https://doi.org/10.65770/LULX4034
ABSTRACT
Background: The global imperative to simultaneously reduce the environmental burden of heavy-duty diesel engines and preserve their fundamental economic utility—powering freight, construction, agriculture, and power generation—has driven an accelerating research investment in machine learning (ML) applications for emissions control. ML techniques offer a fundamentally different paradigm for engine management: rather than pre-programming static relationships between operating parameters and emissions outputs, ML systems learn these relationships from data, adapt to changing conditions, and can simultaneously optimize objectives that are in fundamental tension within conventional calibration frameworks. The pace of regulatory tightening, exemplified by the U.S. EPA Tier 4 Final, Euro VI, and the anticipated 2030 ultra-low NOx standards, has made this shift from static to adaptive intelligence not merely academically interesting but commercially and environmentally essential. Scope and Methodology: This systematic review synthesizes peer-reviewed research published between 2010 and 2024, identified through a PRISMA-adapted screening process of 1,842 records across five major academic databases. A final corpus of 347 publications satisfying pre-defined inclusion criteria—peer-reviewed, quantitative performance results, heavy-duty diesel focus exceeding 75 horsepower, English-language—was analyzed. Publications were classified by ML methodology category (supervised learning, physics-informed hybrid approaches, reinforcement learning, and unsupervised learning) and by application domain (NOx prediction, PM estimation, multi-pollutant trade-off optimization, real-time adaptive control, aftertreatment system coordination, calibration automation, and predictive maintenance). Key Findings: The reviewed literature reveals four generational phases of ML adoption in diesel emissions control, from simple regression-based offline calibration tools to sophisticated real-time physics-informed adaptive systems. The most significant performance advances have been achieved by hybrid physics-ML architectures, which consistently outperform purely data-driven approaches on the out-of-distribution conditions most critical for regulatory compliance—by up to 7 percentage points in transient NOx prediction accuracy. Deep learning-based model predictive control frameworks have demonstrated 15–25% NOx reduction and 8–15% fuel economy improvement in production-validated implementations, including on Cummins diesel engines. Physics-informed neural networks emerge as the architectural direction with the strongest combined case for accuracy, extrapolation fidelity, and regulatory interpretability. Cross-domain evidence from IoT-based environmental monitoring, AI-augmented engineering decision systems, and data-driven energy management frameworks reinforces the structural validity of the adaptive ML approaches applied in the diesel engine context. Recent advances in data-driven bioenergy feedstock optimization for switchgrass and cellulosic biofuels represent an emerging complement to diesel ML systems, as the fuel quality characteristics of bioenergy-derived products directly influence the adaptive correction demands placed on injection control architectures.
Conclusions: Despite demonstrated production-level progress, substantial challenges persist in functional safety certification, regulatory framework adaptation, embedded computational feasibility, and long-term model robustness. Federated fleet learning, edge AI hardware, digital twin integration, and biofuel-adaptive ML are identified as the four highest-priority research directions with the clearest pathways to transformative commercial impact. Practitioners are provided with evidence-based algorithm selection guidance and phased implementation recommendations derived from the systematic analysis of the reviewed corpus.
References
- [1] International Council on Clean Transportation (ICCT), “Trucking’s Climate Imperative,” ICCT, Washington, D.C., 2023.
- [2] S. Environmental Protection Agency, “Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990–2020,” EPA 430-R-22-003, 2022.
- [3] Johnson, T.V., “Review of vehicular emissions trends,” SAE Int. J. Engines, vol. 8, no. 3, pp. 1152–1167, 2015. https://doi.org/10.4271/2015-01-0993
- [4] Heywood, J.B., Internal Combustion Engine Fundamentals, 2nd ed. New York: McGraw-Hill Education, 2018.
- [5] Reitz, R.D. et al., “IJER editorial: The future of the internal combustion engine,” Int. J. Engine Res., vol. 21, no. 1, pp. 3–10, 2020.
- [6] Lapuerta, M. et al., “Effect of biodiesel fuels on diesel engine emissions,” Prog. Energy Combust. Sci., vol. 34, no. 2, pp. 198–223, 2008.
- [7] Bosch, R. et al., Automotive Handbook, 9th ed. Stuttgart: Bentley Publishers, 2014.
- [8] Ewim, D.R.E. et al., “A quick review of the applications of artificial neural networks (ANN) in the modelling of thermal systems,” Paulus Editora, 2021.
- [9] Norouzi, A. et al., “Deep learning based model predictive control for compression ignition engines,” arXiv:2204.00139, 2022.
- Mohammad, A. et al., “Physical-oriented and machine learning-based emission modeling in a diesel compression ignition engine: Dimensionality reduction and regression,” Int. J. Engine Res., 2023. DOI: 10.1177/14680874211070736
- Kumar, S. et al., “Machine learning based combustion model for IC engine using in-cylinder pressure,” Fuel, vol. 280, 118610, 2020.
- Shin, B. et al., “NOx abatement and fuel economy improvement in diesel engine using intelligent dual-loop EGR and VNT control,” SAE Int. J. Engines, vol. 1, no. 1, pp. 1454–1465, 2009.
- Bradford, E. et al., “Efficient multiobjective optimization employing Gaussian processes, spectral sampling and a genetic algorithm,” J. Glob. Optim., vol. 71, no. 2, pp. 407–438, 2018.
- Willems, F. et al., “Is cylinder pressure-based control required to meet future HD legislation?,” SAE Int. J. Engines, vol. 4, no. 1, pp. 1107–1118, 2011.
- Raissi, M. et al., “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear PDEs,” J. Comput. Phys., vol. 378, pp. 686–707, 2019.
- Adesiyan, Y.R. and Alaba, P.A., “Critical review of data-driven breeding and selection on field-grown switchgrass (Panicum virgatum L.) as a bioenergy feedstock,” Appl. Chem. Eng., vol. 8, no. 1, 2025. https://doi.org/10.59429/ace.v8i1.5596
- Adesiyan, Y.R. and Alaba, P.A., “Field-Grown Switchgrass (Panicum virgatum L.) as a Bioenergy Feedstock: A Review of Phenotype Data Collection and Analysis,” J. Biotechnol. Biores., vol. 5, no. 4, JBB.000618, 2024. https://doi.org/10.31031/JBB.2024.05.000618
Download all article in PDF
