A Physics-Informed CFD–Machine Learning Framework for Piston Bowl Optimization Toward Enhanced Tumble in a Small Diesel Engine

Strong in-cylinder tumble motion enhances turbulent mixing and combustion efficiency in compression ignition engines. However, the geometry–tumble relationship remains poorly quantified, particularly in small agricultural diesel engines where design is largely empirical. This study introduces a physics-informed, data-driven framework for combustion chamber optimization in a Vikyno RV165-2 diesel engine operating within a speed range of 1600–2200 rpm, with explicit emphasis on tumble enhancement. A parametric design space defined by six non-conventional geometric variables was systematically explored through 100 high-fidelity CFD simulations using ANSYS ICE. The resulting dataset was used to train a machine-learning-based surrogate model, enabling rapid prediction of the tumble ratio and efficient identification of the optimal design. The results indicate that the optimized configuration (A = 59 mm, B = 35 mm, H = 14.38 mm, L = 8 mm, X = 10 mm, Y = 10 mm) achieved a 56.81% increase in average tumble ratio relative to the baseline. Flow field analysis reveals that this improvement is driven by enhanced flow redirection toward the piston bowl and intensified large-scale vortex structures during the late compression stroke. The proposed CFD–machine learning framework provides an efficient strategy for combustion chamber design, improving geometry–flow coupling and air–fuel mixing.