Multi-Phase System Supplied by Svm Vsi: A New Fast Algorithm to Compute Duty Cycles

Summary Multi-phase drives are more and more often used in industry, especially for high-power applications [17, 18, 20]. Space Vector Modulation (SVM) is now widely implemented and possesses many advantages over carrier-based pulse width modulation (PWM): – natural overmodulation implementation; – easy solution for saturation treatment; – fast and convenient to compute; – easy implementation of switching constraints for example to reduce harmonic currents [19]. Many authors proposed SVM VSI applied to multi-phase drives. For example, [2] and [15] have chosen instantaneous vectors to control dual 3-phase induction machines with low generated harmonic currents, [4] and [6] to control 5-phase machines. The initial space is split onto orthogonal subspaces (d-q and zero-sequence) and the initial reference vector can be expressed at any sampling time in terms of several reference vectors, each one belonging to one subspace (plane and/or line). Each reference vector is located onto a sector, bounded by two active vectors, and decomposed onto these vectors. Once the two vector components are known, duty ratios are determined. Then, zero vectors are chosen and switching sequencing is imposed. Due to the high number of phases, a high number of sectors exist and the location of the different reference vectors leads to a great computational requirement (Fig. 1a). Using the equivalence between a multi-phase machine and a set of fictitious one-phase or two- phase machines which are magnetically independent but mechanically and electrically coupled [1, 13], we propose a new fast algorithm to compute the duty cycles of each VSI leg. This algorithm, based on a vectorial approach of inverters developed in [3, 5, 7, 22], thereby reduces computation time and allows to use low computational requirements. This paper shows that, compared to classical techniques [9, 10, 11, 12], it is no more necessary to find the location of the reference vectors to get explicitly the duty cycle of each leg. Fig. 1 shows the difference between classical algorithm (Fig. 1a) and proposed one (Fig. 1b). This proposed technique is at first illustrated on a 3-phase drive. Geometrical representations allow then to establish links with usual 3-phase SVM. The implementation of the proposed SVM is achieved in the vector control of a 5-phase drive. Experimental results are presented and confirm the theoretical approach.