Intelligent back-propagated neural networks to study nonlinear heat transfer in tangent-hyperbolic fluids

Abstract

This paper presents a novel way to analyze unsteady flow and heat transfer in Magnetohydrodynamic Tangent-Hyperbolic Fluid (MHD-THF) flow over a stretching sheet that uses Recurrent Neural Networks (RNNs) optimized with the Levenberg-Marquardt Algorithm (LMA). The controlling PDEs are similarly converted into nonlinear ordinary differential equations (ODEs), and reference data is generated using the Adam-Bashforth numerical solver. The reference dataset, divided into 70 % training, 15 % testing, and 15 % validation sets, which was separated into three distinct cases for each scenario, was trained, tested, and validated using MATLAB's RNN-LMA technique. The absolute error for each scenario of model instance is approximately 10−06, 10−05, 10−07, 10−05, and 10−08. The numerical solutions for the nonlinear fluid dynamics system were considered to reduce mean square error. The stochastic technique's reliability and competency are demonstrated using comparison configurations of MSE, error histograms, correlation, and regression. A graphical assessment of the model shows that power-law index, unsteadiness, Hartmann number, and Weissenberg number decrease velocity profiles. Meanwhile, the temperature profile reveals an increase with the Eckert number and unsteadiness, whereas the Prandtl number has the reverse impact. This study illustrates the potential of RNN-LMA techniques to handle complicated fluid dynamics problems with more accuracy and efficiency.

Keywords

• Tangent hyperbolic fluid
• Magnetohydrodynamic
• Nanofluids
• Recurrent neural networks
• Levenberg marquardt algorithm
• Adams numerical solver