Leveraging Advanced Machine Learning Algorithms for Energy Efficiency and Carbon Footprint Reduction in Diesel Engines

In this study, machine learning-based models were employed to estimate brake-specific fuel consumption (BSFC) using the data derived from 300 experimental engine tests conducted under varying loads, EGR ratios, and fuel combinations. Eleven alternative algorithms, including AdaBoost, Gradient Boosting, Random Forest, SVM, and Neural Network, were evaluated using datasets derived from real engine data. Statistical tools such as the R² coefficient and Mean Square Error (MSE) were utilized to assess the performance of the models. The best results were derived using Gradient Boosting (R² = 0.9998), AdaBoost (R² = 0.9998), and Random Forest (R² = 0.9997). In the prediction process of the Gradient Boosting model, the minimum absolute error was found to be 0.0000000, and the maximum absolute error was found to be 0.000002. The model's prediction accuracy ranges from 98.30% to 99.39%. This degree of accuracy demonstrates that fuel consumption can be predicted from engine data alone, without installing fuel-measurement equipment. The model's forecasting performance has been examined by analyzing the most and least successful examples. The results simplify fuel-consumption measurement procedures, provide the infrastructure for AI-supported test systems, and reduce time and labor. In this respect, the study significantly advances the digitization of engine testing in both academia and industry.