Thermal full-field prediction of an air-cooled data center using a novel multi-scale approach based on POD and CFD coupling

Abstract

A full-knowledge of the temperature distribution in a data center is crucial for the design and operation to avoid thermal failure. However, it is a great challenge to conduct full-field simulations using the computational fluid dynamics/heat transfer (CFD/HT) method due to the unacceptable computational cost caused by the multi-scale structure of data centers. In this paper, a novel multi-scale model based on POD and CFD coupling is proposed to predict the thermal and fluid characteristics in data centers from room scale to chip scale. The temperature and flow fields in an air-cooled data center are predicted and analyzed under the scales of room, server, IGBT and chip. Based on the CFD results of Room-Server level, 16 sets of design cases’ temperature fields for Server-IGBT level are compared using the POD and CFD methods, which shows that the maximum average absolute deviation and average relative deviation are 0.004 °C and 0.59%, respectively. Subsequently, three sets of off-design cases for Server-IGBT level and IGBT-Chip level are predicted using the aforementioned methods, yielding the maximum relative deviation 7.21% and 0.98%, respectively. The salient outcome is the remarkable 175-fold reduction in CPU processing time achieved with the POD model compared to traditional CFD simulations.