Driven by the rapid expansion of artificial intelligence workloads, data centers have become major global consumers of electricity. Consequently, the efficiency of their thermal management systems is increasingly critical from both economic and environmental perspectives. This thesis investigates the thermal behavior of data center Heating, Ventilation, Air Conditioning, and Refrigeration (HVACR) systems through physics-based modeling and Global Sensitivity Analysis (GSA). In collaboration with Vertiv, an acausal thermodynamic model of a real data room is developed in MATLABSimscape/Simulink. The model captures the facility’s dominant thermal-fluid dynamics while maintaining the computational efficiency required for repeated, simulation-based analysis. To characterize the model’s input-output sensitivity structure, a structured three-stage GSA pipeline is designed and applied, combining the Elementary Effects Test (EET), a two-sample Kolmogorov-Smirnov (KS) test, and the moment-independent PAWN method. This workflow addresses factor screening and prioritization while allowing the methods to cross-validate each other, yielding a highly robust characterization of the system without relying on a single technique in isolation. The resulting sensitivity analysis provides deep insights into the physical mechanisms governing the data room, establishing a quantitative basis to inform model improvements and energy-efficient control strategies. Furthermore, the GSA campaign serves as a qualitative internal consistency check, confirming that expected physical behaviors and topological symmetries naturally emerge from the model’s response. This qualitative assessment complements the necessary quantitative validation against a high-fidelity reference, such as Computational Fluid Dynamics (CFD) or real operational data, which is identified as an essential direction for future work. Overall, this thesis contributes a computationally tractable acausal modeling framework paired with a rigorous GSA methodology, bridging the gap between physical modeling and statistical sensitivity analysis for advanced cooling system design.
Driven by the rapid expansion of artificial intelligence workloads, data centers have become major global consumers of electricity. Consequently, the efficiency of their thermal management systems is increasingly critical from both economic and environmental perspectives. This thesis investigates the thermal behavior of data center Heating, Ventilation, Air Conditioning, and Refrigeration (HVACR) systems through physics-based modeling and Global Sensitivity Analysis (GSA). In collaboration with Vertiv, an acausal thermodynamic model of a real data room is developed in MATLABSimscape/Simulink. The model captures the facility’s dominant thermal-fluid dynamics while maintaining the computational efficiency required for repeated, simulation-based analysis. To characterize the model’s input-output sensitivity structure, a structured three-stage GSA pipeline is designed and applied, combining the Elementary Effects Test (EET), a two-sample Kolmogorov-Smirnov (KS) test, and the moment-independent PAWN method. This workflow addresses factor screening and prioritization while allowing the methods to cross-validate each other, yielding a highly robust characterization of the system without relying on a single technique in isolation. The resulting sensitivity analysis provides deep insights into the physical mechanisms governing the data room, establishing a quantitative basis to inform model improvements and energy-efficient control strategies. Furthermore, the GSA campaign serves as a qualitative internal consistency check, confirming that expected physical behaviors and topological symmetries naturally emerge from the model’s response. This qualitative assessment complements the necessary quantitative validation against a high-fidelity reference, such as Computational Fluid Dynamics (CFD) or real operational data, which is identified as an essential direction for future work. Overall, this thesis contributes a computationally tractable acausal modeling framework paired with a rigorous GSA methodology, bridging the gap between physical modeling and statistical sensitivity analysis for advanced cooling system design.
Modelling, Simulation, and Global Sensitivity Analysis of a Data Center Cooling System
CACCIUTO, EMILIANO
2025/2026
Abstract
Driven by the rapid expansion of artificial intelligence workloads, data centers have become major global consumers of electricity. Consequently, the efficiency of their thermal management systems is increasingly critical from both economic and environmental perspectives. This thesis investigates the thermal behavior of data center Heating, Ventilation, Air Conditioning, and Refrigeration (HVACR) systems through physics-based modeling and Global Sensitivity Analysis (GSA). In collaboration with Vertiv, an acausal thermodynamic model of a real data room is developed in MATLABSimscape/Simulink. The model captures the facility’s dominant thermal-fluid dynamics while maintaining the computational efficiency required for repeated, simulation-based analysis. To characterize the model’s input-output sensitivity structure, a structured three-stage GSA pipeline is designed and applied, combining the Elementary Effects Test (EET), a two-sample Kolmogorov-Smirnov (KS) test, and the moment-independent PAWN method. This workflow addresses factor screening and prioritization while allowing the methods to cross-validate each other, yielding a highly robust characterization of the system without relying on a single technique in isolation. The resulting sensitivity analysis provides deep insights into the physical mechanisms governing the data room, establishing a quantitative basis to inform model improvements and energy-efficient control strategies. Furthermore, the GSA campaign serves as a qualitative internal consistency check, confirming that expected physical behaviors and topological symmetries naturally emerge from the model’s response. This qualitative assessment complements the necessary quantitative validation against a high-fidelity reference, such as Computational Fluid Dynamics (CFD) or real operational data, which is identified as an essential direction for future work. Overall, this thesis contributes a computationally tractable acausal modeling framework paired with a rigorous GSA methodology, bridging the gap between physical modeling and statistical sensitivity analysis for advanced cooling system design.| File | Dimensione | Formato | |
|---|---|---|---|
|
Cacciuto_Emiliano.pdf
Accesso riservato
Dimensione
37.75 MB
Formato
Adobe PDF
|
37.75 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/116320