Multi-objective optimal design of hybrid liquid cooling system with manifold microchannels for data center server thermal management

The emerging compact and high-power data centers require achieving efficient cooling and heat dissipation. The study proposes a hybrid liquid cooling system for a server. A predictive model on the response surface methodology is initiated to quantify effects of key design parameters, including manifold channel width and height, parallel channel width and height, fin spacing, and liquid flowrate, on the maximum temperature, temperature difference, and heat dissipation efficiency of the cooling system. Both individual and interactive influences of these parameters are systematically analyzed. Regression models are constructed for the objectives, and Non-dominated Sorting Genetic Algorithm II is used to perform the multi-objective optimization, and generate a Pareto front for optimal design identification for minimizing the maximum temperature and temperature difference and maximizing heat dissipation efficiency. Under the optimized conditions at 705W thermal design power, the system achieves the maximum temperature of 67.92 °C, temperature difference of 15.99 °C, and heat dissipation efficiency of 17.74, corresponding to improvements of 4.36 %, 16.47 %, and 24.69 % from the basic conditions. The thermal performance predicted by the multi-objective optimization approach shows an average absolute deviation of less than 5 % compared with the data from the CFD model. The diminishing improvements at lower thermal loads indicate that the optimization method is more suitable for high-power servers. The optimized hybrid cooling system demonstrates substantial thermal improvements compared with air cooling. The proposed optimization framework could shed lights on enhancing both thermal management and energy efficiency in the server cooling systems of data centers.