Exergy and energy-based sustainability evaluation of diesel-biodiesel-ethanol blends with emission forecasting using advanced machine learning models

Abstract

The increasing influence of machine learning in engine emission prediction is on rising trend. The present study of thermodynamic analysis of ternary fuel with advanced Machine learning model provides valuable insights and adds significant outcomes to existing analysis. The current work deals with performance and sustainability of binary (diesel-biodiesel) and ternary (diesel-biodiesel-ethanol) fuel blends in a single-cylinder engine. Engine experiments were conducted using a structured design of experiments (DOE) approach, followed by thermodynamic analyses to evaluate key performance parameters, including exergy efficiency, brake thermal efficiency (BTE), and sustainability index. To optimize fuel parameters, the Desirability Function Approach (DFA) integrated with Response Surface Methodology (RSM) was employed. Additionally, advanced machine learning (ML) techniques were utilized to predict these performance characteristics. Notably, the binary blend demonstrated superior performance, achieving a 3.76 % higher BTE, 5.62 % higher exergy efficiency, and a 1.56 % increase in the sustainability index compared to conventional fuel. However, the inclusion of ethanol in the ternary blend (45 % Diesel–45 % Biodiesel–10 % Ethanol) resulted in a slight reduction in the sustainability index, which reached a peak value of 1.28 under full-load conditions. Interestingly, both sustainability index and exergy efficiency exhibited a consistent increase with rising engine load. At 5.2 kW, the blend BDE50 exhibits lower thermal efficiency than D100 and BDE10 by about 14.06 % and 7.36 %. Also, BDE50 blend exhibits lower exergy efficiency than D100 and BDE10 by about17.01 % and 11.66 % respectively. At full load, BDE50 blend possess 2.684 kW thermal loss and 18.583 kW exergy destruction, while BDE10 possess 2.331 kW thermal loss and 14.817 kW exergy destruction respectively. When comparing predictive models, the ML model demonstrated superior accuracy over RSM, as evidenced by higher R² values. Furthermore, desirability analysis confirmed the blends' strong performance and emission characteristics, achieving an optimal desirability rating of 0.777. Among the advanced ML models evaluated, XGBoost outperformed all others across multiple performance metrics, indicating its robustness in predicting fuel blend efficiency and sustainability.