Aplicação de inteligência computacional na previsão do resfriamento de componentes eletroeletrônicos em placas de circuito

Abstract

The increasing miniaturization and rising power density of electronic components have intensified the challenges associated with thermal management in printed circuit boards. Traditional numerical methods, although accurate, often demand high computational cost and long processing time in analyses involving conjugate heat transfer. Thus, computational intelligence techniques emerge as effective alternatives for predicting the thermal behavior of such systems. In this work, Artificial Neural Networks were applied to predict the temperatures associated with the conjugate cooling of discrete heaters mounted on printed circuit board of channels. The database used was obtained from the invariant descriptor which represents the conjugate phenomenon through the thermal influence coefficients grouped into a conjugate matrix. Two Supervised Neural Network Architectures were implemented: the Multilayer Perceptron (MLP) and the Radial Basis Function (RBF) Network, both trained and evaluated using error metrics. The input variables considered were the Reynolds number, the relative heater height (with respect to the channel, and the relative thermal conductivity between the substrate and the fluid. The output consisted of the temperatures of the heaters, which remained within an average percentage error range of 25%. The results obtained through the Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics demonstrated the effectiveness of the proposed models, indicating that Artificial Neural Networks represent a viable approach to complement traditional numerical methods in conjugate forced convection cooling analysis.