Matrix Converter Commutation Strategies With and Without Explicit Input Voltage Sign Measurement

Transcription

1 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 2, APRIL Matrix Converter Commutation Strategies With and Without Explicit Input Voltage Sign Measurement Jochen Mahlein, Jens Igney, Jörg Weigold, Michael Braun, and Olaf Simon Abstract This paper gives an overview to voltage-controlled matrix converter commutation. Conventional converter systems with explicit sign measurement circuits that were necessary for the commutation in the past are discussed. New operation methods eliminating these extra measuring circuits are presented. The paper explains the new methods. Advantages and disadvantages of different approaches are discussed. The robustness of the new methods against disturbance is proved and measurements on a 5.5-kW matrix converter system are shown. Index Terms Bidirectional switch, commutation, critical area, matrix converter, prevention, prohibition, replacement, sign evaluation, sign measurement. I. INTRODUCTION FOR MORE than 20 years, the matrix converter has been discussed academically. The progress on silicon semiconductors has enabled a great advance on this type of converter, making it interesting for industrial applications. The matrix converter can be a simple 3-phase to 3-phase converter (Fig. 1). The number of phases is not limited. By using nine bidirectional switches (BDSs), the matrix converter is able to create a variable output voltage system of a desired frequency and magnitude [1] [14]. Normally, a BDS is built by two collector or emitter-connected insulated gate bipolar transistors (IGBTs). This enables a high switching frequency which is necessary to decouple input and output systems of the matrix converter. Altogether, 18 IGBTs for the BDS are more than 12 IGBTs needed for a dc-voltage link converter with power regeneration. However, it is a fact that the size of IGBT chips needed to build up a comparable matrix converter in power is reduced [4]. The necessary current ratings of the semiconductors and, with it, the silicon material, is reduced to 2/3. The installed switch power for the dc-voltage link converter and the matrix converter is about the same. A further aspect is the reduction of unit volume by renunciation of a chopper resistor and a live time limiting dc-link reactance. If an ordinary L C filter is added, the matrix converter is stressing the grid with sinusoidal current only. As mentioned, the converter offers power regeneration. The input power factor can be adjusted with little reduction to the active power transfer ratio. The sinusoidal voltage transfer ratio is limited to 0.86 on principle. A basic problem of the matrix converter is the absence of passive free wheeling paths for the load. This makes Manuscript received May 23, 2001; revised August 17, Abstract published on the Internet January 9, J. Mahlein, J. Igney, J. Weigold, and M. Braun are with the Elektrotechnisches Institut, University of Karlsruhe, D Karlsruhe, Germany ( mahlein@eti.etec.uni-karlsruhe.de). O. Simon is with Siemens AG, D Erlangen, Germany. Publisher Item Identifier S (02) Fig. 1. Fig. 2. Matrix converter with an L C filter. Single output phase. the matrix converter hard to handle concerning the commutation behavior and at pulse-off. A reliable protection at pulse-off is presented in [9], [12], and [13]. The problem of a safe and easy commutation strategy must be solved. II. COMMUTATION STYLES The knowledge of matrix converter commutation is fundamental in understanding the difficulties of the converter. On a single output phase, as shown in Fig. 2 the commutation process can be studied. In contrast to the well-known dc-voltage link converter, a commutation cannot be initiated without any knowledge of the commutation conditions. To perform a commutation, /02$ IEEE

2 408 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 2, APRIL 2002 Fig. 4. Common structure of a matrix converter system. Fig. 3. Legal commutation sequences. the voltage between the involved BDS or the load current must be measured. For a commutation, only the sign of one of the two quantities is important, as the sign determines the commutation sequence. A well-selected sequence enables a commutation without short circuiting the input voltages or breaking the load current. In Fig. 3, all applicable sequences are listed in its legal quadrants. The shown sequences are four step commutations. This means that, before a commutation, a whole BDS in an output phase is switched on. After a commutation has passed, this BDS is off and a BDS connected to a different input phase is switched on. There are 12 ways to perform a commutation. Fig. 3 explains the paths from a starting situation with BDS 11 switched on. The desired destination is to switch to phase B. The IGBTs of Fig. 2 change their state when they are named. In order to deal with the delay times of the IGBTs, a little dead time is kept between each commutation step. The two sequences on the axis only need the evaluation of the sign of the input voltage but are independent of the current sign. These sequences are named voltage commutation and are explained in [1]. The two sequences on the axis only need the evaluation of the sign of the load current and are named current commutation [3]. The remaining sequences need the evaluation of both signs. This fact makes them harder to handle. As a result, voltage or current commutation should be used. Another idea to achieve a commutation is the two-step commutation. The basic idea is to switch only the IGBT in a BDS which will lead the load current. In addition, all IGBTs in the matrix which will not conduct are switched on. This will reduce the total time of the commutation and opens additional free wheeling paths for the load current in case of error. The two-step commutation can be voltage controlled [6] or current controlled [7]. The shown commutation styles have been presented in many papers [1] [7], [11]. The general structure of the most used Fig. 5. Critical areas of commutation sequence selection. test setups is shown in Fig. 4. Pulsewidth modulation (PWM) generator and commutation logic are divided in separate logic devices. The commutation logic deals only with the switching commands of the PWM generator and the utilized sign of commutation. The detection of the sign is performed by a special circuit arrangement. This circuit has to work between the maximum input voltages or load currents. Therefore, this circuit has to meet high demands and must be very precise, making it difficult to design. For current-controlled commutation, with a four-step commutation the sign of the load current can be detected at high values without problems. A problem occurs if the load current nears the zero crossing. A wrongly detected sign will lead to a wrong sequence breaking the load current. If an inductive load is connected the voltage on the switching IGBT will rise rapidly with. In general, the switching IGBT will be destroyed exceeding its maximum blocking voltage if no snubber circuit is added. Finally, many matrix converters with current commutation have been presented [5], [7], [9], showing that this method is possible to implement. At a voltage-controlled commutation, an error-free detection gets critical at zero crossing of the phase-to-phase voltage between the involved BDS (Fig. 5). A poorly selected sequence will short circuit the input voltage phases. Having a look at the commutation circuit in Fig. 6, a short circuit current can only flow if the voltage is bigger than the sum of the forward voltages of the involved semiconductors. The critical voltage of detection is (1)

3 MAHLEIN et al.: MATRIX CONVERTER COMMUTATION STRATEGIES 409 Fig. 6. Short-circuit situation at voltage-controlled commutation. Fig. 7. Insertion of two additional sequences to phase C in a modulation period at a critical area between phases A and B with the replacement method. Depending on the used semiconductors, the critical voltage can be between [16] [18] Vto V IGBT diode Vto V RBIGBT (Reverse Blocking Insulated Gate Bipolar Transistor). This means that the voltage sign can be wrongly detected in an area around zero crossing between without negative consequences. This fact helps a lot in the design of a reliable voltage sign detection circuit. In case of a wrong detected voltage sign out of a critical area, the IGBT is stressed with a short circuit current during the period of one dead time between a commutation sequence step. The dead time is around 1 2 s. According to the thermal behavior of IGBT, this will not do any harm to the device if not repeated periodically. The voltage commutation seems to be the more practical solution for the matrix converter. In the following, the paper deals with the four-step voltage-controlled commutation if not commented differently. III. COMMUTATION STRATEGIES To reduce the expenses of the converter, an extra sign detection circuit is not desired. An idea is to use the A/D converters as they are necessary to calculate the controlling and the PWM patterns of the converter. Normally, this sign detection works well and a loss-optimized modulation can be used. A loss-optimized modulation is switching between voltages that do not differ much from each other. For example, in the marked uncritical area in Fig. 5, a switching sequence can be B C A and backward. The problem with this idea is again sign detection in the critical areas. There are several possibilities to manage a critical situation of commutation if no precise sign detection is desired. Prohibition: A critical commutation sequence is not executed until the voltages differ enough from each other (or the current sign can be detected without failure concerning the current commutation [5]). The effect on the output is little because the voltages have a small difference. A disadvantage is the shape of the input current. Because of the calculated pulse patterns which will not be executed, the input current will differ from the ideal sinusoidal shape. Evaluation of Current and Voltage Sign: A combination of voltage and current commutation may help in a lot of cases. The sequences used can be chosen from Fig. 3. Before a sequence is initiated, the signs and their reliability are checked. Finally, there will be situations in which both signs are unsure. As a result, the method as described in prohibition has to be chosen. Replacement: A critical area is driven around (Fig. 7). The critical sequence is replaced by two uncritical sequences which will commutate to the remaining third input phase and then to the desired destination phase. This method is described in [8] for a two-step commutation and fits for four-step commutation as well. The method is effective and easy to implement in the commutation logic. Only information about a critical area must be added. A disadvantage is caused by the extra sequences which are inserted. They will increase the switching losses of the converter and double the total time of commutation in a critical area. Therefore, the output voltage and the input current will differ from the ideal sinusoidal shape in a critical area. Especially at small output voltages, this method becomes unsuitable. In noncritical areas, the modulation can be executed without insertion of additional sequences. Prevention: A new idea is to avoid the critical switching patterns in a critical area (Fig. 8). This new method has been found by using the rectifying and inverting vector modulation (RIVM) algorithm [11], [13], [14]. In a critical area, the modulation patterns are reshuffled so that no critical sequence appears. No additional sequences are required compared to the replacement method. In the following, the realization of method prevention and replacement are described in more detail. IV. DESIGN CONSIDERATIONS TO THE ROBUST COMMUTATION METHODS REPLACEMENT AND PREVENTION The strategy of the prevention method is simple and easy to implement. Fig. 8 shows the same demanded average output voltage as in Fig. 7. The modulation patterns demanded

4 410 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 2, APRIL 2002 level has to be chosen this way so that an disturbed input voltage cannot have any influence on the commutation process. The worst case of this assumption is an excitation of the -filter at the input lines of the converter, e.g., at power turn on of the converter. The values of the filter devices are well known and the resonant frequency can be calculated from the formula (2) Fig. 8. Rearrangement of the modulation patterns in a modulation period of the prevention method. Looking at Fig. 11, a short circuit of the input voltage sources can only happen if a disturbed phase-to-phase voltage with a zero crossing is not detected within a certain time and the modulator is not switched to the prevention method or the commutation logic is not working with replacement. By knowledge of the resonant frequency and a maximum assumed voltage magnitude of the oscillating filter, a maximum time of conversion at excitation can be determined. A linearization of the grid sinus and the resonant sinus graph with no phase shift will lead to (3) Fig. 9. Advanced structure of a matrix converter system. by the PWM are rearranged. Instead of using the sequence A B C, the critical step from phase A B is avoided by rearranging the sequence to A C B. Neither the average value of the input current nor the average value of the output voltage are influenced. To implement this method, the system structure has to be modified. The considerations are the same as for replacement. An example is given in Fig. 9. The difference with the former structure is the connection of the A/D converters. Normally, the A/D converters are read out directly by a digital signal processor (DSP) or a micro controller (MC). The resolution of the voltage signal is limited to the sampling time of the processor. To peform the normal PWM, a sampling time of 100 to 50 s is enough. To replace an exact sign detection circuit, the resolution in time of the A/D converters has to be increased. The increase of the sampling time of the processor can be a solution but is an expensive method if the processor is calculating the PWM pattern in every sampling period. Another way is to implement the PWM generator, the commutation logic, the sign detection, and A/D-converter control in one logic device [e.g., field programmable gate array (FPGA)]. Now, the voltage signal can be sampled independently from the demands of the PWM generation. If a sigma delta converter is used, the measured bit stream has to be filtered to analyze the value of the input voltage (Fig. 10). From the voltage signal, the critical and uncritical areas and the voltage signs are calculated. The sampling rate of the sigma delta A/D converter and the necessary size of the critical area are related. A window that characterizes the critical area in the input voltage system has to be defined. The window with the borders of a critical voltage in which and are the grid frequency and the maximum magnitude of the phase-to-phase input voltage. The maximum gradient exists at zero crossing of the phase-to-phase input voltage. Fig. 11 shows this situation with a superimposed 1.7-kHz oscillation of the filter with a magnitude of %. The grid was assumed to be V According to the delay time of the A/D conversion and the processing time of the digital filters, a dead time has to be considered. The tolerant time of conversion is Now, the maximum permissible magnitude of a disturbance by excitation can be calculated from Table I gives a few examples of the reduced time of conversion of the new methods. The resonant frequency of the filter was assumed to khz again. The case marked with cannot be handled with a causal system because the time becomes negative. That means the PWM has to be switched to prevention or the commutation logic to replacement before the voltage is detected. This shows hat the capability of the new voltage sign detection of the matrix converter is limited by the characteristics of the signal processing method used. If a critical area is reached within a modulation period, a sequence already started cannot be reshuffled. The method of prevention cannot work immediately. To overcome this situation, the method of replacement which can be enabled with the detection of the sign is used as a final instance. This shows the combination of prevention and replacement is perfect for matrix converter commutation without explicit voltage sign measurement. Hz (4) (5)

5 MAHLEIN et al.: MATRIX CONVERTER COMMUTATION STRATEGIES 411 Fig. 10. Signal structure of the new sign detection. TABLE I EXAMPLES OF TOLERANT TIME OF REACTION FOR SIGN DETECTION Fig. 11. Zero crossing of the phase-to-phase input voltage (50 Hz) with a filter resonance of 20% (^v at 1.7 khz). An implementation of these strategies with current-controlled four-step commutation is not possible. Under real working conditions, additional aspects of the input voltage caused by higher harmonics or switching transients on the grid will influence the theoretical result. To avoid an error in voltage sign detection, the window marking the critical area should be selected carefully. A good security distance is recommendable. Investigations on an experimental setup (Fig. 12) have shown that a window for a critical area should be chosen to be 20% 27% of. For example, the window for input phase A and B should be defined as Fig. 12. Test setup of the matrix converter. The time of prevention modulation is about 38% 50% of the main period. V. EXPERIMENTAL TEST SETUP The new commutation method has been developed and tested on a 5.5-kW matrix converter test setup as shown in Fig. 12. The BDSs are collector-connected IGBTs which are all placed in a single ECONOMAC chase of EUPEC [15]. This will ensure a compact design of the setup and a low parasitic stray inductance from the filter capacitors to the IGBTs. The filter capacitors are connected in delta configuration. The three-phase filter coils can be seen on the left side of the picture. The two PCB boards contain the gate units and the over-voltage protection circuits. The system is controlled by a TMS 320 C40 DSP system. If a three-phase machine is added to the converter, the drive system can be operated with space vector control. The FPGA contains a PWM generator, commutation logic, and sign detection. A structure, shown in Figs. 9 and 10, is implemented in an EPF

6 412 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 2, APRIL 2002 Fig. 13. Load voltage and load current at a passie R L load. Fig. 14. Grid voltage and grid current at 50 Hz. 10K100 FPGA device from Altera. A combination of prevention and replacement methods is realized. VI. MEASUREMENTS Fig. 13 shows the output voltages and currents of the matrix converter setup at a passive load at standard operation. In Fig. 14, the grid voltage and current taken by the converter with is shown. Fig. 15 shows the change of the PWM algorithm from a loss-optimized modulation. As mentioned before, a critical area could be reached within a modulation period. The first commutation at a change of the modulation is executed by replacement method. The following commutations are executed by the prevention method. At the end of a critical area, the converter is switched back to loss-optimized modulation. To verify the robustness of the combined methods, the converter was tested under a heavy disturbed input voltage with an space-vector-controlled induction machine as a load. The disturbance was created by switching the used RIVM algorithm to over-modulation. This mode is described in [11] and is stimulating the L C filter to oscillations. The shape of the input voltage during the robustness test can be seen in Fig. 16. The critical rate of rise of voltage at zero crossing of the input voltage sinus with addition of the filter oscillation was measured to 3.2 V/ s. The maximum magnitude of the filter oscillation was measured to 175 V. The resonant frequency of the filter is 3 khz. At a voltage window size of 150 V, no commutation short circuit was measured. VII. CONCLUSION This paper gives a summary of the possible commutation styles and strategies. A new robust commutation strategy without an explicit sign measurement is presented. Measurements of a test setup running without explicit sign measurement for the first time are shown. The robustness against disturbance is proven by measurements. The new method works very well.

7 MAHLEIN et al.: MATRIX CONVERTER COMMUTATION STRATEGIES 413 Fig. 15. Change of commutation strategies. Fig. 16. Phase-to-phase input voltage during robustness test. The proposed structure of signal processing is completely digital. New paths in matrix converter design can be tread, now offering all the well-known advantages of digital system integration. [9] P. Nielsen, The matrixconverter for an induction motor drive, Ph.D. dissertation, Aalborg University, Aalborg East, Denmark, [10] L. Huber and D. Borojevic, Space vector modulated three-phase to three-phase matrix converter with power factor correction, IEEE Trans. Ind. Applicat., vol. 31, pp , Nov./Dec [11] J. Mahlein, O. Simon, and M. Braun, A matrix converter with space vector modulation enabling overmodulation, in Proc. EPE, Lausanne, Switzerland, 1999, CD-ROM Paper 394. [12] J. Mahlein and M. Braun, A matrix converter without diode clamped over-voltage protection, in Proc. IEEE IPEMC, vol. 2, Beijing, China, 2000, pp [13] O. Simon and M. Bruckmann, Control and protection strategies for matrix converter, in Proc. SPS/IPC/Drives, Nürnberg, Germany, [14] O. Simon and M. Braun, Theory of vector modulation for matrix converter, in Proc. EPE, Graz, 2001, CD-ROM Paper [15] M. Hornkamp, M. Loddenkötter, M. Münzer, O. Simon, and M. Bruckmann, Economac, the first all-in-one IGBT module for matrix converters, in Proc. PCIM 2001, Nürnberg, Germany. [16] Power Semiconductors, IGBT Modules, Siemens Data Book, Munich, Germany, [17] Power Electronics, Semikron Data Book, Nürnberg, Germany, [18] IXYS Semiconductors Advanced technical information. Data sheet IXRH 50R80/60 IGBT with reverse blocking capability. [Online]. Available: REFERENCES [1] A. Alesina and M. Venturini, Analysis of optimum-amplitude nineswitch direct AC AC converters, IEEE Trans. Power Electron., vol. 4, pp , Jan [2] M. Braun, Ein dreiphasiger Direktumrichter mit Pulsweitenmodulation zur getrennten Steuerung der Ausgangsspannung und der Eingangsblindleistung, Dr.-Ing. dissertation, Technische Hochschule Darmstadt, Darmstadt, Germany, [3] N. Burany, Safe control of four-quadrant switches, in Conf. Rec. IEEE-IAS Annu. Meeting, 1989, pp [4] T. Matsuo, S. Bernet, R. Colby, and T. Lipo, Application of the matrix converter to induction motor drives, in Conf. Rec. IEEE-IAS Annu. Meeting, vol. 1,, San Diego, CA, 1996, pp [5] L. Empringham, P. Wheeler, and J. Clare, Intelligent commutation of matrix converter bi-directional switch cells using novel gate drive techniques, in Proc. IEEE PESC 98, Fukuoka, Japan, pp [6] M. Ziegler and W. Hofmann, Performance of a two steps commutated matrix converter for ac-variable-speed drives, in Proc. EPE, Lausanne, Switzerland, 1999, CD-ROM Paper 258. [7] L. Empringham, P. Wheeler, and J. Clare, Bi-directional switch commutation for matrix converters, in Proc. EPE, Lausanne, Switzerland, 1999, CD-ROM Paper 409. [8] M. Ziegler and W. Hofmann, A new two steps commutation policy for low cost matrix converters, in Proc. PCIM Conf., Nürnberg, Germany, Jochen Mahlein received the diploma engineer title from the University of Karlsruhe, Karlsruhe, Germany, in Since 1996, he has been an Assistant Professor in the Electrotechnical Institute, University of Karlsruhe. He teaches courses in Control of Electrical Drives and Practice of Electrical Drives. His research areas are direct converters, soft-switching converters, and control of electrical drives. Jens Igney was born in Heidelberg, Germany, in He received the Dipl.-Gew. and Dipl.-Ing. degrees with the main subject in electrical engineering from the University of Karlsruhe, Karlsruhe, Germany, in 2000 and 2001, respectively. He is currently an Assistant Professor in the Electrotechnical Institute, University of Karlsruhe. His special interests are direct converters and control of electrical drives.

8 414 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 49, NO. 2, APRIL 2002 Jörg Weigold was born in Ulm, Donau, Germany, in He received the Dipl.-Ing. degree from the University of Karlsruhe, Karlsruhe, Germany in His practical training concerned matrix converter research. His research interests are signal processing, control theory, and application of microcontrollers. Olaf Simon received the Dipl.-Ing. and Dr.-Ing. degrees from the University of Karlsruhe, Karlsruhe, Germany, in 1993 and 1998, respectively. In 1998, he joined Siemens AG, Erlangen, Germany, where he has been working in the Department for Innovation and Technology for electrical drives. His areas of interest are power electronic devices, new converter topologies, and the impact of control on converter system performance. Michael Braun received the Dipl.-Ing. and Dr.-Ing. degrees from the Technische Hochschule Darmstadt, Darmstadt, Germany, in 1978 and 1983, respectively. From 1983 to 1994, he was with Siemens AG, Erlangen, Germany, where he was engaged in the development of power electronics and electrical drives. In 1994, he joined the University of Karlsruhe, Karlsruhe, Germany, where, since 1994, he has been a Professor and Head of the Electrotechnical Institute. His research areas are power electronics, control of electrical drives, and mechatronics.