Machine learning generalization of lumped parameter models for the optimal cooling of embedded systems

Abstract

Smart Retrofitting is an emerging approach that transforms conventional devices into Cyber-physical systems, thereby allowing enterprises to adhere to Industry 4.0. This paper proposes a thermal design methodology for Smart Retrofitting by means of Machine Learning that can be employed in order to generalize the thermal behaviour of embedded heat sources. The entire underlying heat transfer physics is captured by means of temperature dependent characteristics. The total heat transfer rate is estimated based on the software instruction cycle, allowing an accurate selection of the cooling components. This process is formalized by focusing on tangible aspects. Experimental studies are conducted on laboratory prototypes. The novelty of the study lies in the integration of Machine Learning for supporting Lumped Parameter simulations.