NONLINEAR PREDICTIVE CONTROL IN THE LHC ACCELERATOR C. de Prada * , E. Blanco ** , S. Cristea * , J. Casas ** * Dpt. of Systems Eng. and Automatic Control, University of Valladolid, Spain fax: +34 983 42 3161, prada@autom.uva.es, smaranda@autom.uva.es ** LHC Division, CERN, CH-1211, Geneva (Switzerland) fax: +41 22 767 4620, Enrique.Blanco@cern.ch, Juan.Casas@cern.ch Abstract: This paper describes the application of a nonlinear model-based control strategy in a real challenging process. A predictive controller based on a nonlinear model derived from physical relationships, mainly heat and mass balances, has been developed and commissioned in the Inner Triplet Heat Exchanger Unit (IT-HXTU) prototype of the LHC particle accelerator being built at CERN, operating at a temperature of about 1.9 K. The development includes a state estimator with a receding horizon estimation procedure to improve the regulator predictions. Copyright © 2003 IFAC Keywords: Nonlinear Predictive control, Nonlinear estimators, Cryogenic 1. INTRODUCTION MPC strategies have become the preferred control technique for many process control problems, most of the industrial MPC controllers using an internal linear dynamic model. Nevertheless, many common processes exhibit nonlinear behavior and they may be required to operate over a wide range of conditions, therefore, these controllers are often tuned in a conservative way, which can result in serious degradation of controller performance. An alternative is to use a non-linear internal model. Many controllers, using different types of internal models, have been proposed in the literature, but the number of non-linear MPC is still low in industry. In this paper we present an implementation of a non- linear MPC to a challenging process, which requires to operate under strong requirements. It corresponds to the new particle accelerator, the LHC, which is under construction at CERN, Geneva. In its final form it will expand in a circumference of about 27 Km in France and Switzerland. In order to test the proposed design some prototypes such as the String1 (Casas et al., 1998) and the IT-HXTU (Byrns et al., 1998) were built (Fig.1). The aim of the LHC is to accelerate particles at speeds close to the one of the light in order to study the results of its collisions. For this purpose, the particles are driven within the LHC accelerator using very strong magnetic fields, which requires high electrical currents of about 12 kA for its magnets. A practical operation of the magnets requires operating with no electrical resistance in the coils, superconductivity condition, that can be maintained only at extremely low temperatures of around 1.9K. The main aim of the control system presented in this paper is to maintain this temperature in a narrow range, 50 mK, in spite of the unknown disturbances acting on the process. This is required in order to avoid a “quench” that will stop the operation. Fig. 1 Inner Triplet Prototype (length: 30 meters) Several linear control strategies has been tested at String1, including PID and linear MPC (Blanco, 1999; Cristea, 1998) but all of them suffer from the above mentioned problem: their performance is degraded when, due to different heat load charges, the unit must operate in different working conditions. This was the main reason to implement NMPC. This paper presents the non-linear approach, as well as state and disturbance estimations that were not taken into account in previous versions, and it is organized as follows: After the introduction, section 2 describes the String1 and its cryogenic system. Section 3 is devoted to the process model and section 4 to the non-linear controller including the state estimator. Experimental results are given in section 5. The paper ends with some brief conclusions.