Graduation Semester and Year
Spring 2026
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Mechanical Engineering
Department
Mechanical and Aerospace Engineering
First Advisor
Dereje Agonafer
Abstract
The rapid escalation of power densities in high-performance computing and artificial intelligence systems has intensified thermal management challenges in modern data centers. As conventional air cooling approaches its practical limits, liquid-based technologies—including direct-to-chip (D2C) cooling and single-phase immersion cooling (SPIC)—have emerged as viable alternatives. However, achieving energy proportionality, scalable hardware design, and coordinated system-level control under dynamic operating conditions remains a critical challenge.
This dissertation develops an integrated framework for thermal management in liquid-cooled data centers, spanning rack-level flow control, fluid-informed hardware standardization, and hybrid cooling analysis. First, an experimental rack-scale dynamic D2C architecture is established by integrating server-level temperature-based flow control devices (FCDs) with differential-pressure (ΔP) pump modulation. Using a four-server parallel test platform subjected to a 16-state transient load sequence, the coordinated control strategy achieves up to 25% reduction in pumping power while maintaining thermal stability and hydraulic robustness.
Second, a fluid-centric optimization framework for SPIC systems is proposed using Open Compute Project (OCP) figures of merit. A temperature-averaged similarity metric is introduced to correlate fluid properties with cross-fluid thermal performance penalties, enabling quantitative grouping of dielectric fluids and standardization of heatsink geometries without fluid-specific redesign.
Third, numerical investigations of hybrid SPIC–D2C architectures quantify discrepancies between component-level and loop-level heat capture ratios, revealing the impact of secondary heat exchange and flow coupling on system-level measurements.
Finally, transient simulations evaluate hybrid system response under off-nominal events, including pump shutdown and workload burst scenarios, to assess thermal safety margins and short-term robustness.
Collectively, this work advances liquid cooling from isolated component improvements toward coordinated, energy-aware, and scalable system-level architectures, providing practical design and control methodologies for next-generation data centers operating under dynamic workloads.
Keywords
Data center thermal management; Direct-to-chip liquid cooling; Single-phase immersion cooling; Hybrid cooling; Heat capture ratio; Liquid-cooled servers; Pumping power reduction; Dynamic flow control; Coolant distribution unit; Dielectric fluids; Heatsink optimization; Figures of merit; Electronics cooling; High-performance computing; Artificial intelligence data centers
Disciplines
Energy Systems | Heat Transfer, Combustion
License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Sivakumar, Akiilessh, "Thermal Management in Data Centers: A Study of Hybrid-Cooled Servers and Single-Phase Immersion Systems" (2026). Mechanical and Aerospace Engineering Dissertations. 1.
https://mavmatrix.uta.edu/mechaerospace_dissertations2/1