Nonlinear Model-Based Predictive Control applied to Large Scale Cryogenic Facilities

The thesis addresses the study, analysis, development, and finally the real implementation of an advanced control system for the 1.8 K Cooling Loop of the LHC (Large Hadron Collider) accelerator. The LHC is the next accelerator being built at CERN (European Center for Nuclear Research), it will use superconducting magnets operating below a temperature of 1.9 K along a circumference of 27 kilometers. The temperature of these magnets is a control parameter with strict operating constraints. The first control implementations applied a procedure that included linear identification, modelling and regulation using a linear predictive controller. It did improve largely the overall performance of the plant with respect to a classical PID regulator, but the nature of the cryogenic processes pointed out the need of a more adequate technique, such as a nonlinear methodology. This thesis is a first step to develop a global regulation strategy for the overall control of the LHC cells when they will operate simultaneously. For that the controller will have into account the effects of disturbances due to the certain coupling between cells, and also it should provide a common tuning structure along the LHC ring. The recent interest in the design and analysis of nonlinear control systems is due to several factors. First, linear controllers usually perform poorly when applied to highly nonlinear systems or moderately nonlinear that operate over a wide range of conditions. On the other hand, significant progress has been made in the development of model-based controller design strategies for nonlinear systems. These techniques employ the nonlinear model directly in the controller calculation without the need of local linearization around an operating point. Advances in computer science and electronics provided enough powerful computers which have made online implementation of these nonlinear model-based controllers feasible. Model-Based Predictive Control (MBPC) is an optimal-control based methodology to select control inputs by minimizing an objective function. The objective function includes both present and predicted system variables and is evaluated using an explicit model to predict the future process outputs. One of the more attractive features of this technique is the possibility of introducing constraints explicitly in the online calculation. Extensive work has been done with linear model-based predictive control in terms of performance, stability and design issues, and this technique is applied widely and with success in industry, mainly on petrochemical and chemical plants.  Despite of this fact, the nonlinear adaptation of this methodology remains still rather unexploited. This work intends to investigate and demonstrate the validity of a nonlinear model-based control strategy through a real industrial implementation of a nonlinear predictive controller to regulate the superconducting magnet temperature for the LHC accelerator. Two real-scale prototypes have been built to validate, among others, the cryogenics of the LHC accelerator and the designed controller. First, the so-called LHC Test String implementing a typical half cell of the accelerator (about 50 meters length) and, second, the IT-HXTU, representing special regions on the accelerator, the so-called Inner Triplet, with larger dynamic heat loads at 1.9 K. The regulator structure developed uses a nonlinear model derived from physical relationships, mainly heat and mass balances. The development includes a state estimator with a receding horizon estimation (RHE) procedure to improve the regulator predictions by overwhelming the always existing model mismatch and process disturbances. Real operation with the designed controller provided results which corroborate the assumptions of the expected performance improvement and helped in the day-to- day operation by reducing the temperature variability and increasing the overall plant safety.