Abstract: Machine learning (ML) has gained significant traction in engine-related research, particularly because of its potential to improve predictive performance while reducing computational costs. However, most current applications rely on feedforward neural networks (FNNs; e.g., conventional artificial neural networks (ANNs)); these are well-suited for modeling static data and capturing nonlinear relationships, but do not explicitly encode temporal dependencies unless sequence context is introduced via feature engineering. Motivated by this limitation, we evaluate sequence-aware neural surrogates for engine-relevant combustion-chemistry time-series data. Specifically, the temporal evolution of an intermediate product group during polycyclic aromatic hydrocarbon (PAH) formation in C₂H₄/NH₃ pyrolysis is modeled using reactive force field (ReaxFF) molecular dynamics (MD) trajectories, comparing an FNN baseline (with explicit time as an input) against a long short-term memory (LSTM)-based recurrent neural network (RNN). The results show that while the FNN baseline benefits from explicit temporal feature engineering, its predictive performance is inferior to that of the LSTM model, even when the network depth is increased. This behavior is consistent with the architectural limitations of feedforward models, which do not maintain an internal memory state, and therefore, tend to generalize poorly when the target dynamics are history dependent. In contrast, the LSTM model leverages gated memory to learn temporal dependencies and consequently improves the predictive accuracy of combustion-chemistry time-series modeling, providing an efficient surrogate once trained. Overall, our findings delineate the conditions under which sequence-aware recurrent architectures offer advantages over feedforward models for ReaxFF MD time-series surrogate modeling.

面向发动机燃烧反应化学的ReaxFF分子动力学数据驱动神经网络代理模型

[1]AghbashloM, PengWX, TabatabaeiM, et al., 2021. Machine learning technology in biodiesel research: a review. Progress in Energy and Combustion Science, 85:100904.

[2]AliramezaniM, KochCR, ShahbakhtiM, 2022. Modeling, diagnostics, optimization, and control of internal combustion engines via modern machine learning techniques: a review and future directions. Progress in Energy and Combustion Science, 88:100967.

[3]AramburuA, GuidoC, BaresP, et al., 2024. Knock detection in spark ignited heavy duty engines: an application of machine learning techniques with various knock sensor locations. Measurement, 224:113860.

[4]AtkinsonCM, LongTW, HanzevackEL, 1998. Virtual Sensing: a Neural Network-Based Intelligent Performance and Emissions Prediction System for On-Board Diagnostics and Engine Control. SAE International, Warrendale, USA.

[5]BadraJA, KhaledF, TangM, et al., 2021. Engine combustion system optimization using computational fluid dynamics and machine learning: a methodological approach. Journal of Energy Resources Technology, 143(2):022306.

[6]BennettAM, LiuP, LiZP, et al., 2020. Soot formation in laminar flames of ethylene/ammonia. Combustion and Flame, 220:210-218.

[7]DiaoST, LiHT, YuMG, 2024. Atomic insights into the combustion mechanism of DME/NH₃ mixtures: a combined ReaxFF-MD and DFT study. International Journal of Hydrogen Energy, 80:743-753.

[8]HuangQ, LiuJL, 2024. Preliminary assessment of the potential for rapid combustion of pure ammonia in engine cylinders using the multiple spark ignition strategy. International Journal of Hydrogen Energy, 55:375-385.

[9]HuangQ, LiuJL, UlishneyC, et al., 2022. On the use of artificial neural networks to model the performance and emissions of a heavy-duty natural gas spark ignition engine. International Journal of Engine Research, 23(11):1879-1898.

[10]HuangQ, YangRM, LiuJH, et al., 2024. Investigation of the mechanism behind the surge in nitrogen dioxide emissions in engines transitioning from pure diesel operation to methanol/diesel dual-fuel operation. Fuel Processing Technology, 264:108131.

[11]HuangQ, YangRM, LiuJH, et al., 2025a. CFD-based investigation of ammonia combustion and slip behavior in an ammonia-diesel dual-fuel engine. Journal of the Energy Institute, 122:102217.

[12]HuangQ, XieTF, LiuJL, 2025b. Machine learning-assisted reconstruction of in-cylinder pressure in internal combustion engines under unmeasured operating conditions. Energies, 18(19):5235.

[13]KailasanathanRKA, YelvertonTLB, FangTG, et al., 2013. Effect of diluents on soot precursor formation and temperature in ethylene laminar diffusion flames. Combustion and Flame, 160(3):656-670.

[14]KamatAM, van DuinACT, YakovlevA, 2010. Molecular dynamics simulations of laser-induced incandescence of soot using an extended ReaxFF reactive force field. The Journal of Physical Chemistry A, 114(48):12561-12572.

[15]KamatS, KapaseP, JainP, et al., 2023. Modeling Virtual Sensor for Engine Nitrogen Oxides Using Variants of Artificial Neural Networks. SAE International, Warrendale, USA.

[16]KowalikM, AshrafC, DamirchiB, et al., 2019. Atomistic scale analysis of the carbonization process for C/H/O/N-based polymers with the ReaxFF reactive force field. The Journal of Physical Chemistry B, 123(25):5357-5367.

[17]KumarA, YashwanthVH, KumarR, et al., 2024. Machine Learning Approach to Control Thermal Strategies and Mitigate Sensor Failure Penalty on Emissions. SAE International, Warrendale, USA.

[18]Le CornecCMA, MoldenN, van ReeuwijkM, et al., 2020. Modelling of instantaneous emissions from diesel vehicles with a special focus on NO_x: insights from machine learning techniques. Science of the Total Environment, 737:139625.

[19]LiJ, ZhouQ, HeX, et al., 2023. Data-driven enabling technologies in soft sensors of modern internal combustion engines: perspectives. Energy, 272:127067.

[20]LiRZ, HerrerosJM, TsolakisA, et al., 2020. Machine learning regression based group contribution method for cetane and octane numbers prediction of pure fuel compounds and mixtures. Fuel, 280:118589.

[21]LiRZ, HerrerosJM, TsolakisA, et al., 2021. Machine learning-quantitative structure property relationship (ML-QSPR) method for fuel physicochemical properties prediction of multiple fuel types. Fuel, 304:121437.

[22]LiuJL, WangHF, 2022. Machine learning assisted modeling of mixing timescale for LES/PDF of high-Karlovitz turbulent premixed combustion. Combustion and Flame, 238:111895.

[23]LiuJL, HuangQ, UlishneyC, et al., 2022. Comparison of random forest and neural network in modeling the performance and emissions of a natural gas spark ignition engine. Journal of Energy Resources Technology, 144(3):032310.

[24]LiuL, ChenWL, ZhuQ, et al., 2023. Inhibitory mechanisms of ammonia addition on soot formation during n-decane pyrolysis. Fuel, 350:128695.

[25]MishraC, SubbaraoPMV, 2021. Design, development and testing a hybrid control model for RCCI engine using double Wiebe function and random forest machine learning. Control Engineering Practice, 113:104857.

[26]PetrucciL, RicciF, MarianiF, et al., 2020. Engine Knock Evaluation Using a Machine Learning Approach. SAE International, Warrendale, USA.

[27]SilvaM, MohanB, BadraJ, et al., 2023. DoE-ML guided optimization of an active pre-chamber geometry using CFD. International Journal of Engine Research, 24(7):2936-2948.

[28]SokR, JeyamoorthyA, KusakaJ, 2024. Novel virtual sensors development based on machine learning combined with convolutional neural-network image processing-translation for feedback control systems of internal combustion engines. Applied Energy, 365:123224.

[29]TangQX, WangMD, YouXQ, 2019. Effects of fuel structure on structural characteristics of soot aggregates. Combustion and Flame, 199:301-308.

[30]ThompsonAP, AktulgaHM, BergerR, et al., 2022. LAMMPS-a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Computer Physics Communications, 271:108171.

[31]TorregrosaAJ, BroatchA, OlmedaP, et al., 2021. Numerical Estimation of Wiebe Function Parameters Using Artificial Neural Networks in SI Engine. SAE International, Warrendale, USA.

[32]WangY, MaoQ, WangZY, et al., 2023. A ReaxFF molecular dynamics study of polycyclic aromatic hydrocarbon oxidation assisted by nitrogen oxides. Combustion and Flame, 248:112571.

[33]XingZH, ChenC, JiangX, 2023. A molecular investigation on the mechanism of co-pyrolysis of ammonia and biodiesel surrogates. Energy Conversion and Management, 289:117164.

[34]YanYC, XieTF, LiuJL, 2025. Rapid and accurate prediction of molecular dynamics simulations using physics-informed LSTM networks in engine emission analysis: a case study of C₃H₆/NH₃ pyrolysis for PAH formation. Journal of the Energy Institute, 120:102090.

[35]YaoS, WangB, KronenburgA, et al., 2020. Modeling of sub-grid conditional mixing statistics in turbulent sprays using machine learning methods. Physics of Fluids, 32(11):115124.

[36]YoonK, RahnamounA, SwettJL, et al., 2016. Atomistic-scale simulations of defect formation in graphene under noble gas ion irradiation. ACS Nano, 10(9):8376-8384.

[37]ZaherMH, ChuC, DadsetanM, et al., 2023. Experimental and numerical investigation of soot growth and inception in an ammonia-ethylene flame. Proceedings of the Combustion Institute, 39(1):929-937.

[38]ZhangK, XuYS, YuRH, et al., 2024. ReaxFF molecular dynamics study of N-containing PAHs formation in the pyrolysis of C₂H₄/NH₃ mixtures. Combustion and Flame, 270:113774.

[39]ZhangP, ZhangK, ChengXB, et al., 2022. Analysis of inhibitory mechanisms of ammonia addition on soot formation: a combined ReaxFF MD simulations and experimental study. Energy & Fuels, 36(19):12350-12364.

[40]ZhangP, WuH, ZhangK, et al., 2023. Decoupling effects of C₃H₃/C₄H₅/i-C₄H₅/CN radicals on the formation and growth of aromatics: a ReaxFF molecular dynamics study. Journal of Aerosol Science, 171:106185.

[41]ZhaoJ, LinYY, HuangK, et al., 2020. Study on soot evolution under different hydrogen addition conditions at high temperature by ReaxFF molecular dynamics. Fuel, 262:116677.

Similar articles