4th China International Youth Conference on Electrical Engineering (CIYCEE 2023)

137 / 2023-10-19 12:08:06

Neural Network-Based Design-Oriented Thermal Model for Natural Cooling Fin Heatsinks

heatsink,thermal,thermal model

6.AI-driven technology > (3)、Advanced sensing and intelligent computation

Thermal management is an important part of power electronics converter design. Losses from power electronics converter are conveyed through media and finally dissipated into the air with, mostly, heatsink. Depending on the loss to dissipate, natural cooling or forced air cooling could be used. Traditionally, thermal designers use either empirical formulation or computational fluid dynamics software (CFD) to help heatsink design. However, this could be either ineffective or time-consuming. In this paper, we introduce the neural network tool. A three-layer neural network is used to predict the mean/max temperature on the heatsink based on input dimensional parameters. The average validation set error is 5.7% over a wide shape range, a significant improvement compared with the empirical or datasheet fitting method. Finally, we give two applications of the neural network. i) The thermal resistance boundaries are fitted for a given heatsink volume. Designers could easily utilize it to estimate how large a heatsink is required for their specific design.  ii), A simple thermal network considering heatsink thermal capacitance could be obtained from the neural network.