manifold.
  • N2 mass fraction filter margin = 5% of span
    Section 2.3; perturbed N2 mass fractions are kept within +/- 0.05 of the original span, an ad hoc physical plausibility bound.
  • Temperature filter bounds = 290 K to max(T)+100 K
    Section 2.3; accepted perturbed temperatures are restricted to this range with no sensitivity analysis.
  • Heat release rate filter threshold = Unspecified (discard significant negative HRR)
    Section 2.3; states with large negative heat release are removed as nonphysical, but the cutoff is not quantified.
  • BC/PT switching temperature = 1000 K
    Section 3.2; threshold chosen from Figure 7 to separate low-temperature small-target failures from high-temperature accuracy; selected post hoc.
  • Frozen chemistry threshold = 305 K
    Section 3.3; species changes are set to zero below 305 K, just above the 300 K unburnt temperature, with no sensitivity study.
  • MLP architecture = 4 hidden layers x 800 neurons
    Section 2.4; architecture adopted from prior work and not optimized here.
assumptions (6)
  • domain assumption The 1D laminar premixed flame manifold, after interpolation and constrained random perturbations, spans the thermochemical space of the 2D HIT flame.
    Sections 2.2 and 2.3 limit training data to one canonical flame and then augment it; Section 3.3 tests the surrogate in 2D HIT. The generalization claim depends entirely on this coverage.
  • ad hoc to paper Linear interpolation between neighboring flame states in temperature yields physically meaningful thermochemical states.
    Algorithm 1 assumes species mass fractions vary approximately linearly with temperature across the flame front; no equilibration or chemical consistency check is applied to interpolated states.
  • domain assumption The Okafor et al. mechanism is an adequate ground truth for NH3/CH4 combustion.
    Section 2.2 uses the 59-species, 356-reaction mechanism; neural network outputs are trained against CVODE integration of this mechanism, so the mechanism's accuracy bounds the surrogate.
  • ad hoc to paper The DNN-predicted species increments need no explicit mass-conservation constraint (sum of Delta Y equal to zero).
    Section 2.4 describes outputs for all species except argon but does not describe renormalization or enforcement of element conservation; the solver must tolerate or correct any drift.
  • domain assumption A fixed integration step of dt = 1e-6 s is a valid regression target for all species and temperatures.
    Section 2.1 defines the output as Y(t+dt)-Y(t); stiff species may require smaller steps in some zones, and the frozen-chemistry cutoff at 305 K is a related simplification.
  • standard math CVODE integration is the correct reference for the chemical source terms.
    CVODE is used throughout as ground truth; it is a standard, externally implemented solver, so this is a reasonable but unvalidated reference for the surrogate's accuracy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing deep learning of ammonia/natural gas combustion kinetics via physics-aware data augmentation and scale separation." pith.science (2026). https://pith.science/paper/SGLOIMYS

@misc{pith2026250708277,
  author       = {Pith},
  title        = {Pith review of: Enhancing deep learning of ammonia/natural gas combustion kinetics via physics-aware data augmentation and scale separation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SGLOIMYS}},
  note         = {Machine review of arXiv:2507.08277}
}
read the original abstract

Accurate and efficient numerical simulation of ammonia combustion is critical for advancing ammonia-based energy systems, where turbulent flame dynamics and pollutant formation strongly affect practical applicability. However, such simulations are hindered by the need to solve high-dimensional stiff chemical ordinary differential equations (ODEs), which constitute the primary computational bottleneck. To address this challenge, this study explores Deep learning for solving Flame chemical kinetics with stiff ODEs (DFODE) in ammonia/natural gas combustion. Thermochemical training data are obtained from one-dimensional (1D) freely propagating premixed laminar flames, and a physics-aware augmentation strategy combining interpolation of neighboring states with constrained random perturbations is introduced to overcome sampling imbalance near steep flame-front gradients. In addition, transformation strategies for model target formulation were evaluated, and the prediction accuracy in low-temperature regimes was notably enhanced through scale separation for targets spanning multiple orders of magnitude. Validation in 1D laminar flames confirms the effectiveness of these refinements, while a posteriori evaluation in a two-dimensional (2D) propagating flame under homogeneous isotropic turbulence (HIT) demonstrates that the trained models generalize to unseen conditions. The DNN surrogates reproduce flame characteristics with high fidelity and deliver up to a 20x speedup in end-to-end CFD simulations. These results highlight the potential of deep learning-based chemical kinetics to accelerate ammonia/natural gas combustion modeling, supporting efficient and scalable high-fidelity simulations for emerging zero-carbon energy systems.

Figures

Figures reproduced from arXiv: 2507.08277 by the authors.

Figure 1
Figure 1. Overview of the training procedure and flow chart for the dfLowMachFoam [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Comparison of thermochemical data before and after flame structure interpo [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Comparison of interpolated thermochemical data before and after perturbation [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Comparison of target value formulations and model performance colored by [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: Temperature, heat release rate (HRR), and species mass fractions profiles in [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: Heat release rate (HRR), and species mass fractions profiles in progress variable [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]
Figure 7
Figure 7. Figure 7: Scatter plots of BC model predictions for targets with magnitudes less than [PITH_FULL_IMAGE:figures/full_fig_p022_7.png]
Figure 8
Figure 8. Figure 8: Temporal evolution of flame area in 2D HIT simulations: comparison between [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: Conditional-averaged heat release rate (HRR) and C [PITH_FULL_IMAGE:figures/full_fig_p026_9.png]
Figure 10
Figure 10. Figure 10: Contour plots of species mass fractions (NH [PITH_FULL_IMAGE:figures/full_fig_p027_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 30 canonical work pages

  1. [1]

    Ammonia for power

    Valera-Medina A, Xiao H, Owen-Jones M, David W, Bowen P. Ammonia for power. Prog in Energy and Combust Sci. 2018;69:63–102.https: //doi.org/10.1016/j.pecs.2018.07.001

    doi:10.1016/j.pecs.2018.07.001 2018

  2. [2]

    Review on Ammonia as a Potential Fuel: From Synthesis to Economics

    Valera-Medina A, Amer-Hatem F, Azad A, Dedoussi I, De Joannon M, Fernandes R, et al. Review on Ammonia as a Potential Fuel: From Synthesis to Economics. Energy & Fuels. 2021;35(9):6964–7029.https: //doi.org/10.1021/acs.energyfuels.0c03685

    work page doi:10.1021/acs.energyfuels.0c03685 2021

  3. [3]

    Ammonia as a hydrogen energy carrier

    Kojima Y, Yamaguchi M. Ammonia as a hydrogen energy carrier. Int 29 J of Hydrog Energy. 2022;47(54):22832–22839.https://doi.org/10. 1016/j.ijhydene.2022.05.096

    work page 2022

  4. [4]

    A review of ammonia as a compression ignition engine fuel

    Dimitriou P, Javaid R. A review of ammonia as a compression ignition engine fuel. Int J of Hydrog Energy. 2020;45(11):7098–7118.https: //doi.org/10.1016/j.ijhydene.2019.12.209

    work page doi:10.1016/j.ijhydene.2019.12.209 2020

  5. [5]

    Unlocking ammonia engines: pre-chamber ignition with partial ammonia cracking

    Sun J, Tang Q, Zhu X, Wen M, Huang L, Ming Z, et al. Unlocking ammonia engines: pre-chamber ignition with partial ammonia cracking. Energy. 2025;335:137961.10.1016/j.energy.2025.137961

    work page arXiv 2025

  6. [6]

    A review on clean ammonia as a potential fuel for power generators

    Yapicioglu A, Dincer I. A review on clean ammonia as a potential fuel for power generators. Renew and Sustain Energy Rev. 2019;103:96–108. https://doi.org/10.1016/j.rser.2018.12.023

    work page doi:10.1016/j.rser.2018.12.023 2019

  7. [7]

    Ammonia as Effective Hydrogen Storage: A Review on Production, Storage and Utilization

    Aziz M, Wijayanta A, Nandiyanto A. Ammonia as Effective Hydrogen Storage: A Review on Production, Storage and Utilization. Energies. 2020;13(12):3062.https://doi.org/10.3390/en13123062

    doi:10.3390/en13123062 2020

  8. [8]

    Emission and Combustion Characteristics of Ammonia/Methane Mixtures for Carbon Reduction and Alternative Fuel Development

    Jurić F, Ban M, Baleta J, Honus S, Duić N, Vujanović M. Emission and Combustion Characteristics of Ammonia/Methane Mixtures for Carbon Reduction and Alternative Fuel Development. Energy. 2025;:13817010. 1016/j.energy.2025.138170

    work page arXiv 2025

Show all 46 references
  1. [9]

    Effects of injection pa- rameters on the combustion characteristics of ammonia-diesel dual-fuel direct-injection (ADDI) mode and combustion enhancement at high ammonia energy ratio

    Wu H, Jin Z, Qian Y, Zhou D, Shi L, Lu X. Effects of injection pa- rameters on the combustion characteristics of ammonia-diesel dual-fuel direct-injection (ADDI) mode and combustion enhancement at high ammonia energy ratio. Energy. 2025;327:136449.10.1016/j.energy. 2025.136449

    2025

  2. [10]

    Renew and Sustain Energy Rev

    ChaiW,BaoY,JinP,TangG,ZhouL.Areviewonammonia, ammonia- hydrogen and ammonia-methane fuels. Renew and Sustain Energy Rev. 2021;147:111254.https://doi.org/10.1016/j.rser.2021.111254

    2021

  3. [11]

    Propagation and emissions of premixed methane-ammonia/air flames

    Ku J, Ahn Y, Kim H, Kim Y, Kwon O. Propagation and emissions of premixed methane-ammonia/air flames. Energy. 2020;201:117632.10. 1016/j.energy.2020.117632

    2020

  4. [12]

    Toward accommodating realistic fuel chemistry in large- scale computations

    Lu T, Law C. Toward accommodating realistic fuel chemistry in large- scale computations. Prog in Energy and Combust Sci. 2009;35(2):192– 215.https://doi.org/10.1016/j.pecs.2008.10.002. 30

    2009 doi

  5. [13]

    Revealing the mechanisms of ammonia dual- fuel combustion for decarbonization in marine transportation

    Liu X, Tang Q, Xu L. Revealing the mechanisms of ammonia dual- fuel combustion for decarbonization in marine transportation. Energy. 2025;334:137611.10.1016/j.energy.2025.137611

    2025

  6. [14]

    Numerical simulation of ammonia/methane/air combustion using reduced chem- ical kinetics models

    Mikulčić H, Baleta J, Wang X, Wang J, Qi F, Wang F. Numerical simulation of ammonia/methane/air combustion using reduced chem- ical kinetics models. Int J of Hydrog Energy. 2021;46(45):23548–23563. https://doi.org/10.1016/j.ijhydene.2021.01.109

    2021 doi

  7. [15]

    Predictions of NO and CO emissions in ammonia/methane/air combus- tion by LES using a non-adiabatic flamelet generated manifold

    Honzawa T, Kai R, Okada A, Valera-Medina A, Bowen P, Kurose R. Predictions of NO and CO emissions in ammonia/methane/air combus- tion by LES using a non-adiabatic flamelet generated manifold. Energy. 2019;186:115771.10.1016/j.energy.2019.07.101

    2019 doi

  8. [16]

    Machine Learning for Chemical Reactions

    Meuwly M. Machine Learning for Chemical Reactions. Chem Rev. 2021;121(16):10218–10239.https://doi.org/10.1021/acs.chemrev. 1c00033

    2021 doi

  9. [17]

    A multi-scale sampling method for accurate and robust deep neural network to predict combustion chemical kinetics

    Zhang T, Yi Y, Xu Y, Chen Z, Zhang Y, E W, et al. A multi-scale sampling method for accurate and robust deep neural network to predict combustion chemical kinetics. Combust and Flame. 2022;245:112319. https://doi.org/10.1016/j.combustflame.2022.112319

    2022

  10. [18]

    Machine learning for com- bustion chemistry

    Echekki T, Farooq A, Ihme M, Sarathy S. Machine learning for com- bustion chemistry. In: Machine learning and its application to reacting flows: ML and combustion. Springer International Publishing Cham; 2023, p. 117–147

    2023

  11. [19]

    Artificial intelligence as a catalyst for combustion science and engineering

    Ihme M, Chung W. Artificial intelligence as a catalyst for combustion science and engineering. Proc of the Combust Inst. 2024;40(1-4):105730. https://doi.org/10.1016/j.proci.2024.105730

    2024

  12. [20]

    Numerical simulation of tur- bulent combustion: Scientific challenges

    Ren Z, Lu Z, Hou L, Lu L. Numerical simulation of tur- bulent combustion: Scientific challenges. Sci China Phys, Mech & Astron. 2014;57(8):1495–1503.https://doi.org/10.1007/ s11433-014-5507-0

    2014

  13. [21]

    Machine learning tabu- lation of thermochemistry in turbulent combustion: An approach based on hybrid flamelet/random data and multiple multilayer perceptrons

    Ding T, Readshaw T, Rigopoulos S, Jones W. Machine learning tabu- lation of thermochemistry in turbulent combustion: An approach based on hybrid flamelet/random data and multiple multilayer perceptrons. 31 Combust and Flame. 2021;231:111493.https://doi.org/10.1016/j. combustfl...

    2021

  14. [22]

    Criteria to switch from tabulation to neural networks in computational combustion

    Nikolaou Z, Vervisch L, Domingo P. Criteria to switch from tabulation to neural networks in computational combustion. Combust and Flame. 2022;246:112425.https://doi.org/10.1016/j.combustflame.2022. 112425

    2022 doi

  15. [24]

    On-the-fly artificial neural network for chemical kinetics in direct numerical simulations of premixed combus- tion

    Chi C, Janiga G, Thévenin D. On-the-fly artificial neural network for chemical kinetics in direct numerical simulations of premixed combus- tion. Combust and Flame. 2021;226:467–477.https://doi.org/10. 1016/j.combustflame.2020.12.038

    2021

  16. [25]

    Chemistry reduction using machine learning trained from non-premixed micro-mixing modeling: Application to DNS of a syngas turbulent oxy-flame with side-wall ef- fects

    Wan K, Barnaud C, Vervisch L, Domingo P. Chemistry reduction using machine learning trained from non-premixed micro-mixing modeling: Application to DNS of a syngas turbulent oxy-flame with side-wall ef- fects. Combust and Flame. 2020;220:119–129.https://doi.org/10. 1016/j.comb...

    2020

  17. [26]

    Artificial neural network chem- istry solving for high-pressure hydrogen–air combustion

    Béroudiaux A, Vervisch L, Domingo P. Artificial neural network chem- istry solving for high-pressure hydrogen–air combustion. Int J of Hydrog Energy. 2025;109:669–683.https://doi.org/10.1016/j.ijhydene. 2025.02.054

    2025 doi

  18. [27]

    Chatzopoulos A, Rigopoulos S. A chemistry tabulation approach via Rate-Controlled Constrained Equilibrium (RCCE) and Artificial Neu- ral Networks (ANNs), with application to turbulent non-premixed CH4/H2/N2 flames. Proc of the Combust Inst. 2013;34(1):1465–1473. https://doi.or...

    2013 doi

  19. [28]

    Tabulation of combus- tion chemistry via Artificial Neural Networks (ANNs): Methodology and application to LES-PDF simulation of Sydney flame L

    Franke L, Chatzopoulos A, Rigopoulos S. Tabulation of combus- tion chemistry via Artificial Neural Networks (ANNs): Methodology and application to LES-PDF simulation of Sydney flame L. Com- bust and Flame. 2017;185:245–260.https://doi.org/10.1016/j. combustflame.2017.07.014. 32

    2017 doi

  20. [29]

    Simulation of turbulent premixed flames with machine learning - tabulated thermochemistry

    Readshaw T, Franke L, Jones W, Rigopoulos S. Simulation of turbulent premixed flames with machine learning - tabulated thermochemistry. Combust and Flame. 2023;258:113058.https://doi.org/10.1016/j. combustflame.2023.113058

    2023

  21. [30]

    A feasibility study on the use of low-dimensional simulations for database generation in adaptive chem- istry approaches

    Newale A, Sharma P, Pope S, Pepiot P. A feasibility study on the use of low-dimensional simulations for database generation in adaptive chem- istry approaches. Combust Theory and Model. 2022;26(7):1239–1261. https://doi.org/10.1080/13647830.2022.2137062

    2022

  22. [31]

    Simplifying chemical kinetics: Intrinsic low- dimensional manifolds in composition space

    Maas U, Pope S. Simplifying chemical kinetics: Intrinsic low- dimensional manifolds in composition space. Combust and Flame. 1992;88(3-4):239–264.https://doi.org/10.1016/0010-2180(92) 90034-M

    1992 doi

  23. [32]

    Graphics process- ing unit/artificial neural network-accelerated large-eddy simulation of swirling premixed flames

    Zhang M, Mao R, Li H, An Z, Chen ZX. Graphics process- ing unit/artificial neural network-accelerated large-eddy simulation of swirling premixed flames. Phys of Fluids. 2024;36(5):055147.10.1063/ 5.0202321

    2024

  24. [33]

    Comprehensive deep learning for combustion chemistry integration: Multi-fuel gener- alization anda posteriorivalidation in reacting flow

    Li H, Yang R, Xu Y, Zhang M, Mao R, Chen Z. Comprehensive deep learning for combustion chemistry integration: Multi-fuel gener- alization anda posteriorivalidation in reacting flow. Phys of Fluids. 2025;37(1):015162.https://doi.org/10.1063/5.0248582

    2025 doi

  25. [34]

    Experimental and numerical study of the laminar burning velocity of CH4–NH3–air premixed flames

    Okafor E, Naito Y, Colson S, Ichikawa A, Kudo T, Hayakawa A, et al. Experimental and numerical study of the laminar burning velocity of CH4–NH3–air premixed flames. Combust and Flame. 2018;187:185–198. https://doi.org/10.1016/j.combustflame.2017.09.002

    2018 doi

  26. [35]

    Cantera: An Object-oriented Software Toolkit for Chemical Kinetics, Thermody- namics, and Transport Processes

    Goodwin D, Moffat H, Schoegl I, Speth R, Weber B. Cantera: An Object-oriented Software Toolkit for Chemical Kinetics, Thermody- namics, and Transport Processes. 2022.https://doi.org/10.5281/ ZENODO.6387882

    2022

  27. [36]

    DeepFlame: A deep learning empowered open-source platform for reacting flow simulations

    Mao R, Lin M, Zhang Y, Zhang T, Xu Z, Chen Z. DeepFlame: A deep learning empowered open-source platform for reacting flow simulations. Comput Phys Commun. 2023;291:108842.https://doi.org/10.1016/ j.cpc.2023.108842. 33

    2023

  28. [37]

    An integrated framework for accelerating reactive flow simulation using GPU and ma- chine learning models

    Mao R, Zhang M, Wang Y, Li H, Xu J, Dong X, et al. An integrated framework for accelerating reactive flow simulation using GPU and ma- chine learning models. Proc of the Combust Inst. 2024;40(1-4):105512. 10.1016/j.proci.2024.105512

    2024

  29. [38]

    DeepFlame 2.0: A new version for fully GPU-native machine learning accelerated reacting flow simulations under low-Mach conditions

    Mao R, Dong X, Bai X, Wu Z, Dang G, Li H, et al. DeepFlame 2.0: A new version for fully GPU-native machine learning accelerated reacting flow simulations under low-Mach conditions. Comput Phys Commun. 2025;312:109595.10.1016/j.cpc.2025.109595

    2025

  30. [39]

    Machine learning tabulation of ther- mochemistry of fuel blends

    Ding T, Rigopoulos S, Jones W. Machine learning tabulation of ther- mochemistry of fuel blends. Appl in Energy and Combust Sci. 2022;12:100086.10.1016/j.jaecs.2022.100086

    2022

  31. [40]

    Delving into Deep Imbal- anced Regression

    Yang Y, Zha K, Chen Y, Wang H, Katabi D. Delving into Deep Imbal- anced Regression. 2021.10.48550/arXiv.2102.09554

    2021 doi arXiv

  32. [41]

    Data-driven simulation of ammonia combustion using neural ordinary differential equations (NODE)

    Saito M, Xing J, Nagao J, Kurose R. Data-driven simulation of ammonia combustion using neural ordinary differential equations (NODE). Appl in Energy and Combust Sci. 2023;16:100196.https://doi.org/10. 1016/j.jaecs.2023.100196

    2023

  33. [42]

    Clustering-Enhanced Deep Learning Method for Computation of Full Detailed Ther- mochemical States via Solver-Based Adaptive Sampling

    Chen X, Mehl C, Faney T, Di Meglio F. Clustering-Enhanced Deep Learning Method for Computation of Full Detailed Ther- mochemical States via Solver-Based Adaptive Sampling. Energy & Fuels. 2023;37(18):14222–14239.https://doi.org/10.1021/acs. energyfuels.3c01955

    2023 doi

  34. [43]

    Combust Theory and Model

    BlascoJ,FueyoN,DopazoC,ChenJ.Aself-organizing-mapapproachto chemistry representation in combustion applications. Combust Theory and Model. 2000;4(1):61–76.https://doi.org/10.1088/1364-7830/ 4/1/304

    doi

  35. [44]

    An improved approach towards more robust deep learning models for chemical kinetics

    Han X, Jia M, Chang Y, Li Y. An improved approach towards more robust deep learning models for chemical kinetics. Combust and Flame. 2022;238:111934.https://doi.org/10.1016/j.combustflame.2021. 111934

    2022 doi

  36. [45]

    Ding T, Jones W, Rigopoulos S. An improved machine learning method for thermochemistry tabulation, with application to LES-PDF simu- lations of piloted diffusion and swirl-bluff-body stabilised flames with 34 NOx formation. Combust and Flame. 2025;277:114130.10.1016/j. combust...

    2025

  37. [46]

    Efficient machine learningmethodforsupercriticalcombustion: Predictingreal-fluidprop- erties and chemical ODEs

    Cai Y, Yang R, Li H, Xu J, Xiao K, Chen Z, et al. Efficient machine learningmethodforsupercriticalcombustion: Predictingreal-fluidprop- erties and chemical ODEs. Aerosp Sci and Technol. 2025;159:110034. https://doi.org/10.1016/j.ast.2025.110034

    2025

  38. [47]

    DNSLab: A gateway to turbulent flow sim- ulation in Matlab

    Vuorinen V, Keskinen K. DNSLab: A gateway to turbulent flow sim- ulation in Matlab. Comput Phys Commun. 2016;203:278–289.https: //doi.org/10.1016/j.cpc.2016.02.023. 35

    2016 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.