Fast-track decarbonization of existing diesel engines via Hydrogen-Dodecane dual-fuel retrofit: A machine learning and combustion kinetics framework

This research introduces a hybrid modeling framework that combines intricate combustion dynamics with machine learning (ML) to facilitate the decarbonization and retrofitting of traditional diesel engines utilizing Hydrogen-Dodecane dual-fuel mixtures. A Cantera-based kinetic model was used to predict combustion stability, heat release, expansion power, efficiency, and emissions over a diverse array of fuel blends and injection timings. The findings indicated that injection angles duration (10°–15°) and moderate hydrogen enrichment produce maximum combustion performance, achieving a balance among ignition reliability, thermal efficiency, and emissions reduction. Optimal efficiency (56%) and power output (21.2 kW) were recorded at mid-range blending ratios under lean conditions, highlighting a significant interaction between fuel composition and combustion timing. Ensemble-based machine learning models were developed to enhance predictive capability by forecasting heat release and classifying combustion stability. Of the systems evaluated, the random forest regressor attained the highest predictive accuracy (R 2 = 0 . 895 ). The innovation of this work lies in the integration of high-fidelity kinetic modeling and machine learning for real-time predictive control of hydrogen-retrofitted engines, providing a technically sound approach to clean and adaptable decarbonized combustion systems. • Introduces a combined Cantera-based kinetic and machine learning framework for hydrogen-dodecane retrofitting. • Identifies optimal injection angles and hydrogen ratios for stable and efficient dual-fuel combustion. • Demonstrates high accuracy in predicting heat release and combustion stability using ensemble ML models. • Provides a scalable digital tool for real-time engine control and low-emission diesel retrofits.