Document Type

Dissertation

Degree

Doctor of Philosophy (PhD)

Major/Program

Mechanical Engineering

First Advisor's Name

Cheng-Xian Lin

First Advisor's Committee Title

Committee Chair

Second Advisor's Name

George Stevo Dulikravich

Second Advisor's Committee Title

Committee member

Third Advisor's Name

Yiding Cao

Third Advisor's Committee Title

Committee member

Fourth Advisor's Name

Arindam Chowdhury

Fourth Advisor's Committee Title

Committee member

Fifth Advisor's Name

Xiaobing Liu

Fifth Advisor's Committee Title

Committee member

Sixth Advisor's Name

Shaolei Ren

Sixth Advisor's Committee Title

Committee member

Keywords

CFD, data center, thermal management, reduced order modeling, response surface methodology, rapid flow simulation, hybrid turbulence modeling, optimization, tile modeling

Date of Defense

6-28-2019

Abstract

Computational fluid dynamics (CFD) has become a popular tool compared to experimental measurement for thermal management in data centers. However, it is very time-consuming and resource-intensive when used to model large-scale data centers, and may not be ready for real-time thermal monitoring. In this thesis, the two main goals are first to develop rapid flow simulation to reduce the computing time while maintaining good accuracy, and second, to develop a whole building energy simulation (BES) strategy for data center modeling. To achieve this end, hybrid modeling and model training methodologies are investigated for rapid flow simulation, and a multi-zone model is proposed for BES.

In the scope of hybrid modeling, two methods are proposed, i.e., the hybrid zero/two-equation turbulence model utilizing the zone partitioning technique and a combination of turbulence and floor tile models for the development of the composite performance index. It shows that the zero-equation coupled with either body force and modified body force tile models have the best potential in reducing the computing time, while preserving reasonable accuracy. The hybrid zero/two-equation method cuts down the computing time in half compared to the traditional practice of using only two-equation model.

In the scope of model training, reduced order method via proper orthogonal decomposition (POD) and response surface methodology (RSM) are comprehensively studied for data center modeling. Both methods can quickly reconstruct the data center thermal profile and retain good accuracy. The RSM method especially shows numerous advantages in several optimization studies of data centers. Whether it is for the tile selection to control the server rack temperature difference or impacting the decision for the input design parameters in the early stage of data center infrastructure design, RSM can replace the costly experiments and the time-consuming and resource-intensive CFD simulations.

Finally, for the whole BES study, the proposed multi-zone model is found to be much more effective compared to the common use single zone model. The location factor plays an important role in deciding whether some of boundary conditions are affecting the cooling electricity consumption. In addition, the effect of supply temperature and volumetric flow rate have significant effects on the energy consumption.

Identifier

FIDC007790

Share

COinS
 

Rights Statement

Rights Statement

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/
This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).