CyboSoft has introduced a Model-Free Adaptive (MFA) control solution for coolant distribution units (CDUs) in artificial intelligence data center liquid cooling systems.

The technology targets cooling stability, control performance, energy consumption and system reliability under rapidly changing computing loads.

According to CyboSoft, fluctuating graphics processing unit (GPU) workloads create challenges for maintaining stable coolant temperatures and pressures. Conventional proportional-integral-derivative (PID) controllers can struggle with nonlinear process behavior, time delays and interactions among control loops.

The MFA solution addresses these challenges through adaptive control of coolant temperature and differential pressure. It regulates secondary coolant supply temperature by adjusting the primary cooling valve. It also adjusts secondary pump speed to maintain coolant circulation.

These controls manage two coolant circuits separated by a heat exchanger. The primary circuit transfers heat to the facility cooling system. The secondary circuit circulates coolant through servers to remove heat from GPUs and central processing units (CPUs).

CyboSoft offers several MFA controller types, including Nonlinear MFA, Anti-delay MFA and Multi-Input-Multi-Output (MIMO) MFA controllers.

The company reports that its Nonlinear MFA controller maintained control as a process setpoint changed from 5% to 95%, despite a 40-fold change in process gain. In the same comparison, PID control experienced significant oscillations.

The MFA solution can run on an industrial personal computer using CyboCon or MFA Control Toolset for LabVIEW software. It also supports integration with existing distributed control systems (DCS) and programmable logic controllers (PLCs) through Open Platform Communications (OPC) or Modbus TCP. Manufacturers can also embed MFA controllers directly into equipment control systems.