Design and Implementation of an AC to DC Matrix Converter with High-Frequency Isolation and Power Factor Correction (for particle accelerator applications)
|University of Valencia
|Dr. J. M. Espí
|Ph. D. in Electronics Engineering
This thesis (written in Spanish) presents the analysis, design and implementation of a four-quadrant power supply with high-frequency isolation, which is expected to be used to feed the low-energy correction magnet of a particle accelerator. In particle accelerator applications the magnetic field during beam acceleration may be either positive or negative, and true bipolar power converters are needed. The selected bipolar topology consists of a bidirectional three-phase to single-phase reduced matrix converter (RMC) with power factor correction and a bidirectional active rectifier. Main features of this power converter are the ability to regenerate energy back to the utility when the magnet acts as generator, unity power factor at the mains and reduction of volume and weight thanks to the inclusion of the isolation transformer at the switching frequency. A space vector modulation (SVM) technique was used to achieve unity power factor at the input and output current regulation simultaneously. This was done while a symmetrical pure AC profile is generated at the primary side of the isolation transformer. The secondary AC signal is then rectified into a positive or negative voltage, according to the desired output current sign, and later filtered to obtain the output DC current in both polarities. The active rectifier used permits reverse current flow to the primary side when driving an inductive load. By synchronising the commutation of both converters and adding a saturable inductor and a blocking capacitor it is possible to achieve soft commutation for most of the semiconductor elements.
An all-digital control based on a Digital-Signal-Processor (DSP) and a Field-Programmed-Gate-Array (FPGA) was used to implement space vector modulation and output current regulation. Output current regulation is performed on a powerful 32-bit fixed-point DSP of Motorola, and was implemented by means of an observer based optimum state feedback control (LQR -- Linear Quadratic Regulator). A reduced order observer was implemented to estimate the output filter inductor current, reducing the number of sensors. Experimental results of a 1.5 kW, 20 kHz prototype are presented to illustrate the performance of the proposed topology.