November 30, 2025Open Access

AI-driven performance and emission forecasting in diesel–hydrogen dual-fuel engines: a machine learning approach

CBChiranjit Bhowmik MDMadhujit Deb ATAnirban Tarafdar

Key Points

Performance improved by 23% in diesel-hydrogen engine at full load using optimized injection strategies.
Machine learning models, including ensemble learning, achieved R-squared of 0.98 in predicting emissions.
Gradient boosting and random forest models were effective in forecasting outcomes in diesel-hydrogen dual-fuel engines.
Emissions compliance remains a key challenge alongside performance enhancement in hydrogen-based systems.

Abstract

Abstract Environmental concerns and the rising demand for fossil fuels have encouraged the exploration of alternative fuels like hydrogen. Hydrogen, with its clean-burning properties, can improve engine performance and meet strict emission standards when used in a dual-fuel mode with diesel. In this study, a four-stroke single-cylinder diesel engine (SCDE) was experimentally tested under diesel–hydrogen dual-fuel operation at 20–100% load conditions. The results showed that the diesel–H 2 strategy at 7500 μs (DH3) injection strategy demonstrated a 23% improvement in brake thermal efficiency (BTE) compared to baseline diesel operation at full load, though oxides of nitrogen (NOx) emissions peaked at 7.5 g kW -1 h -1 at full load under the DH3 approach. Soot emissions are minimal at low loads but increase sharply at higher loads, particularly for DH3. Unburned hydrocarbon (UHC) emissions are higher for hydrogen-enriched strategies compared to pure diesel under low-load conditions, with diesel–H 2 strategy at 5500 μs (DH1) and diesel–H 2 strategy at 6500 μs (DH2) showing peaks at 29 g kW -1 h -1 and 22 gkW -1 h -1 , respectively. The diesel–H 2 strategy at 8500 μs (DH4) achieved a maximum 85% reduction in UHC emissions and significantly minimized soot formation under high-load conditions. However, they also present challenges in terms of NOx emissions, particularly at higher loads. Therefore, optimizing injection strategies is critical to achieving a balance between enhanced performance and emissions compliance for sustainable engine operation. Furthermore, eight machine learning (ML) models—decision trees (DT), random forest (RF), support vector regression (SVR), gradient boosting (GB), AdaBoost (ADA), linear regression (LR), K-nearest neighbors (KNN) and ensemble learning (ENSEMBLE)—were employed for predictive analysis of performance and emissions. The ENSEMBLE model achieved the best prediction accuracy ( R -squared ( R 2 ) = 0.98, lowest root mean square error (RMSE) = 0.012, mean absolute error (MAE) = 0.009), confirming its robustness and generalization capability in dual-fuel engine prediction.

Ask AI

Helpful

Bookmark

View Full Paper