Development of machine learning algorithm for assessment of heat transfer of ternary hybrid nanofluid flow towards three different geometries: Case of artificial neural network

The focus of this paper revolves around the examination of flow of ternary hybrid nanofluid, specifically the Al(2)O(3)–Cu-CNT/water mixture, with buoyancy effect, across three distinct geometries: a wedge, a flat plate, and a cone. The study takes into account the presence of quadratic thermal radiation and heat source/sink of non-uniform nature. To develop the model, the Cattaneo–Christov theory is utilized. The equations governing the flow are solved by applying similarity transformations and employing the "bvp4c function in MATLAB” for numerical analysis and solution. Conventional methods for conducting parametric studies often face challenges in producing significant conclusions owing to the inherent complex form of the model and the method involved. To address the aforementioned issue, this paper explores the potential of machine learning methods to foresee the conduct of the flow characterized by multiple interconnected parameters. By utilizing simulated data, an artificial neural network is trained using the Levenberg-Marquardt algorithm to learn and comprehend the underlying patterns. Subsequently, the trained neural network is employed to estimate the Nusselt number on the surfaces of all three geometries. This approach offers a promising alternative to traditional parametric studies, enabling more precise predictions and insights into the behavior of complex systems. The Nusselt number is highest for THNF flow over the cone. The mean squared error (MSE) values for the ANN algorithm, across all analyzed cases, range from 0 to 0.03972. The findings contribute to an improved understanding of the characteristics and dynamics of ternary hybrid nanofluid flow in various geometries, assisting in the design and optimization of heat transfer systems involving such fluids.