A hybrid machine learning-NSGA-II-TOPSIS framework for multi-objective optimization of CI engine performance and emissions using ternary biodiesel-butanol-diesel blends

The rising global energy demand and strict emission rules have accelerated the investigation of sustainable alternative fuels for transportation. Biodiesel-alcohol blends exhibit significant potential due to their inherent oxygen content. However, their application results in highly nonlinear combustion characteristics and complex trade-offs between engine performance and emissions. To address these challenges, this study integrates machine learning-based modeling with multi-objective optimization. Experimental investigations were conducted on a CI engine powered with diesel/cottonseed biodiesel/n-butanol blends under various engine operating conditions. Based on the experimental dataset, three tree-based ensemble ML models (Random Forest (RF), Extra Tree (ET), and Gradient Boosting (GB)) were developed to predict engine performance and emissions characteristics. Among all models, the GB model exhibited superior predictive performance, with high accuracy as R 2 of 0.951 for BTE, 0.975 for BSFC and more than 0.84 for all emission parameters with an overall MAPE of 10.3%. A hybrid framework integrating the NSGA-II and TOPSIS was utilized for multi-objective optimization. The optimal solution was obtained at 88.3% engine load, fuel type 5 (D40B40Bu20), and fuel injection timing of 27 °bTDC, corresponding to a BTE of 36.84%, BSFC of 0.286 kg/kWh, CO of 0.048% vol, CO 2 of 2.01% vol, HC of 17.64 ppm and NOx of 709.85 ppm. Experimental validation showed good agreement with model predictions, with deviations within ±7% for all performance and emission parameters. The results confirm that the integration of ML with NSGA-II and TOPSIS offers a robust and efficient framework for multi-objective optimization for CI engines powered with oxygenated ternary blends.