ABSTRACT

The theoretical analysis of heat and fluid flow problems originated in the 19th century through the analytical solution of Navier-Stokes and energy equations, which was expanded by 20th-century’s empirical correlations based on fundamental dimensionless numbers. However, the rise of complex applications rendered these analytical and correlational approaches insufficient. While computational methods (CFD) emerged to handle intricate geometries, the increasing complexity of modern systems (such as 3D Lattice Metal Frames) leads to prohibitively long simulation run times. This dependency on expensive and case-by-case thermal analyses necessitates a paradigm shift. This work proposes that defining novel and complex dimensionless parameters, and integrating them with advanced machine learning techniques offers the necessary breakthrough.

The study investigates convective heat transfer in a channel containing 3D LMF structures. The Nusselt number and friction factor were computationally calculated for twelve different LMF geometries such as Pin Fin, Kagome, Octahedron, and Octet. An Artificial Neural Network (ANN) model was subsequently developed using MATLAB based on these numerical results. Crucially, the model employs only six dimensionless input parameters, the Reynolds number and five other novel dimensionless parameters to define the complete geometrical features of all structures.

The developed ANN correlation successfully and accurately predicts both the Nusselt number and friction factor for all studied 3D LMF structures. This outcome verifies the primary aim: the successful development of new and complex dimensionless parameters capable of generalizing geometrical complexity. By combining these parameters with the trained ANN model, the Nusselt number and friction factor values can be determined rapidly, providing a powerful, cost-effective ANN correlation that overcomes the time limitations of traditional numerical simulation.

BIOGRAPHICAL NOTE