DRL for conjugate heat transfer

In this study, we gauge the ability of deep reinforcement learning (DRL) techniques to assist the control of conjugate heat transfer systems. We make use of a single-step proximal policy optimization algorithm to optimize different set-ups of natural and forced convection.

In the above configuration, we consider the cooling of a workpiece using 3 injectors. Their positions are optimized following different strategies to obtain minimal thermal gradients.

Here, a similar 3D configuration is considered. Optimal configurations are obtained within 40 to 60 episodes, depending on the considered strategy.

Related publications

Deep reinforcement learning for the control of conjugate heat transfer with application to workpiece cooling

E. Hachem, H. Ghraieb, J. Viquerat, A. Larcher, P. Meliga

Submitted to Nature Machine Intelligence