Data-driven controller design using the CERN Power Converter Control Libraries (CCLIBS)

The data-driven control approach is a control methodology in which a controller is designed without the need of a model. Parametric uncertainties and the associated unmodeled dynamics are therefore irrelevant; the only source of uncertainty comes from the measurement process. The CERN Power Converter Control Libraries (CCLIBS) have been updated to include data-driven H-infinity control methods recently proposed in literature. In particular, a two-step convex optimization algorithm is performed for obtaining the 2-degree-of-freedom controller parameters. The newly implemented tools in CCLIBS can be used both for frequency response measurement of the load and for controller synthesis. A case study is presented where these tools are used for an application in the CERN East Area Renovation Project for which a high-precision 900 A trapezoidal current pulse is required with 450 ms flat-top and 350 ms ramp-up and ramp-down times. The tracking error must remain within +/- 100 parts-per-million (ppm) during the flat-top (before the ramp-down phase starts). The magnet considered in the case study is of non-laminated iron type, hence the necessity of data-driven techniques since the dynamics of such a magnet is difficult to be modeled accurately (due to eddy currents losses). The power converter used is a SIRIUS 2P (with a current and voltage rating of 400 Arms and 450 V, respectively) whose digital control loop is regulated at a sampling rate of 5 kS/s.