3D Double diffusive natural convection in an open pyramidal cavity: CFD study and machine learning prediction

The present study corresponds to a CFD based analysis of the three-dimensional double-diffusive natural convection in an open pyramidal cavity. The equations governing the fluid flow, heat and mass transfer and entropy generation within the cavity are solved using the Finite Element Method. In addition, machine learning (ML) models were developed to estimate the average Nusselt number Nu av , the average Sherwood number Sh av and the total generated entropy S tot as functions of governing parameters in order to enable fast and reliable prediction of the system's thermal and mass transport performance. Thus, the Random Forest regression model is trained on the CFD results as a surrogate predictor. The accuracy of the predictive model is excellent, ensuring R 2 > 0.96 for all output targets, and the main findings indicate that the optimal configuration occurs at high buoyancy ratio (N > 1) and wide cavity openings, which leads to enhanced heat and mass transfer. The minimal entropy generation occurs at N = –1, where the opposing thermal and solutal forces reduces the irreversibilities. Overall, it can be mentioned that the integration of machine-learning approach with CFD leads to robust predictive models allowing fast and accurate design of the double-diffusive convection in 3D open geometries. Increasing the cavity opening width L significantly enhances heat and mass transfer: for example, in a strongly opposing-buoyancy case (N = –2) Nu av more than doubles (from 3.8 to 8.1) and Sh av nearly quintuples (from 1.7 to 8.7) as L increases from 0.1 to 1.0, with similarly high absolute values achieved under aiding buoyancy (e.g., Nu av ≈8.5 and Sh av ≈12.1 at L = 1.0 for N = +2).