Optimizing Performance in Switched Reluctance Drives J. Reinert R. Inderka M. Menne R.W. De Doncker Institute for Power Electronics and Electrical Drives Aachen University of Technology, Jagerstr. 17-19,52066 Aachen, Germany www.rwth-aachen.de/isea Abstract - It is demonstraited how the overall performance of a switched reluctance drive can be improved by using different control strategies for different regions of the torque-speed diagram. For a 4-phase 30 kW machine these strategies are fitted by connecting different optimizing routines with a simulation program. The importance of a correct tuning of the control parameters like turn-on angle, turn-off angle, reference current and size of the hysteresis band is demonstrated for a current controlled drive. Furthermore, the test bench and the measurement procedure for the experimental verification and comparison to the simulation are presented. I. INTRODUCTION Due to their rugged brushless design switched reluctance (SR) drives combine a high reliability and an outstanding performance over a wide speed range. The SR-machine has been built for drives ranging from a few watts up to 300 kW. Taking into account some detailed design and control considerations, the SR machine is ideally suited fix high performance applications at low costs. The performance of an elec.trica1drive can be characterized by efficiency, torque, torque-ripple, noise, power-weight ratio and power density. The importance of each of these criteria is weighted according to the drive application. In addition, having a specific application, the weight of each criterion can change over the working area of the drive. An optimized performance of a drive can only be reached with a correct balance of the criteria, because the improvement of one criterion will most surely worsen another one. Analyzing the losses in SR machines is complicated, because of the non-sinusoidal waveforms of the flux-linkage as well as current and voltage. A low volt-ampere requirement can only be reached if the iron is subjected to substantial saturation. Therefore, the core is expo!jed to high average remagnetization velocities and extreme non-linearities. The optimization of the performance of the drive is based on the calculation of the losses in the machine and converter. In optimizing a certain performance criterion, i.e. torque ripple or efficiency, it is absolutely necessary to consider the machine design, the controller abilities and switching strategies together. A low torque ripple for example can be obtained by either 0-7803-4340-9/98/$10.00 (3 1998 IEEE. implementing current profiling [ 11 or by adapting the design of the laminations [2], but even better, by a combination of both. In addition, it is advantageous to have a flexible controller structure as different control strategies can be implemented to optimize further drive performance. With different values of the control parameters turn-on angle, turn-off angle, reference current and size of the hysteresis band ireDAi), it is possible to obtain the same operation point (torque-speed) thereby influencing the performance criteria, such as efficiency, torque ripple, noise, etc. In this paper, a prototype drive (4 phase, 8/6 configuration) for an electric vehicle (EV) is presented, where the efficiency is the main performance criterion. However, the torque ripple, particularly at low speeds, is also important. Attention is drawn to the implementation of different controller strategies and their effect on the drive performance, while having a fixed hardware design of machine and controller. During the design stage, however, the interaction of machine geometry and controller structure was considered. Section I1 of this paper treats the requirements to EV-traction drives and the consequences to the control strategies of a SRdrive. The selection of these control strategies for different operating regions in the torque-speed diagram is also discussed. To test different control strategies a specialized simulation program was constructed which is outlined in section 111. The simulation results and the implementation of the optimizing strategies are discussed in section IV. The measurement methods and the test bench to verify the simulation results are presented in section V of the paper. 11. DESCRIPTION OF CONTROL STRATEGIES In many applications, the drive is operated over the entire operating area as, for example, in EV-traction drives. The converter and the machine are subjected to a wide range of requirements, with the associated efficiencies for each working point. For this application, the energy consumption of a specific driving cycle has to be minimized and it is therefore necessary to implement control strategies with which the optimal drive efficiency qd is reached for each working point. Although the efficiency is a dominant criterion for an EV-drive over the entire operating region, other drive performance criteria, as for example the torque ripple factor, 765 - - : " . E have a larger weight in some of the operating areas. Evaluating the weight of different criteria leads to distinguished areas over the torque speed range, as shown in Fig. 1 (here only the first quadrant is shown). The boundaries between these areas are obviously not as clear as in the figure, but can be identified satisfactorily. \ 200 , I 7 80 3 U 40 0 10 20 30 Rotor angle (deg) 40 50 60 0 10 20 30 Rotor anglc (dcg) 40 50 60 \ ! V i Fig. 2: Fig. 1: Identified areas of different control strategies Different switching strategies As mentioned before, it is compulsory in the design of high performance SR drives to consider the machine design (lamination geometry and thickness, air gap, phase- and pole number), the power converter (topology, semiconductor devices) and the flexibility of the controller (microcontroller or DSP) together. For the EV-application described here it was found, that a good torque ripple factor at satisfactory efficiency can best be achieved with a four phase machine. Once having designed the machine and converter and having obtained the drive criteria arrangement, the switching strategies of the controller guarantee the highest degree of freedom to optimize performance. In area1 a low torque ripple is required, to prevent speed oscillations during low-speed operation. A low torque ripple, while using current controlled operation, can be achieved by additional profiling of the phase current waveforms to obtain a smooth total instantaneous torque as seen in Fig. 2. This current profiling can reduce the efficiency considerably depending on the number of poles and phases of the SRM. Therefore, it is important to adapt the weights of the performance criteria so as to optimize the efficiency, while keeping the ripple below a predefined percentage. Example of operation point with torque ripple factor kpipp=28.5 %, N = 500 rpm; T = 68.3 Nm The difference between I1 and I11 is that the drive operates with a conventional current fed chopping mode in area I1 while it is in single-pulse mode in area 111. Therefore, in region I1 the control parameters, i.e. turn-on (O,,) and turn-off (e,,) angles, reference current iref and size of hysteresis band Ai can be set for each working point in the torque-speed diagram to obtain the maximum efficiency. In area111 the performance can be optimized only by changing e,, and e,, At higher speeds and lower loads as in area IV a technique has been found to minimize motor losses. It corresponds to a region where the material starts to saturate. By switching only every second phase of the machine these active phases are carrying higher currents, thereby driving the material deeper into saturation, as seen in Fig. 3. 0.2: -s" In contrast to area I, the importance of a low torque ripple factor in area11 is reduced. At higher speeds, a high torque ripple does not lead to noticeable speed oscillations, due to the kinetic energy in the system, which is proportional to the square of the speed. In the regions 11, I11 and IV the efficiency is the key criterion for the drive performance. 766 I 1 I 10 20 30 I I I I I 60 70 80 0.; h I m ) d .^ 0.1: X Ls 0. I 0.0' ( Fig. 3: 40 50 Current (A) 4-phase versus 2-phase operation at N and T,,, = 17 Nm = 4300 rpm After the saturation poinit the iron losses of one excitation do not increase significantly. However, due to the fact that only every second phase is activated, the overall iron losses are reduced considerably. The overall copper losses increase slightly due to their squared dependency on current. Simulation and measurement have shown, that the total losses in the machine for this region ciln be reduced by up to 4%. Since the drive of an EV is regularly being driven in this region, an increased drive efficiency lowers the energy demand notably. The high torque ripple piroduced during this operation can be tolerated because at these speeds the inertia of the mechanical system cancels out unwainted speed oscillations. Fig. 3 shows an operating point of 4700 rpm for the operation with four phases (smaller loop) and with two phases active (bigger loop). The torque produced with one phase is doubled by going from the 4-phase to the 2-phase operation. No chopping occurs in the 2-phase operation, due to the fact that the induced voltage equals roughly the supply voltage minus the resistive drop over the phase windings in lhis operating point. This mode of operation is not possible at high speeds, see area IV in Fig. 1, because the induced voltage is to high. Iv. IMPLEMENTATION OF OPTIMIZING STRATEGIES AND SIMULATION RESULTS To reach an optimal drive performance over the entire torquespeed range as discussed in section 11, the control parameters can be precalculated with a simulation program or, alternatively, adjusted on-line as done in [4]. In this section a method is outlined which optimizes the control parameters in advance by simulation. The optimized values can then be stored as look-up-tables in the microcontroller unit. For the generator operation the values of the control parameters have to be optimized in a similar way. In optimizing the parameters for the different areas two strategies can be defined. One strategy for area1 of Fig. 1 where the torque ripple factor has to be minimized and the other for the areas 11, 111, IV where the drive efficiency is maximized. 111. SIMULATION The use of an accurate sirnulation program in the design of SR drives is indispensable. The simulation as well as the optimization method is implemented in MatlabTM.With this custom tailored program it is possible to fit the structure and program features to specific needs. As basis of the simulation program the magnetic properties of the machine are used. These can be obtained by a finite element analysis or by measurements on existing machines. The simulation does not only calculate the machine behavior, it also considers the different loss mechanisms in a SR drive. For the calculation of the iron losses a novel method has been implemented [3] using the average remagnetization velocity. The losses in the power devices are calculated b s e d on measurements of 600Vl600A IGBT devices used in the converter. Information from the data sheets was not used as this tends to be somewhat optimistic with respect to conduction losses. The calculated losses occurring in the machine and the converter are used in the optimization routine. With this procedure a fast calculation for different operating situal ions is possible. These simulation techniques offer a considerable time reduction compared to a pure experimental method, because the measurements need to be carried out in fewer predefined operating points. The drive parameters for an optimiil machine behavior can readily be determined. Changes in the controller hardware can be evaluated and considered beforehand. The different switching strategies are simulated and analyzed, which can be obtained experimentally only with c;onsiderable effort. 767 Rotor angle (de@ Fig. 4: Torque production capability for three phases. Contours of constant current are shown In Fig. 4 the torque production capability of the drive is depicted for 3 phases with 3 different constant current values each. In a 816 configuration, 24 excitation-pulses per mechanical revolution are needed (4 phases times 6 rotor poles) and thus each pulse should contribute to torque over 15". The rotor and stator poles have pole arcs of 20" each. However, due to the leakage flux a positive torque can be developed over 30' with one phase. Therefore two phases could always contribute to the average torque. It can be seen in Fig. 4 that at 0" phase 1 is in the aligned position (not able to produce a positive torque) while the phase 2 already has a high torque production capability. Minimizing the torque ripple the currents in the two phases during a commutation process are calculated such that the best overall efficiency is obtained. By consecutively adjusting a pair of currents the sum of the torque produced with each phase is held equal to the average torque. This simulation is repeated for each rotor positions. To reduce calculation time the current in the phase with the higher torque production capability is always increased first while the other current is decreased. At the same time, the rate of change of current which is possible at the corresponding speed and rotor position is considered in the calculation. the energy ratio [5] is maximized. The higher copper losses, caused by the very high current in comparison to the two other waveforms, are more than compensated. The torque ripple is of course much worse than that of the other curves, but this performance criterion has no weight in these regions. The low tum-on angle and the restricted current maximum of curve B are the reasons for the poor efficiency, because a lot of reactive power is needed between the angle of 4" and 9". The value of the tum-on angle at waveform C is too high to build up the current to its maximum level when the rotor reaches the position with the highest torque-current ratio. Nevertheless, the efficiency of control method C is still higher than with control B because the reactive power is much smaller. In Fig. 2 an example is given. It can be seen that the current increases at the end of the conduction phase, thereby exploiting the superior torque production capability of this phase compared to the following phase for the corresponding rotor positions. Here a specified torque ripple was tolerated in order to keep an acceptable efficiency. For the remaining areas of Fig. 1 the key performance criteria is efficiency. Finding the best value for the control parameters, provided speed and voltage are constant, is difficult because it possible to attain the same working point with different control parameters. For example to obtain the working point 13.5 Nm 4000 rpm three different simulated current waveforms are shown in Fig. 5. This example shows the procedure for optimizing one performance criterion. Together there are four control parameters to be adapted, all having an influence on the drive behavior. A consideration of all the possibilities to change the drive performance leads to high calculation times. To reduce the calculation time it is advantageous to use a gradient-type algorithm. All gradient-type algorithms require suitable starting values for the control parameters. To obtain good starting values a prestudy of the influence of the parameters on the performance criteria must be made. The influence of the tumon and tum-off angles on drive efficiency and torque is presented in Fig. 6 and 7. - 0 I Fig. 5 : Type A B C I 30 40 Rotor angle (deg) I 50 60 I I I I I PI I I 30 I-^ 40 Rotor angle (deg) 50 - I I I 20 10 I I I 20 10 I \/ 0 I I * I In these two figures all control parameters (voltage, reference current, hysteresis band) and speed are constant. By changing the turn-on and turn-off angle a maximum of efficiency can be found (eon= 6"; eOff = 21") in Fig. 6 . In comparison to Fig. 7 it is obvious that the torque is increasing by shifting the tum-on angle to lower rotor angles than 6 degrees. Consequently, a larger angle produces a higher average torque with worse ratio of torque to current. At lower rotor angles a lot of reactive power is needed leading to a lower efficiency. 60 Three currents and resulting torques producing the same average torque of 13.5 Nm at N = 4000 rpm eon eoff irefinax 3 4 18.5 23 115 - 39 35 22 123 I- irefmin Taw 13.4 13.5 113.6 qd 93.4 88.8 190.1 ~~ ~~~~ ~ Table 1: Control parameters and corresponding efficiency for the current waveforms of Fig. 5 In Table 1, the control parameters e,,, irefmax, irehln and the two drive criteria efficiency and average torque are presented. Of all three waveforms shown the best efficiency is reached with the parameters of curve A. Indeed, with these parameters Fig. 6: 768 Efficiency versus e,,, eOff at 3000 rpm. Other control parameters are constant L" 10 Fig. 7: " Torque versus turn-on and turn-off angle at 3000 rpm. Other control parameters are constant Fig. 9: To obtain a higher average torque it is advantageous to increase the reference current as long as possible and if the maximum current (mostly restricted by the converter or by the maximum winding temperature) i5i reached the turn-on angle must be reduced. However, as shown before, another reference current or hysteresis band with other switching angles may lead to higher efficiency at the same torque. Hence a consideration of all four parameters is necessary for the current feed operation mode. This typical trade-off between efficiency and torque is also shown in Fig. 8 and 9 where the reference current and the turnon angle are the two changeable parameters while keeping the other parameters constant. It can be seen that at low currents the turn-on angle has no big influence on the torque or efficiency. This changes considerably at higher currents, where a precise turn-on angle is required. By a gradient-type algorithm the performance criteria in each defined area (Fig. 1) are optimized. The simulation results are verified by experiments. These are carried out on a test bench described in the next section. V. TESTBENCH To verify the simulation results and to optimize the control strategy, the prototype 4-phase machine (8/6) with a nominal power of 30 kW is installed on a test bench [6] as shown in Fig. 10. Setting the speed of the load machine and the torque of the SRmachine via a master computer any point in the operating region can be reached. By means of torque- and speed measurements the mechanical output power is obtained. The input power delivered to the entire drive as well as the motor are measured via a high precision power meter. Therefore, the efficiency of the machine, the converter and the overall efficiency can be determined and compared to the simulation results. 0.95 6 .-3 0.9 ;. n 0.85 Fig. 8: Torque versus current reference and turn-on angle at 3000 rpm. Other control parameters are constant Efficiency versus current reference and turn-on angle at 3000 rpm. Other control parameters are constant An oscilloscope with 2,5 GSampledsec connected to the evaluation unit is also recording the torque of the drive versus time. This measured value can be compared with the simulation results. The torque-meter has a bandwidth of up to 1500 Hz. Hence, the torque-ripple of frequencies up to about 500 Hz in the low speed region can be measured. 769 REFERENCES N' oscilloscope 4200rpm 550 Nm evaluation 100 kW L I 1 M. Alakiila, L. Sjoberg, P. Johansson: A 40 kW Switched Reluctance Engine StartedGenerator System f o r an Electric Hybrid Vehicle, EPE' 97, Vol. 4, pp. 717 - 720. 300V DC 300A battery current battery voltage master comwler Fig. 10: '' I 121 T. Schencke: Drehmomentenglattung von geschalteten Reluktanzmotoren durch eine angepaJ3te Blechschnittgestaltung, Dissertation TU Ilmenau, February 1997. [31 J. Reinert, R. Inderka, R. W. De Doncker: A Novel Method for the Prediction of Losses in Switched Reluctance Machines, EPE '97, Vol. 3, pp. 608 - 612. [41 Kjaer P.C., Nielsen P., hdersen L., Blaabjerg F.: A new Energy Optimizing Control Strategy for Switched Reluctance Motors, Proc. of APEC '94, pp. 48-55, 1994. Test bench for measurements The switching strategies influence the efficiency of both the machine and the converter and also the torque-ripple. Each of these influences can be detected by the measurements described above. Different switching strategies can therefore be tested in the whole operating region of the drive. The best switching strategy can be chosen and the area of its implementation defined. [51 T. J. E. Miller: Converter Volt-Ampere Requirements of the Switched Reluctance Motor Drive, IEEE Transactions on Industry Applications, Vol. U-21, No. 5, pp. 1136-1144, [61 P. Mauracher: System-Optimization of the Drive Train of Electric Vehicles to Reduce the Energy Consumption, EVS 13, Vol. 1, pp. 70 - 77, 1996. 1985. Furthermore, the control parameters of the SR drive can be changed in each operating point. This tuning is implemented on-line, while the test bench is running, by means of the user interface of the SR drive's control unit. Thereby, the optimal control parameters obtained by simulation can be verified. VI. CONCLUSION In this paper a technique to optimize performance of switched reluctance drives is presented. The strategy is implemented on a 4-phase 30 kW drive for electric vehicles. The operating area of the drive is divided into four regions, where different control strategies have to be adapted in order to maximize a chosen performance criterion. It is shown how these criteria can be influenced by adapting the control parameters e,,, 8,ff, iref,Ai. In helping to find the best possible combination of control parameters for each working point a optimization program was developed and successfully implemented. 770