Abstract

Data centers (DCs) are the backbone of modern digital infrastructure, but their complexity, energy intensity, and need for continuous uptime pose major maintenance challenges. This dissertation reviews current Operations and Maintenance (O&M) practices and develops three novel optimization models specifically tailored to DCs.
The literature review highlights research gaps: existing O&M models rarely address DC-specific issues such as balancing cost and high availability. It underscores the importance of integrating reliability and availability analyses into predictive maintenance, energy efficiency, and system optimization.
Building on these findings, three optimization models are introduced. The first applies Dynamic Programming (DP) and the multiple knapsack problem to prioritize maintenance of UPS units under budget constraints, achieving a 61% improvement in availability with optimized resource allocation. The second model develops an availability-based cost optimization framework for UPS units in k-out-of-n configurations. Case studies show optimal selection (k = 5 of 10), delivering 99.991% availability above Tier I standards while minimizing costs. Sensitivity analysis confirmed that availability is more affected by repair times than failure rates, and the validation under varying conditions confirmed the robustness of the results. The third model, formulated as a Mixed-Integer Nonlinear Program (MINLP), optimizes maintenance in Series-Parallel systems. Results demonstrate 99.974% availability (above Tier I targets) while maintaining costs within budget.
Together, these models advance DC maintenance by integrating cost, reliability, and availability into scalable decision-making frameworks. The contributions provide operators with practical tools to reduce costs, improve reliability, and ensure compliance with strict uptime standards.