Optimization of microchannel heat sinks with flexible vortex generators using GWO-SVR: A CFD and machine learning approach

This study presents an innovative approach incorporating flexible vortex generator clusters along microchannel sidewalls while utilizing computational fluid dynamics to investigate thermal dissipation behavior. We developed an advanced predictive framework-combining GWO with Support Vector Regression (SVR)-to accurately model heat transfer performance in microchannel heat sinks. The regression model underwent comprehensive interpretability analysis through Pearson's coefficient and Partial Dependency Plot (PDP)/Individual Conditional Expectation (ICE) methodologies. Our results demonstrate the GWO-SVR model's exceptional capabilities, achieving a 0.96531 coefficient of determination (R2) and minimal 0.00114 mean squared error (MSE) on training data, surpassing conventional machine learning algorithms. Through systematic evaluation, we determined that parameter influence followed a clear hierarchy-microchannel performance was most significantly affected by spacing adjustments, moderately responsive to Height variations, with Width and Young's modulus demonstrating equivalent but lesser impacts. PDP/ICE visualizations further clarified how each design parameter contributes to thermal efficiency, revealing their functional mechanisms and parameter interactions. Based on this comprehensive analysis, we established optimal configuration values for the flexible vortex generator array: spacing (394.7 mu m), height (195 mu m), width (21 mu m), and Young's modulus (7,900,000pa).