TITLE:
Machine Learning-Driven Prediction and Analysis of Magnetohydrodynamic (MHD) Natural Convection in Nanofluid-Filled Trapezoidal Cavities with Variable Obstacle Shapes, Wall Corrugations, and Inclination Angles
AUTHORS:
Sree Pradip Kumer Sarker, Md. Mahmud Alam
KEYWORDS:
Magnetohydrodynamic (MHD) Convection, Nanofluid, Machine Learning Prediction, Trapezoidal Cavity, Thermal Performance Analysis
JOURNAL NAME:
Advances in Nanoparticles,
Vol.14 No.4,
November
13,
2025
ABSTRACT: This study proposes a machine learning-enhanced framework to predict and analyze magnetohydrodynamic (MHD) natural convection within nanofluid-filled trapezoidal cavities featuring variable obstacle shapes, wall corrugations, and inclination angles. A comprehensive dataset was generated through Finite Element Method (FEM) simulations covering a wide parametric range, including Rayleigh numbers (103 - 106) and Hartmann numbers (0 - 50), for Cu-H2O nanofluids. Key thermal performance metrics Nusselt number (Nu), entropy generation (ST), and Ecological Coefficient of Performance (ECOP), were extracted and used to train supervised machine learning models: Support Vector Regression (SVR), Decision Tree (DT), and Random Forest (RF). Among them, the RF model achieved superior performance, yielding R2 scores of 0.991 (Nu), 0.982 (St), and 0.989 (ECOP), with mean prediction errors under 1.5%. The results show excellent agreement between FEM and ML outputs across diverse configurations, including different obstacle geometries (star, square, triangular), wall undulations (sinusoidal, square, triangular), and inclination angles (15˚, 30˚, 45˚). The integration of ML significantly reduced computational cost while preserving high accuracy, thus demonstrating its viability for rapid prediction and optimization. This hybrid FEM-ML methodology offers a powerful tool for real-time thermal system analysis and design in advanced MHD nanofluid applications.