Thermal analysis of extended surfaces using deep neural networks

Abstract

The complexity of thermal analysis in practical systems has emerged as a subject of interest in various fields of science and engineering. Extended surfaces, commonly called fins, are crucial cooling and heating mechanisms in many applications, such as refrigerators and power plants. In this study, by using a deterministic approach, we discuss the thermal analysis of conduction, convection, and radiation in the presence of a magnetic force within an extended surface. The present study develops a deep neural network to analyze the mathematical model and to estimate the contributions of each dimensionless model parameter to the thermal dynamics of fins. The deep neural network used in this study makes use of a feedforward architecture in which the weights and biases are updated through backward propagation. The accuracy of the neural network model is validated using results obtained from a spectral-based linearization method. The efficiency rate of the extended surfaces is computed using the neural network and spectral methods. The results obtained demonstrate the accuracy of the neural network-based technique. The findings of this study in relation to the novel mathematical model reveal that utilizing materials with variable thermal conductivity enhances the efficiency rate of the extended surface.