Data-driven optimization of turbulent kinetic energy and tumble-y in combustion engines: A comparative study of machine learning models

• Combustion process was optimized by AI-ML techniques with CFD. • AI-ML predictions were evaluated against measured data. • The target optimization was assessed using the Euclidean distance. • The optimization parameters enhanced efficient combustion process. This paper presents an innovative approach to optimising the cold flow dynamics in combustion engines by integrating machine learning (ML) techniques with computational fluid dynamics (CFD). The research focuses on predicting and optimising critical pre-combustion parameters, such as turbulence kinetic energy (TKE) and tumble-y, which are pivotal for enhancing the air–fuel mixing during the intake and compression phases. Three ML models, Random Forest Regression (RFR), Gaussian Process Regression (GPR), and Neural Networks (NN), are evaluated for their predictive capabilities. The GPR model outperforms the others, demonstrating superior accuracy and reduced uncertainty, as highlighted by metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), Pearson Coefficient (PC), and R-squared (R 2 ). Additionally, the ML-based approach achieves a remarkable 21.6x speedup compared with traditional CFD solvers, significantly reducing the computational costs while maintaining high fidelity in capturing momentum and thermal characteristics. The optimization results underscore the critical role of TKE and tumble-y in creating favourable conditions for efficient combustion. For instance, as demonstrated in Design #1 (TKE = 396.56 J/kg, Tumble-y = -0.1535, Temp. = 846.42 K, Pres. = 1.52 bar) and Design #2 (TKE = 366.77 J/kg, Tumble-y = -0.1535, Temp. = 549.59 K, Pres. = 2.81 bar), higher TKE and optimized tumble-y values enhance air motion dynamics, promoting better fuel–air mixing and thermal performance. The rigorous assessment of optimization results using the Euclidean distance as a fitness function validates the reliability of the predictions and highlights the potential of ML models for efficient, scalable and cost-effective design exploration. Therefore, the present work provides a beneficial relationship between CFD simulation and experimental findings on cold flow dynamics and how these might play a leading role in pre-combustion process. Results provide a frame-shifting pathway toward optimization of engine design for the improvement of thermal efficiency, and meeting sustainability targets.