单相浸没冷却数据中心的多尺度模型研究

To address the challenges traditional cooling methods face in effectively managing the high power density heat dissipation demands of data centers, as well as issues such as local hotspots that can lead to decreased equipment performance and reliability, a multiscale numerical model for single-phase immersion cooling in data centers is proposed and developed. First, resistance curves for servers are derived from detailed models of single-phase immersion servers. Then, a multiscale simulation model for single-phase immersion cooling in data centers is constructed through transferring boundary conditions across different levels. Finally, the effects of different flow allocation methods in immersion chambers on the multiscale model are studied using the validated multiscale model. Numerical results indicate that the constructed model substantially reduces computational complexity and resource requirements. Compared to the “opposite-side inlet-outlet” method, the use of a “distribution chamber + manifold” method leads to a 74. 67% reduction in flow distribution unevenness and a chip node temperature drop of 3. 71 ℃. Moreover, the chip node temperature is proportional to the inlet liquid temperature of the chamber and inversely related to the inlet liquid velocity, thus achieving a complete simulation chain from the immersion chamber level to the chip level in data centers.