Electrothermal convection from a hexagonal body: LB simulations and machine learning predictions

The present work focuses on the interaction between electric body forces and buoyancy driven motion in a square enclosure housing a centrally aligned hexagonal body. Electro thermo hydrodynamic simulations are carried out using the Lattice Boltzmann Method across a range of T values from 100 to 600 and P r values from 4 to 10. Three geometric orientations, namely 0 ° , 15 ° , and 30 ° , are examined. The results indicate that an increase in T amplifies convective activity and promotes charge separation, with unsteady flow appearing beyond T = 400 . At P r = 10 , the N u ¯ M increases from 2.14% at T = 300 to about 111% at T = 400 relative to T = 100 , whereas at P r = 4 the enhancement remains limited to 27%–43%, underscoring the strong P r control on electrothermal heat transfer. Low P r values lead to early instability, while high P r values help maintain steady behaviour for extended ranges. Orientation influences local thermal gradients and flow structure but contributes less compared to the other two parameters. Several machine learning models are trained on simulation data to estimate the mean Nusselt number, N u ¯ M . Among them, AdaBoost and a meta model deliver the most reliable predictions over different test proportions, offering both accuracy and consistency. The SHAP based analysis reveals that T governs most of the model’s output, followed by P r , while the orientation angle shows limited impact. The scatter plots display a strong agreement between actual and predicted values, with most data points confined within a ± 10% error margin. Further, the LGBM-Optuna framework, trained for temperature and charge contour predictions, generates accurate maps within seconds, offering a drastic reduction in computation time compared to LBM simulations. This capability provides a practical surrogate for rapid parametric assessment of T , P r , and θ , and accelerates thermal system optimization. • Study explores electrothermal convection in an enclosure with hexagonal body. • Electrothermal forces trigger vortices and unsteady flow beyond T = 400. • Lower Pr accelerates transient onset; higher Pr sustains steady convection. • ML models predict flow and heat transport trends with strong accuracy. • Optuna-tuned models capture temperature and charge contours with reliability.