Machine learning and experimental emission assessment in high temperature air premixed charged compression ignition engines using the Pugh matrix

This research explores the performance, exhaust emissions, and combustion properties of a Premixed Charged Compression Ignition (PCCI) engine using combinations of Andropogon narudus (AN) and Sapota Oil Methyl Ester (SOME) blended as substitute fuels. A split-fuel injection system was used, supplying 70% of the fuel with direct injection and 30% through the intake air manifold. The test fuels considered were D100 (commercial diesel), AN20 + D80, SOME20 + D80, and their corresponding mixture with nano-additives CeO₂ and Al₂O₃ (10 ppm).Performance analysis showed that the highest brake thermal efficiency (BTE) was attained by the SOME20 + D80 mixture with Al₂O₃, which rose by 2.5% with respect to diesel, and AN20 + D80 with Al₂O₃ with a 2.3% rise in BTE. Brake-specific fuel consumption (BSFC) was reduced by 0.10 g/kWh for AN20 + D80 with Al₂O₃ with respect to diesel due to its lower viscosity. Emission analysis showed a hydrocarbon (HC) emission decrease of up to 7 ppm for all the blends tested, although CO₂ and NO_x emissions were higher in AN and SOME fuel blends with nano-additives. Combustion studies revealed that AN20 + D80 with Al₂O₃ had the maximum peak pressure and net heat release rate, which supports the effect of fuel properties on combustion behavior. For improving predictive accuracy, machine learning-augmented modeling was utilized using Multiple Linear Regression (MLR), Random Forest Regression (RF), and Support Vector Machine Regression (SVMR). The RF model performed better with predictive efficiency, giving R² = 0.97 for NO_x, 0.99 for Smoke, and 0.95 for CO, capturing nonlinear relationships well. MLR had good fits for BTE (R² = 0.99) and BSFC (R² = 0.94), whereas SVMR had poorer predictions (e.g., Smoke: R² = 0.19, CO₂: R² = 0.30).A sustainability ranking situated AN20 + D80 at the most viable biofuel position, particularly with the addition of Al₂O₃. Predictive analytics derived from ML in the study focus on the role of achieving maximal alternative fuel mixtures, less reliance on huge experimental trials, and more cleaner and efficient burning systems.