Ensemble Learning for Predicting Temperature of Heat Sources for Minimizing Electronic Failures

Abstract

This paper uses machine learning algorithms as a viable alternative for simulation software for the electronic cooling problem to select a suitable configuration and minimize failures due to improper heat dissipation. A combination of machine learning algorithms such as XG Boost, Support Vector Regression, Random Forest Regression is used temperatures prediction of silicon heat source which resembles to IC chips on FR4 substrate board for cooling under horizontal forced convection. Three-dimensional steady state conjugate heat transfer module with non-isothermal fluid flow is selected from COMSOL Multiphysics 5.4 to study heat transfer and fluid flow. For cooling constant heat flux of 5000 W/m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , 2.5 m/s air velocity is selected. A total of 100 simulations have performed, from which 80 configurations are used for training and rest 20 are used for testing. Python language is used for programming for Ensemble Learning. Results have shown that the errors obtained between predicted and simulated values are below 15%.