Passive flow regulator for achieving uniform flow in cooling systems for artificial intelligence servers under different pressures

Flow instability and maldistribution are key challenges in multichannel liquid-cooling systems within the coolant distribution units of artificial intelligence servers. Although prior research has examined microchannel design and flow restrictors, macroscale imbalance among cold plates remains insufficiently explored. Therefore, this study explores the use of spring-loaded passive pressure-reducing valves (PRVs) as flow self-regulating devices that do not require external control. A computational fluid dynamics model with mesh deformation was developed and validated through single-channel experiments, in which mean absolute percentage errors of less than 10.16% were obtained for pressure and flow rates. Parametric studies reveal that an increase in the spring constant (ks) causes an increase in the outlet pressure setpoint (Pset). Moreover, a greater precompressed spring length (l0) reduces the activation pressure for flow regulation. Parallel-channel experiments confirm that PRVs improve flow uniformity but increase pressure loss; thus, pump and energy budget optimization must be conducted in pressure-sensitive systems.