<p>Multi-Objective Optimization of Thermal Management Systems for Artificial Intelligence Server Racks Using Exergy Analysis and Life-Cycle Assessment</p>

Engineering notes

Engineering notes will be added by the Power for Solar editorial team.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为光伏效率、钙钛矿太阳能电池、储能技术、太阳能热利用、BIPV、并网技术等高价值论文补充中文说明。

Original abstract

Artificial intelligence (AI) workloads have drastically increased heat generation in current data centers, thus efficient and environmentally friendly cooling systems have become a necessity. Traditional thermal management (TMS) system design methods only focus on optimizing one performance metric like energy efficiency and overlook the environmental impacts throughout the system life cycle. This paper develops a multi-objective optimization framework for an AI server rack TMS by simultaneously maximizing exergy efficiency and minimizing the life-cycle environmental impact. Through second-law thermodynamics and a life-cycle assessment, a physics-based mathematical model is created. The Non-Dominated Sorting Genetic Algorithm II is used to solve the optimization problem and identify the feasible optimum under the imposed thermodynamic, economic, and environmental constraints. The optimization results yielded a feasible optimum with a scaled exergy-performance index of 329.34 and a life-cycle impact of 159881.57 normalized impact units. The exergy-performance value is not interpreted as a conventional percentage efficiency; rather, it represents the value of the adopted exergy objective function under the present mathematical formulation. In the present formulation, the knee-point solution coincided with this optimum, indicating that the best thermodynamic and environmental performance was achieved at the same operating condition. The proposed framework offers a structured preliminary model-based framework for sustainable thermal management design of AI server racks.

Links and sources

• PDF from original source

Comments