Multi-scale modeling and fast inference for thermal environment analysis of air-cooled data center

Both the thermal environment of a data center room and the internal thermal environment of a server are of great concern, but it is extremely difficult to build and solve full-scale computational fluid dynamics (CFD) simulation models from room scale to chip scale. In this paper, a multi-scale simulation method for data centers is proposed to achieve this aim by coupling room-level, rack-level, and server-level models. Simulations of the designed working conditions were carried out. The results showed that aisle containment systems improve the server's thermal environment, resulting in a drop in the maximum temperature of the server from 87.5 °C to 73.1 °C. A multi-scale model of the data center can be used to optimize system operation parameters conveniently and accurately. In this simulation, air conditioning parameters that contribute to energy savings with higher supply air temperatures and lower supply air volumes while guaranteeing safe server operation were determined. Moreover, fast and accurate prediction of chip temperature by radial basis function (RBF) networks has been proven to be feasible.