Session: 12-13-02: Heat Transfer in Electronic Equipment II

Paper Number: 166949

A Fusion Approach to Modeling Surface Roughness Effects on Nucleate Pool Boiling Heat Transfer via Machine Learning and CFD 

The increasing power densities in modern electronic devices necessitate advanced thermal management solutions to ensure their reliability and performance. Two-phase cooling techniques, such as two-phase immersion cooling, microchannel boiling, and two-phase jet impingement boiling, have gained significant attention due to their high heat removal capabilities. Among these techniques, nucleate pool boiling is a fundamental heat transfer mechanism, where surface roughness plays a crucial role in dictating boiling performance. Understanding the effects of surface roughness on boiling heat transfer is vital for designing efficient cooling solutions. Computational Fluid Dynamics (CFD) has emerged as a powerful tool for investigating such complex multiphase flow phenomena, providing insights into the underlying flow dynamics. However, while CFD can model various boiling physics, it may not always be possible to include all the intricate physical interactions and develop comprehensive models for every scenario. To bridge this gap, a fusion approach that integrates Machine Learning (ML) with CFD and experimental data is proposed to enhance the predictive accuracy of nucleate pool boiling models.

The primary contribution of this research is the development of a hybrid framework that leverages ML to refine CFD-based predictions by incorporating insights from experimental data. This fusion approach aims to address the limitations of standalone CFD modeling by integrating data-driven corrections derived from experimental observations. The work advances the field of boiling heat transfer by providing a more accurate and computationally efficient methodology for predicting the influence of surface roughness on nucleate boiling performance. By bridging the gap between empirical correlations, high-fidelity simulations, and experimental data, our approach offers a robust and scalable solution for optimizing two-phase cooling technologies.

Our methodology involves a multi-step process. First, a baseline CFD model is developed using an Eulerian multiphase framework to simulate nucleate pool boiling over surfaces with varying roughness parameters. While the CFD model captures key aspects of the boiling process, certain microscale interactions and complex phenomena may not be fully represented. To address this, experimental data from controlled boiling tests on engineered roughness surfaces is incorporated. Machine learning models are trained on the discrepancies between CFD predictions and experimental observations, allowing them to refine and enhance the predictive capability of the CFD model. These ML-enhanced corrections are then integrated back into the simulation framework to improve overall accuracy.

Preliminary results indicate that the ML-augmented CFD approach significantly enhances predictive accuracy while maintaining computational efficiency. The hybrid model effectively captures key surface roughness effects on boiling heat transfer, including variations in nucleation site density, bubble departure frequency, and heat transfer coefficients. The ML-enhanced CFD simulations exhibit closer agreement with experimental data compared to conventional standalone CFD models. Furthermore, the approach demonstrates the ability to generalize across different surface roughness conditions, providing a scalable tool for predicting boiling performance in practical applications.

In conclusion, this research presents a novel fusion approach that combines ML, CFD, and experimental data to model the effects of surface roughness on nucleate pool boiling heat transfer. The proposed methodology enhances the predictive capabilities of CFD while addressing the challenges associated with modeling all boiling physics explicitly. This makes it a valuable tool for designing next-generation two-phase cooling systems.

Presenting Author: Sreenivas Viyyuri Ansys Inc.

Presenting Author Biography: Sreenivas Viyyuri is a Lead Application Engineer at Ansys. He has over 20 years of experience in performing Computational Fluid Dynamics (CFD) simulations. He started his career in 2004 as a consulting and support engineer at Ansys India. During 2007-2011, he worked as a simulation specialist for Linde Gas in Germany supporting global application development teams for metallurgy, manufacturing, food & beverage, and water & wastewater industry segments. His focus areas are multiphase & reacting flows, and has worked extensively in modeling applications within oil & gas, energy, chemical and process, and pharmaceutical industries. In his current role, he is actively involved in developing solutions and best practices for applications in Energy and Hi-Tech industry. Sustainability is his key focus areas and is actively involved in leveraging simulations for building sustainable solutions.

Authors:

Sreenivas Viyyuri Ansys Inc.