ScholarWorks@중앙대학교

초록

Efficient thermal management is essential for maximizing the performance and reliability of power semiconductor devices, particularly in data centers, electric vehicles, drive motors, and photovoltaic systems. Multi-jet impingement cooling has emerged as a promising solution, offering enhanced heat transfer from thinning boundary layers, and localized cooling capabilities, but its optimization remains challenging due to the complexity and high dimensionality of the design parameter space. This study develops machine learning framework coupled with a genetic algorithm to efficiently predict and optimize the thermal and hydraulic performance of multi-jet impingement cooling. We trained scalar and multimodal machine learning models using comprehensive computational fluid dynamics simulations, incorporating physical principles to enhance predictive accuracy. As a result, prediction errors decreased substantially, from 408.2% to 10.3% for pressure drop and from 25.9% to 3.7% for maximum temperature. Compared to conventional computational approaches, our proposed methodology significantly reduces computational effort and accelerates the identification of optimal cooling configurations. This study presents a robust and efficient strategy for advancing thermal management solutions critical to next generation high power semiconductor applications.

키워드