Language:
    • Available Formats
    • Options
    • Availability
    • Priced From ( in USD )
 

About This Item

 

Full Description

Thermal health of electronic equipment in a data center, cooled by multiple Computer Room Air Conditioners (CRACs), is a strong function of their supply flow rate combinations. Control application of such data centers needs fast and accurate prediction of the flow/thermal behavior for varying CRAC supply flow rates and temperatures. Influence mass fractions, which quantitatively represent the air flow contributions from each CRAC to the servers, have been used to gain control insights previously. Though accurate, they fail to provide fast predictions due to the prerequisite of a Computational Fluid Dynamics (CFD) simulation for each new CRAC supply flow rate combination. Proper Orthogonal Decomposition (POD) models typically predict velocity or temperature fields for a given CRAC supply conditions. However, the models do not provide direct useful insights from control perspective. Here, a simplified reduced order model is proposed that integrates influence mass fractions with the POD framework, to provide fast and accurate prediction of control insights in a data center with variable CRAC supply flow rate combinations. The model enables prediction in terms of explicit influence of CRACs at desired server inlets and server inlet temperatures. The paper also suggests a new approach for selecting POD system snapshots that enables qualitative tuning of the prediction accuracy with the corresponding offline CFD effort. The prediction of influence mass fractions for a test data center is found to be in excellent agreement with corresponding CFD simulations. The model accurately predicts the maximum influencing CRAC for 90% of the servers in most test cases. The average error in prediction of server temperatures is shown to be as low as 0.94°C (1.69°F) for a sample case. Fast prediction of such insightful flow metrics with a limited offline CFD effort makes the model suitable for an online optimization and control application.