Speaker
Description
Cooling plants are energy-intensive systems which enable ideal thermodynamical conditions for the smooth operation of accelerator facilities. Among this class of plants, chilled water production systems, used to bring cooling water from high to low temperatures, are commonly found. The automatic control system of such plant is usually complex due to the large number of correlated control inputs available, rendering it particularly challenging to minimize its energy consumption when solely relying on conventional control methods (e.g., PID and/or if-based logic). In 2024, CERN's Engineering department partnered with the university Politecnico di Milano to develop an energy optimal model-predictive controller (MPC) for one of the critical chilled water production plants used for the cooling of CERN's flagship accelerator, the Large Hadron Collider. This 18-month collaboration is well underway, and this paper explores the motivational and organizational aspects of the project, as well as highlights the technical solution proposed, the challenges faced to date, and how these were overcome. Simulation-based results are presented for a detailed performance comparison between MPC and the currently used rule-based logic controller. Finally, the architecture of controls and operator interface for the MPC deployment in the real plant are discussed, in view of extending this optimal control solution to numerous similar systems at CERN.
