Optimization of chimney design for an extruded radial heat sink using experiments and neural network models

Abstract

This study investigates chimney-assisted natural-convection cooling for an Extruded Radial Heat Sink (ERHS) in high-power Light Emitting Diode (LED) applications. Although chimneys have been applied to Forged Radial Heat Sinks (FRHSs), their application to ERHSs remains insufficiently studied. Direct attachment of a chimney to the flat top of the ERHS obstructs the internal airflow path, requiring a geometry tailored to the ERHS. Accordingly, the effects of chimney geometry and inlet shape were investigated through experiments and steady-state CFD simulations. The sharp-edged inlet induces flow separation and a vena contracta, whereas the filleted inlet is introduced to reduce this inlet loss and maintain attached upward flow. An Artificial Neural Network (ANN) surrogate model with five inputs was trained to predict thermal resistance ( R th ) as its sole output, capturing the nonlinear coupling among the geometric variables with maximum relative prediction errors of 2.4% for the ERHS with Sharp-Edged Chimney (ERHS-SEC) and 2.3% for the ERHS with Filleted Chimney (ERHS-FC). Multi-objective optimization showed that the ERHS-FC Pareto front was consistently shifted toward lower thermal resistance ( R th ) and smaller chimney volume than the ERHS-SEC front. At T hs − T amb = 20 K, the minimum- R th ERHS-FC design reduced thermal resistance by 27.4% relative to the heat-sink-only reference. A separate compact ERHS-FC design maintained a 3.1% thermal benefit while reducing chimney volume by approximately 50% relative to the reference chimney model.