Artificial intelligence has been widely applied in many scientific and engineering fields. Artificial neural networks (ANNs) show their advantages of fast and high accuracy for multiple input and multiple output predictions and optimizations. The computational costs are usually high when deal with conventional computational heat transfer problems. Hence deep learning based on artificial neural networks have been introduced to speed up the computations. Recently, a new architecture named Physics Informed Neural Networks (PINNs) are introduced to guarantee the conservation laws for filed predictions.  Automatic differentiation algorithm is applied to obtain each component of the thermal transport equations.  Thermal energy transport equation residues are included in the loss function during back propagation of the weights training. Examples for transient two dimensional thermal energy transport fields are shown for both predictions with ANNs and PINNs.  Both ANNs and PINNs can predict the transient temperature fields as well as the global Nusselt numbers. Comparing to the conventional ANNs, the average error for predictions of a two dimensional temperature fields with PINNs is 4.8% lower.

Dr. Su is currently an associate Professor of Electromechanical Engineering in University of Macau. She received his PhD degree from the Department of Mechanical Engineering from University of Minnesota in 2006.  In year 2000 and 2003, she obtained her Bachelor of Thermal Engineering from Tsinghua University, and Master of Mechanical Engineering from Hong Kong University of Science and Technology, respectively. Dr. Su’s research interests include Thermal Dispersion in Porous Medium, Renewable Energy Systems, Oscillating Flows, and Indoor Air Quality. She also serves as the director of the solar energy laboratory of University of Macau and associate editors of Journal of Heat Transfer (ASME) and Case Studies in Thermal Engineering (Elsevier).