随着人工智能与高性能计算任务的迅速发展, 数据中心对服务器散热能力的需求不断增长, 传统风冷技术已难以应对高功率密度带来的热管理挑战。为充分结合风冷与液冷系统各自的优势, 该文面向高功率密度数据中心散热与节能需求, 提出了2种服务器级相变风辅液冷系统技术方案, 分别为基于风冷改造方案和热管复合空调结合方案。2种方案均采用无泵驱动的气液相变换热机制, 兼具模块化结构与灵活部署特性。搭建了基于风冷改造方案的系统, 该系统在50kW热负荷下运行稳定, 气液管平均温度为39.85℃, 5个不同高度的蒸发器表面平均最大温差为1.9℃, 验证了其良好的散热能力与温度均匀性。开展了针对中国5个典型气候区域城市的节能分析, 冷源侧采用基于热管复合空调结合方案的风辅液冷系统, 该系统相较于传统计算机房空气处理系统在节能效果上提升可达约30%, 在北京和西安的全年电力使用效率低至1.16, 且系统在不同气候区均具备节能优势。

Objective: With the rapid advancement of artificial intelligence and high-performance computing, data centers are increasingly challenged to achieve effective thermal management and high energy efficiency. Traditional air-cooling methods struggle to handle rising power density, while liquid cooling offers superior heat transfer capabilities but often lacks deployment flexibility. To fully leverage the advantages of air and liquid cooling systems, this study proposes two server-level phase-change air-assisted liquid cooling (AALC) systems to enhance cooling performance, maintain modular deployment, and reduce energy consumption in high-density data centers. Methods: Two technical configurations were investigated: an air-cooling retrofitting scheme (System 1) and a hybrid solution incorporating a heat pipe composite air conditioner (System 2). Both systems utilized a pump-free, gas-liquid phase-change heat transfer loop to enable passive fluid circulation and simplify the structural design. System 1 was designed to combine the structural simplicity and deployment flexibility of air-cooling systems with the high heat transfer efficiency of liquid cooling. It enabled seamless upgradation of high-density cabinets within limited space in existing air-cooled data centers without requiring large-scale infrastructure modifications. An experimental platform simulating a 50kW cabinet heat load was developed using five electric heating modules. The thermal performance of System 1 was evaluated by monitoring the average temperature and surface uniformity across multiple evaporators. To further improve the energy-saving capability and year-round deployment potential across different climatic zones of China, System 2 was developed based on System 1 by incorporating a heat pipe composite air conditioner on the cold source side. This configuration retained the core advantages of passive two-phase cooling while enhancing the effectiveness of natural cooling. For System 2, annual power usage effectiveness (PUE) was simulated using hourly meteorological data from five representative Chinese cities. The total power consumption of the cooling system—including terminal fan power, outdoor fan power, and compressor power—was modeled using empirical polynomial-based expressions. In addition, economic metrics, such as annual energy savings and payback period, were evaluated. Results: System 1 exhibited stable operation under a 50kW cabinet heat load. The average temperature of the two-phase loop reached 39.85 ℃, and the maximum surface temperature difference among five vertical evaporators was only 1.9℃, demonstrating excellent thermal uniformity and heat dissipation capacity. The simulation results of System 2 showed up to 30% energy savings over conventional computer room air handler (CRAH) systems because of the enhanced use of natural cooling. In cities such as Beijing and Xi'an, the annual PUE was reduced to as low as 1.16. The hybrid operation considerably lowered fan and compressor energy consumption. Economic analysis indicated a payback period of approximately 1.89 years, confirming the financial viability of System 2. Conclusions: The proposed AALC systems provide a practical and scalable solution for thermal management in high-density data centers. System 1 verifies the feasibility of passive phase-change cooling for flexible cabinet-level retrofitting.