Experimental and machine learning-assisted optimization of diesel engines using swirl-generating piston designs and operating variables

Abstract

This study explores the influence of swirl-inducing piston modifications and key engine operating parameters on the performance characteristics of a diesel engine. Experiments were carried out using a single-cylinder direct injection (DI) diesel engine with a fixed compression ratio of 17.5. The investigation covered a range of injection pressures from 210 to 270 bar in 30-bar increments and injection timings from 19° before top dead center (bTDC) to 27°bTDC in 4° intervals and three custom-designed piston heads incorporating three, six, and nine swirl grooves were employed. To enhance in-cylinder air motion, three custom-designed piston heads incorporating three, six, and nine swirl grooves were employed. A Design of Experiments (DOE) methodology based on Response Surface Modeling (RSM) was applied to evaluate the statistical relationships among the input variables and engine responses. To complement and validate the RSM-based findings, several supervised machine learning algorithms namely Linear Regression, AdaBoost, Huber Regression, and XGBoost were implemented to predict performance and emission metrics. Among these, XGBoost exhibited superior predictive capability, yielding a low Mean Squared Error (MSE) of 0.288, a Root Mean Squared Error (RMSE) of 0.537, and a Mean Absolute Error (MAE) of 0.433. Notably, the configuration with five grooves and an injection pressure of 245.22 bar resulted in the most efficient combustion and the lowest hydrocarbon (HC) emissions. Additionally, a desirability-based multi-objective optimization approach was employed to identify the optimal combination of parameters. The integrated use of experimental testing and predictive modeling offers a reliable framework for engine performance optimization and provides valuable insights for enhancing combustion efficiency in diesel engines.