Linear Model-Based Predictive Control of the LHC 1.8 K Cryogenic Loop

Abstract

The LHC accelerator will employ 1800 superconducting magnets (for guidance and focusing of the particle beams) in a pressurized superfluid helium bath at 1.9 K. This temperature is a severely constrained control parameter in order to avoid the transition from the superconducting to the normal state. Cryogenic processes are difficult to regulate due to their highly non-linear physical parameters (heat capacity, thermal conductance, etc.) and undesirable peculiarities like non self-regulating process, inverse response and variable dead time. To reduce the requirements on either temperature sensor or cryogenic system performance, various control strategies have been investigated on a reduced-scale LHC prototype built at CERN (String Test). Model Based Predictive Control (MBPC) is a regulation algorithm based on the explicit use of a process model to forecast the plant output over a certain prediction horizon. This predicted controlled variable is used in an online optimization procedure that minimizes an appropriate cost function to determine the manipulated variable. One of the main characteristics of the MBPC is that it can easily incorporate process constraints; therefore the regulation band amplitude can be substantially reduced and optimally placed. An MBPC controller has completed a run where performance and robustness has been compared against a standard PI controller (Proportional and Integral).