Support vector regression method for predicting temperatures of heat sources cooled by forced convection in a horizontal channel

• Published in: Thermal science and engineering progress 2020-12, Vol.20, p.100725, Article 100725
• Main Authors: Durgam, Shankar, Bhosale, Ajinkya, Bhosale, Vivek, Deshpande, Revati, Sutar, Pankaj
• Format: Article
• Language: English
• Subjects: Forced convection; Heat sources; Numerical simulation; Optimal configuration; Support Vector Regression
• ISSN: 2451-9049; 2451-9049
• DOI: 10.1016/j.tsep.2020.100725

•SVR technique is used to predict temperature of heat sources to identify optimal configuration.•To investigate the suitability of problem for small size of substrate board subjected to high heat flux electronic package system.•Full three-dimensional simulations have been carried out using COMSOL Multiphysics 5.4 to study heat transfer and fluid flow characteristics.•Computer simulation of optimal configuration and validating results obtained by SVR.•Using three dummy components along with six heat sources for heat transfer enhancement. This paper presents the novel concept of Support Vector Regression method for predicting the temperatures of rectangular heat sources with dummy components cooled by forced convection in a horizontal channel. The substrate board considered for study is FR4 equipped with silicon heat sources resembles IC chips on printed circuit board. Six heat sources with three dummy components of four different sizes are considered for study. Three dimensional steady state numerical simulations have been performed using finite element method based commercially available software COMSOL Multiphysics 5.4. Laminar non-isothermal fluid flow with conjugate heat transfer module is used to study heat transfer in solids and fluid region. Air cooling of heat sources with dummy components is carried out to study heat transfer characteristics. Flowing air velocity of 2.5 m/s with uniform heat flux value of 5000 W/m2 is considered for study. Support Vector Regression programming is performed using Python programming language to predict temperatures of heat sources to explore optimal configuration. Random hundred different configurations were studied out of which 80% were used in training and 20% used for validation. Results show that the simulated and predicted temperature values are in agreements of ⩽10 %.