Session: 15-01-01: ASME International Undergraduate Research and Design Exposition

Paper Number: 151924

151924 - Optimization and Machine Learning Control of Multi-Temperature, Multi-Module Thermal Energy Storage Systems for Single-Phase Immersion Cooling Applications 

Data center thermal management systems which rely on air cooling are no longer able to keep pace with the power densities of modern processors. Single-phase liquid immersion cooling offers a solution for high power density applications with improved energy efficiency, low cost, and minimal complexity compared to other liquid cooling methods. Phase change thermal energy storage can be integrated into single-phase immersion cooling systems to provide flexibility in utilizing renewably sourced electricity and improve reliability by acting as a backup system when the primary cooling system fails. Multi-temperature, multi-module thermal energy storage ensembles are comprised of multiple phase change materials combined in series and parallel configurations which allow for optimization and flexibility in the operation of the thermal storage. While previous studies have focused on applying such systems to hot storage for concentrated solar power plants, the present study demonstrates their application to cold storage for single-phase immersion cooling systems in data centers. The cold thermal storage ensemble is optimized and controlled by an artificial neural network which is trained using a multi-objective optimization. The optimization uses the epsilon-constraint method to balance competing objectives to match a target cooling supply fluid temperature leaving the cold storage while also minimizing the instantaneous rate of exergy destruction during the heat transfer process. The performance of the thermal energy storage ensemble was compared to two baseline thermal storage systems: a single module system with one phase change material and a cascaded system with multiple phase change materials. Typically, cascaded systems demonstrate an improvement in exergy efficiency compared to single module systems but are difficult to design and control due to their inability to adapt to dynamic operating conditions. Over the range of expected operating conditions for an immersion cooling application, the results show that the ensemble reduces the rate of exergy destruction by 52% compared to the single module system and achieves a 5°C improvement in performance for matching the target supply fluid temperature to the servers compared to the cascaded system. Therefore, the ensemble system presented in this study is the ideal thermal energy storage solution as it achieves the exergy efficiency of a cascaded system while maintaining the flexibility needed to respond to dynamically changing operating conditions. The adaptability and potential for optimization of the thermal energy storage ensemble makes it an excellent solution for improving the efficiency of data center cooling systems with many variable parameters including fluctuating thermal loads, daily and seasonal changes in outdoor temperatures, and availability of renewably sourced electricity.

Presenting Author: Guelor Kabeya San Francisco State University

Presenting Author Biography: Guelor Kabeya is a senior undergrad student in mechanical engineering at San Francisco State university. He is interested in engineering design and machine learning.

Authors:

Guelor Kabeya San Francisco State University
Alanna Cooney San Francisco State University