Using Neural Networks for Thermal Analysis of Heat Conduction

Abstract

Predicting heat transfer mechanisms through solids and fluids is a continuously demanding research topic since accurate and fast temperature calculation is crucial in many engineering and industrial applications. This article presents a new model to calculate the temperature variation of solids and fluids instantly, in less than 0.04 s, for the whole simulation period based on a novel computational framework of deep learning. The partial differential equation, such as the heat transfer equation, can be solved directly at any point according to a well-known boundary condition point without the need for domain discretization. Therefore, instant and accurate temperature calculation is achieved with the minimum computational resources. The proposed deep learning model can be applied in many engineering applications and products by using it in online thermal monitoring or digital twin technology. The new model is well validated by comparing the temperature values obtained from the deep learning model with the experimental temperature measurements. Moreover, a computational cost comparison with other numerical models is conducted to prove the high efficiency of the proposed deep learning model.