Experimental investigation of emissions and machine learning–based prediction of exhaust temperature and heat rate in a hydrogen-enriched heavy-duty internal combustion engine

Exhaust gas recirculation (EGR) and hydrogen enrichment are promising approaches for improving the performance and environmental sustainability of compressed natural gas (CNG) engines. This study experimentally investigates the combined effects of hydrogen fraction, EGR ratio, spark timing, engine load, and engine speed on the exhaust characteristics and emissions of a stoichiometric hydrogen-enriched compressed natural gas (HCNG) engine. Tests were performed using hydrogen fractions of 0–20 %, EGR ratios of 0–4.5 %, spark timings of 8°–59 o CA bTDC, engine loads of 25 %, 50 %, and 75 %, and engine speeds of 900, 1200, and 1500 rpm. The experimental results show that hydrogen enrichment enhances combustion, resulting in higher exhaust temperature and exhaust heat rate. At 20 % hydrogen and 18 °CA bTDC, the maximum exhaust temperature and exhaust heat rate increased by 3.69 % and 15.27 % respectively compared with CNG operation. In contrast, increasing the EGR ratio to 4.5 % reduced the peak exhaust temperature by 1.18 % and the exhaust heat rate by 3.68 % because of charge dilution and lower combustion temperatures. Hydrogen enrichment also reduced CO, CO 2 , and THC emissions by promoting more complete combustion, while EGR effectively suppressed NO x formation through a decrease in peak flame temperature. To complement the experiments, five machine learning models—Interaction Linear (IL), Medium Tree (MT), Medium Gaussian Support Vector Machine (MGSVM), Bagged Tree (BT), and Exponential Gaussian Process Regression (EGPR)—were developed to predict exhaust temperature and exhaust heat rate using different input combinations and testing data ratios. Model performance was assessed using the coefficient of determination (R 2 ), root mean square error (RMSE), mean absolute error (MAE), and training time. Among all models, EGPR demonstrated the highest prediction accuracy, achieving RMSE values of 0.7874 for exhaust temperature and 0.3394 for exhaust heat rate with a 20 % test dataset. The developed predictive model provides an efficient framework for estimating HCNG engine exhaust characteristics and can support future engine calibration, intelligent control strategies, and the design of waste-heat recovery and hydrogen-production systems. These findings contribute valuable experimental data and predictive capabilities for improving the efficiency and environmental performance of HCNG engines.