DSP implementation for the axial flux motor

Transcription

1 DSP implementation for the axial flux motor J.A.G. Wouters DCT Traineeship report Coach(es): Supervisor: G. Heins Dr. ir. F. de Boer Prof.dr.ir. M. Steinbuch Technische Universiteit Eindhoven Department Mechanical Engineering Control Systems Technology Group Eindhoven, September, 2007

2 ii

3 Contents Abstract v 1 Introduction to the axial flux motor 1 2 System overview 3 3 Pulse width modulator Effort conversion Deadband PWM frequency Analog digital convertor Calibration Noise in current Shielding Calibration circuit ADC frequency Reference generation Position measurement Current reference generator Slow start ADC conversion Current control Controller design Bode diagram of the plant Read data out of the DSP Real time data exchange Store data on memory Conclusion 29 A M-files 31 A.1 Real time data exchange B C-code adjustments 33 B.1 C-code adjustments to write to memory B.2 C-code adjustments to read from memory C Simulink model 37 iii

4 iv CONTENTS

5 Abstract At the Charles Darwin University the goal of their research is to achieve a constant torque with the axial flux motor. To achieve constant torque a specific current reference must be followed, the control of this current is performed by a digital signal processor. To control the current a software model is made which contains the reference generation, analog digital convertor, controller and pulse width modulator. To implement this software model on the digital signal processor Matlab/Simulink is used to create a C-code which is transferred to the digital signal processor by Code-Composer-Studio. The implementation of the software model onto the digital signal processor is considered in this research. v

6 vi ABSTRACT

7 Chapter 1 Introduction to the axial flux motor The axial flux motor is different from conventional electric motors due to the different path of the magnetic flux. In conventional motors coils are oriented parallel with the motor axle. Cylindrical shaped magnets are put around the axle to create a flux that flows radially trough the air gap. In the axial flux motor the flux lines are oriented parallel to the motor axle. To generate those flux lines the coils in the stator are oriented radially, therefore the rotor has the shape of a disk. As result the motor is very short in length and has a much larger diameter than a conventional electric motor. At the Charles Darwin University the goal of their research is to achieve a constant torque with the axial flux motor. To achieve constant torque a specific current reference must be followed, the control of this current is performed by a digital signal processor (DSP). The set up for the control of the axial flux motor consists of power electronics, current sensors, motor and Texas Instruments TMS320F2812 DSP. In this report only the DSP implementation of the software model is treated, the other components are not discussed. The software model, which contains the reference generation, analog digital convertor, controller and pulse width modulator consists of a C-code which is created with Simulink. For the implementation of the model on the DSP the Embedded Target for TI C2000 T M DSP platform of Matlab/Simulink is used. In this platform all the software of the DSP components is put in several Simulink blocks, which are used to create the model. When the model is created in Simulink it will be transferred to the DSP with the use of Code-Composer-Studio. Code-Composer-Studio is the link between Simulink and the DSP. This program converts the Simulink model into a C-code and loads this C-program into the DSP. In chapter 2 of this report an overview of the whole set up is given. All the components that are present in the set up are briefly discussed. The steps that were carried out to achieve a correct implementation of the pulse width modulator are mentioned in chapter 3. The calibration of the analog digital convertor is treated in chapter 4, where also the current noise is reflected. In chapter 5 the implementation of the reference generation is given. Followed by the current control which will be discussed in chapter 6. At last, in chapter 7, two methods are given to get data out of the DSP. Followed by some conclusions and recommendations to improve the working of the torque control. 1

8 2 CHAPTER 1. INTRODUCTION TO THE AXIAL FLUX MOTOR

9 Chapter 2 System overview The set up of the axial flux motor has several components. Together all these components make it possible to control the torque. To achieve constant torque a specific, position dependent, current reference must be followed. To follow the current as well as possible a current control loop is created. A schematic overview of the set up is given in Fig In this figure it becomes clear that there are five main components present in the set up. Figure 2.1: Schematic overview of the set up The first component is the DSP, which is the main component in this report. The DSP, shown in Fig. 2.2, contains the main parts of the current loop. Inside the DSP the reference generation, controller, analog digital controller (ADC) and pulse width modulator (PWM) are present. The reference generation and controller are normal parts of the control loop. For each of the three phases the current is controlled by a single control loop. The ADC converts the voltage signal from the current sensors into a digital signal inside the DSP. The last part on the DSP is the PWM, which converts the digital control signal into a switching voltage output. This voltage signal is the input of the power board, which is visible in Fig

10 4 CHAPTER 2. SYSTEM OVERVIEW Figure 2.2: Photograph of the DSP Figure 2.3: Photograph of the power board

11 5 Figure 2.4: Photograph of the Axial Flux Motor This power board magnifies the voltage signal from the PWM to a maximum of 24 volt and passes that signal on to the axial flux motor which is shown in Fig In this figure the rotor has been removed from the motor, so the stator coils are visible. In the motor the high frequency switching voltage between 0 and 24 volt is smoothed out due to the inductance of the motor. To measure the current that flows through the motor a current sensor is present in each of the three phases. The output of the current sensor is a voltage signal that is fed to the ADC of the DSP. In Fig. 2.5 the electronic board with the current sensors is visible. Figure 2.5: Photograph of the current sensor board The last component in the set up of the axial flux motor is the Labview measuring board. This electronic board is the link between the control loop and Labview, which is used to perform measurements. In Fig. 2.6 a photograph of that electronic board is shown. In this report only the implementation of the DSP is treated, therefore only the parts that are inside the DSP are discussed in the next chapters.

12 6 CHAPTER 2. SYSTEM OVERVIEW Figure 2.6: Photograph of the Labview measuring board

13 Chapter 3 Pulse width modulator The pulse width modulator converts a digital signal into a analog signal with a maximum of 3.3 volts. To create a varying voltage signal the width of the pulses is modulated. The input signal of the PWM has to be given in a number of clock cycles. To make the input signal independent of the chosen PWM frequency a effort conversion is added in front of the PWM block. This will be discussed in section 3.1. If the input signal is zero the pulse is low all the time. When the input signal is at its maximum the pulse is high the whole time, except for the deadband time. Why deadband is necessary will be explained in section 3.2. In between these extreme cases the width is modulated via an asymmetric waveform type. Finally the differences due to varying PWM frequencies will be discussed in section Effort conversion As mentioned before the output signal of the PWM has a maximum of 3.3 volts. This 3.3 volt signal is magnified on the power board. This is done in such a way that the signal is inverted, so if the PWM block has an input of zero the voltage output of the power board is at its maximum. To check whether the PWM works satisfying a sine wave is taken as input for the PWM block. The current output is measured with a current clamp in combination with a oscilloscoop. This measured signal is plotted together with the input signal in Fig When the sine wave is high the current output is low, so there can be concluded that the PWM works all right. PWM Sinus wave Figure 3.1: Testing of PWM 7

14 8 CHAPTER 3. PULSE WIDTH MODULATOR Since the inversion in the power board is annoying another inversion is added inside the effort conversion which counteracts the first inversion. This conversion is created in a way that the input is between 0 and 1000 and PWM frequency independent. Because the current has to be positive and negative two PWM s are necessary for each phase of the motor. One that drives the positive current and another one for the negative current. In this effort conversion the signal of each phase is divided over those two PWM s. In total there are six PWM s necessary to drive the motor, these six PWM s all drive a different module from the power board. The Simulink model of the effort conversion is given in Fig Figure 3.2: Simulink model effort conversion 3.2 Deadband As mentioned before the PWM output is never maximum whole the time due to deadband. Deadband is necessary to avoid overlap at the beginning of particular PWM pair signals. To investigate the influence of this deadband a sinusoidal input is put on the PWM and the output is measured with Labview. A PWM frequency of 75 khz is chosen, which are 1000 clock cycles, due to the fact that the DSP is running at 75Mhz. Because the output consists of pulses with varying width it is difficult to interpret this signal. Therefore not the actual signal is measured, but the time that the PWM is high. This is divided by the total PWM period time. This division is a quantification of the current signal. Those measurements are performed with Labview by recording the time when the PWM output has a rising or falling edge. The period is the time between 2 rising edges and the pulse time is the time between a rising and a falling edge. Each time that there is a rising edge a sample is taken. If samples are taken, the logging time should be exactly one second. In Fig. 3.3 the time signals of the PWM output are visible. With a deadband of 1 it seems like there has been measured for 1 second. When a deadband of 240 or 480 clock cycles is used the logging time becomes longer, although the PWM period is still 1000 clock cycles. So it appears that the deadband time is added to the period time. If the deadband is increased the time between two rising pulses will be larger and therefore the logging time longer. A closer look at the signal with deadband 1 shows that this signal is actually logged for seconds. Another remarkable effect is the amplitude of the signals. These change when the deadband is adjusted, which is logical because the pulse time stays the same while the period time is changed due to the deadband. Consequently the division becomes different which results in different amplitudes, as shown in Fig PWM frequency The PWM frequency has nothing to do with the created signal, it is the frequency of the created pulses. If this frequency is increased the output current gets a smoother shape. The disadvantage of a higher PWM frequency is the increased number of calculations that the DSP has to perform. As long as the frequency is higher than the frequency domain of the driven motor, it will be sufficient.

15 3.3. PWM FREQUENCY db1 db240 db Pwm output [%] Time [s] Figure 3.3: Time signals with varying deadband To check whether it really does not matter which PWM frequency is used some simple experiments are done. As input a sine wave with a frequency of 500 hz is utilized, which is converted in a analog signal by the PWM and measured as described in section 3.2. In Fig. 3.4 fourier transformations of the output signals are plotted. All the signals exist of one main frequency, which is normal. Yet the output signals which are created with a PWM frequency larger than 30 khz do not have a frequency of 500 Hz. If an input signal of 1000 Hz is used the output signals are also all right until the PWM frequency goes above 30 khz. Than the DSP is not able to perform all the necessary calculations in the required time anymore. The time it takes to perform all those calculations increases with a rise of PWM frequency. This explains why the output frequency decreases when the PWM frequency is increased kHz 25kHz 30kHz 35kHz 50kHz 75kHz Y(f) Frequency [Hz] Figure 3.4: Fast fourier transformation of PWM output

16 10 CHAPTER 3. PULSE WIDTH MODULATOR

17 Chapter 4 Analog digital convertor An analog digital convertor is necessary to transform the measured current signal into a digital signal. The current flows through a current sensor which gives an output in volts which is related to the measured current. This voltage is the input of the analog digital converter which converts this voltage signal into a digital signal between 0 and The relation between the current and the digital signal must be known to compare the measured signal with the reference. Therefore a calibration is performed which will be discussed in section 4.1. With the PWM and ADC working it was possible to close the current loop, but than some problems did occur. These problems are discussed in section 4.2. Finally the differences due to varying the ADC frequency are treated in section Calibration The calibration is performed by forcing current to flow through the sensor and through a multimeter. A constant current is put through the sensor and measured with the multimeter. The digital signal is measured with the RTDX for 1 second, the RTDX will be discussed in chapter 7. The average of the measured digital signal is plotted against the current in Fig This is done for currents between -10 and 10 amperes for all of the three phases. For each phase a linear fit is used to describe the relation between the current and the ADC output. These fits are plotted in the figure as well Phase A Polyfit Phase A Phase B Polyfit Phase B Phase C Polyfit Phase C ADC value Current [ma] x 10 4 Figure 4.1: Calibration of the ADC 11

18 12 CHAPTER 4. ANALOG DIGITAL CONVERTOR 4.2 Noise in current With a constant reference there are peaks in the current in closed loop. To detect the origin of these peaks several measurements are done. The voltage of the current sensor is measured with Labview to get a better understanding of the peaks in the signal. The measurements are performed at different PWM and loop sample frequencies and shown in Fig In the open loop case there are no peaks at all. With a sample frequency of 1 khz it looks like the peaks are gone as well, but actually there are just no peaks in that particular second. It is not possible to draw a conclusion from these measurements because the number of peaks are dependent of both the PWM and loop sample frequency. The only certainty is that the open loop case has no peaks. The main difference between the open en closed loop model is the analog digital convertor which is not present in the open loop case. It is possible that the ADC is the origin of the peaks, because a wrong ADC value will result in a bad control signal which results in an incorrect current. Voltage [V] Voltage [V] Voltage [V] CL, f s = 10 khz, f pwm = 5 khz CL, f = 10 khz, f = 10 khz s pwm CL, f s = 10 khz, f pwm = 25 khz Time [s] CL, f s = 1 khz, f pwm = 5 khz CL, f = 1 khz, f = 10 khz s pwm OL, f s = 10 khz, f pwm = 25 khz Time [s] Figure 4.2: Peaks in current Shielding To make sure that there are no peaks in the current, the current sensor is removed from the current loop, so the sensor is measuring zero ampere all the time. The whole model is kept as it was to keep the circumstances the same as they were before. When the main power supply from the power board is turned off the peaks are almost gone, this is visible in Fig If the power supply is turned on again the peaks have a much larger amplitude, which is also shown in Fig This means that the power board introduces the peaks in the ADC signal. To improve this problem the cable from the current sensor to the ADC must be shielded. With the shielded cable the peaks are gone even when the power supply is turned on. The difference between shielded or unshielded cable is plotted in Fig. 4.4, here the power supply is turned on and the sensor is still removed from the current loop. Also when the current sensor is placed back in the loop again, so there is current flowing, the peaks are not present in the current anymore.

19 4.2. NOISE IN CURRENT Sensor output, Powerboard off ADC number Sensor output, Powerboard on ADC number Time [s] Figure 4.3: Difference due to power board 2500 Sensor output, Unshielded cable ADC number Sensor output, Shielded cable ADC number Time [s] Figure 4.4: Difference due to shielding

20 14 CHAPTER 4. ANALOG DIGITAL CONVERTOR Calibration circuit Another improvement can be made by using a calibration circuit [Tessarolo, 2004]. This electric circuit feeds two known constant voltages into two extra ADC channels. One of those voltages is 2.5 volt and the other one is volt. Each time the ADC takes a sample of the sensor it also measures those voltages. With these measured samples it is possible to calculate the output voltage of the current sensor. Because this is done each sample the fluctuations inside the ADC are eliminated. In Fig. 4.5 a schematic overview of the used calibration calculations is drawn Ideal gain Actual gain yh Adc value [ ] yl 0 0 xl xh 3000 Calibrated value [mv] Figure 4.5: Graph of calibration equations The equations to calculate the calibrated output are: where: x = calibrated measurement y = measured ADC output ma = actual gain b = actual offset x = y b ma (4.1) (yh yl) ma = (xh xl) (4.2) b = yl xl ma (4.3) where: xl = known reference low input xh = known reference high input yl = reference low ADC output yh = reference high ADC output In Fig. 4.6 the raw ADC signal and the calibrated ADC signal are plotted. The measured ADC signal has amplitude changes of approximately 35 ADC counts although it generally is a constant signal. The calibrated ADC signal has a reduced amplitude change of approximately 25 ADC counts. Relative to the total range of the signals these amplitude changes have the same magnitude, so the signal is not improving with the calibration circuit. Therefore the calibration circuit will not be used in the current measurement.

21 4.3. ADC FREQUENCY Raw adc signal 2380 ADC [ ] Calibrated adc signal ADC [mv] Time [s] 4.3 ADC frequency Figure 4.6: Difference due to calibration circuit Since there are fluctuations of approximately 35 ADC counts in the measured current signal this will still introduce noise in the current. The only other variable which can be chosen in the ADC block parameters is the sample frequency. It is possible to have a slower or faster ADC frequency than loop sample frequency, therefore a rate transition must be inserted after the ADC. Multiple measurements are done where the current is controlled in closed loop at a loop frequency of 10 khz. The ADC runs at different frequencies which results in different current outputs who are measured with a current clamp in combination with a oscilloscoop. The varying current signals are shown in Fig. 4.7, here it is visible that the best current is generated with the ADC running at 30 khz. The rate transition works as a fast-to-slow transition to connect the ADC with the rest of the loop.

22 16 CHAPTER 4. ANALOG DIGITAL CONVERTOR (a) ADC frequency = 10 khz (b) ADC frequency = 20 khz (c) ADC frequency = 25 khz (d) ADC frequency = 30 khz (e) ADC frequency = 35 khz Figure 4.7: Current controlled at different ADC frequencies

23 Chapter 5 Reference generation In the reference generation part of the model the current reference is constructed. To create a constant torque a specific current shape is necessary. This specific current shape is dependent of the angular position of the rotor. The amplitude of the current determines the velocity. The method to generate the right current reference is not part of this report. The Simulink model to implement the current reference is visible in Fig It exists of 4 components, starting with the angular position measurement, which is discussed in section 5.1. The measured angular position is the input for the current reference generator. This generator uses a lookup table to implement the reference signal, the way this lookup table is implemented in the model is treated in section 5.2. Another aspect to avoid startup problems is the addition of a slow start procedure. This slow start is introduced behind the current generator and will be talked over in section 5.3. The last part in the reference generation is the conversion to ADC values, which is necessary to compare the reference with the measured current. This ADC conversion is discussed in section 5.4. Figure 5.1: Simulink model reference generation 5.1 Position measurement The angular position measurement is performed by an 12 bits encoder, which are 4096 steps per revolution. This encoder is based on the gray system, which is a binary code with the property that only one digit changes each time. With real mechanical switches that measure the encoder it is very unlikely that switches will change states exactly in synchrony. In the transition between two states a misread is possible. Because only one digit changes with the gray system a miss read does not give a great output error in contradistinction to the binary system. This advantage is also a disadvantage of the gray system, because the binary numbers are interchanged. Therefore the order of the binary numbers is not corresponding with the angular position. On the DSP, there is a bus for additional external memory. To read this external memory the DSP sends an address to the external memory and the external memory puts the data stored at that location on the data bus. Since the additional external memory is not used, the data signal from the encoder is connected to this external memory bus. When the encoder data has to be accessed the on board memory sends an address to the external memory. As result the DSP will read the data on the external memory bus, which is the encoder data. 17

24 18 CHAPTER 5. REFERENCE GENERATION Before the external memory puts the data on the bus it has to be asked for this data as explained before. Therefore the DSP sends a pulse to the on-board memory which sends an address to access this data. The on-board memory detects this pulse when it is high for at least 30 ns. In the default case the pulse is only high for 25 ns, so the memory does not detect the pulse. As result the encoder data will not be read. The time of the pulse is increased by changing the C-code manually. Therefore the program is not build and executed at once, but only the program is created in Code Composer Studio. After that we can adjust the C-file MW-C28XX-cls.c by changing the variable XintRegs.XTIMING.bit.X2TIMING from 0 to 1. Now the pulse is high for 50 ns, which is long enough to be detected by the on-board memory. 5.2 Current reference generator As mentioned before the generation of the current reference in not part of this research. The reference is provided as function of the angular position. Because the order of the output from the encoder is not corresponding with the angular position the provided current reference has to converted from binary to gray code order. In Fig. 5.2(a) an example of a current reference is shown as function of the binary number. This reference is stored on the DSP with a lookup table. The table consists of one column which contains the reference data, the input of the table is the index of the column. To convert this signal to gray code numbers the order of the data in the lookup table must be interchanged in the same way as the conversion from binary to gray numbers is performed. Therefore the index of a specific binary number in the gray code order is located and the reference value belonging to that specific number is put on the located index in a new column. When this procedure is performed for all the binary numbers from 0 to 4095 the whole new column is sorted out in gray code order. The resulting lookup table is plotted in Fig. 5.2(b) as function of gray code number Current reference [ma] Current reference [ma] Encoder number [ ] (a) Current signal as function of binary number Encoder number [ ] (b) Current signal as function of gray code number Figure 5.2: Current reference signals The lookup table is stored on the memory of the DSP. When the lookup table does not fit on the internal memory of the DSP it should be stored on the external memory on the DSP. In the beginning of this research the internal memory map was used, but it is necessary to store the lookup table on the external memory. Therefore a specific address on the external memory is allocated to store the lookuptable. When the lookup table fits on the available internal memory the table will be stored on the internal memory although it is set to be written on the external memory. The lookup tables for the three phases will not fit on the internal memory so it is necessary to specify a external memory address. To implement this in the C-code it is necessary to adjust the code manually. Later on in this research it became clear that the full memory map had to be used because the ADC did not work on the internal memory map. If the full memory map of the DSP is used, the external memory of the DSP will be used automatically. Then it is not necessary to adjust the C-code manually.

25 5.3. SLOW START Slow start Due to startup problems an addition of a slow start procedure is necessary. When a large current reference is requested the power supply is not able to provide the current necessary to startup. As a result the motor does not rotate and there is a possibility that the power board will break down. If the requested current starts small and is increased slowly the power supply is able to provide the necessary current. To create this slow start the large current reference is multiplied with a factor, which is plotted in Fig Multiplication factor [ ] Time [s] Figure 5.3: Multiplication factor to create slow start 5.4 ADC conversion The last component in the reference generation part of the model is the conversion from milliamperes to ADC-numbers. To construct a correct error signal the reference and measured current must have the same units. When the calibration circuit from section is used the measured current has the unit of milliampere. Then it is not necessary to convert the reference signal into ADC-values. However it has been decided not to use the calibration circuit, so the output of the current measurement is given in ADC-numbers. Therefore the reference also has to be converted into ADC-numbers. The conversion from milliamperes to ADC-numbers is performed by a multiplication with the linear fits from Fig With this conversion the requested reference is transformed into ADC-numbers to compare the reference with the measured current.

26 20 CHAPTER 5. REFERENCE GENERATION

27 Chapter 6 Current control In this chapter the establishing of the controller for the current loop is treated. When the transfer function between voltage and current is considered the axial flux motor is a first order system, therefore a start is made with a proportional controller. The consecutive steps to create a better controller are described in section 6.1. To create a good controller which improves the performance it would be pleasant to know the characteristics of the plant. Therefore the exact transfer of the plant must be determined, the steps that were taken to determine this transfer are discussed in section Controller design To test the DSP implementation of the closed loop model it is necessary to use a reasonable controller. This controller must be able to make the current follow the reference shape. The controller does not have to be able to suppress the noise, because it is possible that there are still components present who create the noise. As reference a square block wave with a period of 0.2 seconds is chosen. As mentioned earlier a proportional controller was initially used because the axial flux motor is a first order system. With a proportional gain of 1 the current follows a block wave but does not reach the specified amplitude. Increasing the proportional gain results in an amplitude rise. With a gain of 2 the current follows the reference. The disadvantage of increasing the proportional gain is the amplification of the noise. In Fig. 6.1 the controlled current is shown, there is so much noise present that it is hard to use this controller to test the implementation of the other components. Figure 6.1: P controlled current 21

28 22 CHAPTER 6. CURRENT CONTROL If the proportional gain is decreased the amount of noise will reduce, but than the reference will not be followed. This can be overcome with the use of a integral controller. In the beginning a PI controller with proportional gain of 0.3 and integral gain of 3 is taken. The resulting current shape is plotted in Fig The integral action causes the current to grow to the specified amplitude, but this takes too much time. An increase of integral gain results in a faster reach of the amplitude. With a integral gain of 100 and a proportional gain of 0.1 the current follows the reference better, this is visible in Fig Now a mainly integral controller is able to control the current, which can be used to test the implementation of the other components. Figure 6.2: PI controlled current, small integral action Figure 6.3: PI controlled current, larger integral action 6.2 Bode diagram of the plant For the construction of the bode diagram of the plant multiple experiments are done. Several sine waves are used as input for the plant, starting with a sine wave of 10 Hz. Continuing with a sine wave of 100, 200 and so on until 1000 Hz. Each frequency constructs a point of the bode diagram by measuring the input and output sine wave. The difference in amplitude of the wave determines the magnitude of the bode plot and the shift in phase determines the phase of the bode plot. The input of the plant, which is the output of the PWM is measured with the method described in section 3.2.

29 6.2. BODE DIAGRAM OF THE PLANT 23 The output is the current that flows through the motor, which is measured with Labview as well. The samples are taken when the PWM has a rising edge. The measured sine waves are fourier transformed to calculated the magnitude and phase for one frequency of the bode plot. This is done for all the frequencies to construct the whole bode diagram. The constructed bode diagram is plotted in Fig The magnitude plot looks like a first order system, but the phase is very strange. It seems like the fourier transformation failed with respect to the phase. Therefore we constructed the bode diagram also by hand. Both of the time domain signals were shown in one plot and by measuring the length of the period and the phase delay the phase plot could be verified. Also the magnitude plot was verified by measuring the difference in amplitude of the both time domain signals. The manual made bode diagram is plotted in Fig. 6.4 as well, now the magnitude is the same as with the fourier method but the phase is different. Nevertheless the phase still has unexplained variations. Magnitude [db] fft method hand method Phase [deg] Frequency [Hz] Figure 6.4: Bode diagram axial flux motor When the experiments were done for a second time it appears that the phase is totally different. Also the manual made bode diagram is different than the one which was constructed the first time. It appears that the phase shift is caused by a synchronization problem from Labview. Both of the measured signals were not sampled at the same time, so it is impossible to construct a proper bode diagram by this measuring method. An attempt to solve the synchronization problem of Labview did not succeed. To check whether the phase is indeed as expected for a first order system another measuring method is used. Now the PWM output is just measured as a analog signal on a time based frequency, instead of measuring the time between the rising and falling edge. This time based analog signal consists of pulse of 24 volt with varying width. To construct a sine wave out of this analog signal the rising and falling edges have to be detected in the data. This is done by comparing the nearby data points with each other. If there is a larger difference between those data points a edge is detected. Because the PWM has a frequency of 10 khz and there is measured with a sampling frequency of 750 khz each PWM period contains only 75 data points. Due to the fact that the amplitude is only 25 procent of the total range at its maximum and 5 procent at its minimum, a measured pulse contains minimal 5 and maximal 20 data points. Therefore the constructed sine wave does not have a smooth shape. In Fig. 6.5(a) the constructed PWM sine wave of 10 Hz is visible together with the current. The current is measured with a time based sampling frequency instead of a trigger based frequency. Due to this time based measuring the current is not as smooth as it was before. As expected the phase delay is now 0 degree. For the 800 Hz signal the phase delay is 90 degrees, which is visible in Fig. 6.5(b). This is again as expected, so the plant is indeed a first order system. The method which compares the input and output sine waves gives a global bode diagram, but it is a very rough one, because it is only done for eleven different frequencies. To create a smoother bode diagram a step input is put on the system. Looking at the response of that step a bode diagram can be constructed. Because the plant is a first order system only the parameters of the transfer function G(s) = k τs+1 have to be tuned. This is done as well as possible to get the same step response with the

30 24 CHAPTER 6. CURRENT CONTROL 8 6 pwm current 8 6 pwm current (a) 10 hz sinus (b) 800 hz sinus Figure 6.5: Input PWM and output current theoretical model and the actual plant. In Fig. 6.6 the step responses of the theoretical model and the actual plant are shown. With this model the bode diagram of Fig. 6.7 can be made, which is comparable with the bode diagram which was constructed before. This bode diagram shows the transfer from the output of the PWM to the actual current. The controller is working inside the DSP and does not work with this in- and output. Several transitions are made in the PWM and ADC which effect the in- and output. To make sure that these transitions contain only amplifications and do not have dynamics inside them the data from the DSP must be read out. In Chapter 7 will be explained how it is possible to get data out of the DSP I actual I model x 10 3 Figure 6.6: Step response 1 st order system

31 6.2. BODE DIAGRAM OF THE PLANT Sine method Step method 8 Magnitude [db] Frequency [Hz] Figure 6.7: Comparison bode diagram sine wave method and step method

32 26 CHAPTER 6. CURRENT CONTROL

33 Chapter 7 Read data out of the DSP In this chapter will be explained how data can be read out of the DSP. There are two methods which can be used to do this. The first one is the RTDX, i.e. real time data exchange. This is a function that creates a link to send the data back to the computer in real time. With a Matlab script this data can be read from the DSP and written to a variable in the workspace. How this is done is treated in section 7.1. The second method to read data out of the DSP is basically split into two parts. First the data will be written to the external memory during the experiment. After that another Simulink program downloads the data from the memory to Matlab with a RTDX function. The way this is implemented in Simulink will be discussed in section Real time data exchange The Real Time Data Exchange function creates a link between the DSP and the computer. Through this link data can be exchanged while the program is running. In the Simulink model the signals that we want to know are connected to a mux and the output of the mux is connected to the RTDX. The model is working normally until the Matlab script starts, then a channel is opened and the data exchange is started. The sampling frequency of the Simulink model determines the frequency of the data exchange. The Matlab script prescribes the number of samples that have to be taken and therefore the logging time. The Matlab script for the real time data exchange is given in appendix A.1. Eventually it became clear that the RTDX did not work in real time. The data which was logged with the RTDX suggests that it works all right, but in reality the sampling time became larger. This came into sight when an open loop sine wave was logged. On the electricity meter of the power supply it is possible to detect the generated sine frequency. When the logging started this frequency became less. This means that the RTDX did affect the working of the whole model and therefore could not be used to read data out of the DSP. 7.2 Store data on memory To transfer the data from the DSP to Matlab it is necessary to use the RTDX, but this function cannot be used while the program is running. Therefore the data is stored on the DSP while the program is running and transferred to Matlab later on. To store the data on the memory of the DSP the signals that we want to know are connected to a To Memory block. This block allocates the memory address where the data has to be written to. In another Simulink model a From Memory block is used to read the data from the memory and transfer it to Matlab with a RTDX function. 27

34 28 CHAPTER 7. READ DATA OUT OF THE DSP A To Memory block can only write to one allocated memory address, so it will overwrite itself each time. This problem is solved by adjusting the c-code manually. An extra variable is created to write the data to, this variable is not allocated to a specified memory address, but increments its address each time a sample is taken. In this way the address where the data is written to changes each time, so multiple data points could be written to the memory. A counter is added to the c-code which makes it possible to determine the start and stop time of the logging process. It is preferable to wait with logging until the motor is started up. Chosen is to start logging after 14 seconds and log for 1 second. Because the logging frequency is 5 khz, data has to be written to the memory while the counter is between and In appendix B.1 the adjustments of the c-code are gone through as well as the way to adjust the code. When the data is stored on the memory another model transfers the data from the memory to Matlab. To read out the memory a From Memory block is used, but this block can only read out one memory address. To read out multiple memory addresses the c-code has to be changed manually. Again the address is incremented each time a data point is read out. How the c-code has to be adjusted is treated in appendix B.2.

35 Chapter 8 Conclusion A DSP is used for the torque control of the axial flux motor. The C-code implementation of the software on the DSP is performed by Code-Composer-Studio which converts the Simulink model into a C-code. The construction of the software that is necessary to control the axial flux motor is treated in this report. To control the torque a specific current has to be followed, therefore a current control loop must be created in Simulink. The control loop consists of a current reference that is implemented by a lookup table. This lookup table uses the angular position as input and the output is the position dependent current reference. The angular position is sampled at the same rate as the control loop, 10 khz, with the use of a "From memory" block. The current reference is controlled by a PID-controller, which is a standard Simulink block from the Embedded Target for TI C2000 T M DSP toolbox. To measure the actual current the output of the current sensor is converted into a digital signal with the ADC of the DSP. The use of an external calibration circuit which feeds two known constant voltages into extra ADC channels did not eliminate the fluctuations inside the ADC. Changing the sampling frequency of the ADC improves the working of the ADC, the best output is present when the ADC runs at 30 khz. A rate transition reduces the sampling frequency back to the 10 khz loop frequency, so it works as a fast-to-slow transition. The output of the controller is the input of the PWM which varies the voltage output of the DSP. This voltage is transformed into a current inside the power board. The PWM frequency has to be at least the same as the loop frequency. The PWM works all right when the frequency is below 30 khz, until that frequency the DSP is able to perform all the calculations in time. Above this 30 khz the frequency of the modulated signal does not have the same frequency as it should be. A current control loop is created by the combination of all these components. With the right lookup table implemented for all the three phases the axial flux motor is now rotating. Measurements of the velocity with Labview show that the motor is not rotating constantly enough. Therefore the controller has to be optimized and more noise reduction is necessary. To facilitate the optimization of the controller two methodes are investigated to read data out of the DSP. The best method to read data out of the DSP is basically split into two parts. With this method the data is stored in real time and transferred to Matlab later on. 29

36 30 CHAPTER 8. CONCLUSION

37 Appendix A M-files A.1 Real time data exchange clear all ; % DISABLE RTDX to make sure that there is no previous data % Construct the link between target board and processor to avoid an error % just incase it has already been closed cc= ccsdsp ( boardnum,0) % disable the RTDX channels cc. rtdx. disable ( all ); % disable RTDX cc. rtdx. disable ; % close all RTDX channels, cc. rtdx. close ( all ) % make sure CCS IDE is visible if cc. isvisible, cc. visible (1); end % close all your remaining links to CCS IDE and release all the % associated resources. clear ( all ); % Calls the link destructors to remove all links % DEFINE VARIABLES TotalSamples = 5000; % ENABLE RTDX % Construct the link between target board and processor cc= ccsdsp ( boardnum,0) % enable the RTDX interface cc. rtdx. enable ; % Define and enable read channels cc. rtdx. open ( ochan, r ); cc. rtdx. enable ( ochan ); % Configure buffers in RTDX to store the data until MATLAB % can read it into workspace. cc. rtdx. configure (1024,1); % Check link configuration cc. rtdx 31

38 32 APPENDIX A. M-FILES % Check that channels are enabled ochan_enabled = cc. rtdx. isenabled ( ochan ) disp ( logging data to buffer... ) % WAIT % Check the number of available messages to read from the target. num_of_rtdx_data = cc. rtdx. msgcount ( ochan ); while num_of_rtdx_data <= TotalSamples num_of_rtdx_data = cc. rtdx. msgcount ( ochan ); end % disable the RTDX channels - hope this works disp ( disabling channels... ) cc. rtdx. disable ( all ); % read in the data disp ( transferring data to PC... ) for i = 1 : TotalSamples RTDX_vector = cc. rtdx. readmsg ( ochan, int16,[1 3],1); RTDX_data (i,1) = RTDX_vector (1); RTDX_data (i,2) = RTDX_vector (2); RTDX_data (i,3) = RTDX_vector (3); end % save the data to a file disp ( saving data to a. mat file... ) save RTDX_data. mat RTDX_data % CLEAN UP RTDX disp ( cleaning up RTDX... ) % Empty the remaining messages from the queue and verify that the queue is empty. cc. rtdx. flush ( ochan, all ) num_of_msgs = cc. rtdx. msgcount ( ochan ) % disable the RTDX channels cc. rtdx. disable ( all ); % disable RTDX cc. rtdx. disable ; % close all RTDX channels, cc. rtdx. close ( all ) % make sure CCS IDE is visible if cc. isvisible, cc. visible (1); end % close all your remaining links to CCS IDE and release all the % associated resources. clear ( all ); % Calls the link destructors to remove all links

39 Appendix B C-code adjustments B.1 C-code adjustments to write to memory Adjusting the c-code is necessary to store the data on the memory of the DSP. Some variables are added to the program, which have to be initialized at the top of the c-code. This is done by adding a Custom block in the Simulink model which adds the following lines at the top of the code. Uint32 delay = 0; int16_t * ownmemind = ( int16_t *) U; int16_t * ownmemind1 = ( int16_t *) U; int16_t * ownmemind2 = ( int16_t *) U; The variable delay is the counter which makes it possible to determine the start and stop time of the logging process. The global variables ownmemind, ownmemind1 and ownmemind2 are allocated to a specific address on the external memory. To write the data to the memory the code which is present in the loop of the To Memory part has to be adjusted. In the unchanged c-code the address of the variable memind is reset every loop cycle. Therefore it will overwrite the same address each time. To overcome this problem a self made variable, called ownmemind is introduced. The address of this variable is not reset each time, so the address will increment each loop cycle. In this way more data can be stored on the memory. The data writing to the variable ownmemind is only done when the counter delay has a value between and Therefore the counter is incremented each time at the beginning of the loop. For the variables ownmemind1 and ownmemind2 the same procedure is used to write the data to other parts of the memory. The code which is put inside the loop of the To Memory blocks is shown below. These extra lines are added to the code using the Manual code option of the To Memory blocks. /* S- Function Block : <Root >/ C28x To Memory ( smemsnk ) */ { /* Memory Mapped Output */ const int16_t * memind = ( int16_t *) U; /* custom code to increment address */ delay ++; } *( int16_t *)( memind ++) = ( int16_t ) rtb_i16_temp35 ; if ( delay > 70000) if ( delay < 75000) *( int16_t *)( ownmemind ++) = ( int16_t ) rtb_i16_temp35 ; 33

40 34 APPENDIX B. C-CODE ADJUSTMENTS /* S- Function Block : <Root >/ C28x To Memory1 ( smemsnk ) */ { /* Memory Mapped Output */ const int16_t * memind = ( int16_t *) U; /* custom code to increment address */ } *( int16_t *)( memind ++) = ( int16_t ) rtb_i16_temp35 ; if ( delay > 70000) if ( delay < 75000) *( int16_t *)( ownmemind1 ++) = ( int16_t ) rtb_i16_temp35 ; /* S- Function Block : <Root >/ C28x To Memory2 ( smemsnk ) */ { /* Memory Mapped Output */ const int16_t * memind = ( int16_t *) U; } /* custom code to increment address */ *( int16_t *)( memind ++) = ( int16_t ) rtb_i16_temp35 ; if ( delay > 70000) if ( delay < 75000) *( int16_t *)( ownmemind2 ++) = ( int16_t ) rtb_i16_temp35 ; B.2 C-code adjustments to read from memory To read the data from the memory the same problem occurred as when the data is written to the memory. With a To memory block the address is reset each time, so this block can only read out one address. To avoid this, the line which resets the address has to be placed outside the loop. In all the three From Memory blocks a local variable memind is used for the address. If the reset line is placed out of the loop this local variable has to be changed into a global variable. For all the loops another global variable name is needed. This code is adjusted manually in Code Composer Studio, after Simulink has created the c-code. First the initialization of the global variables at the beginning of the code is done with the next lines. const int16_t * memind = ( int16_t *) U; const int16_t * memind1 = ( int16_t *) U; const int16_t * memind2 = ( int16_t *) U; In the loop of the From Memory blocks the line that resets the address is deleted and the name of the local variable is changed into the corresponding global variable. Inside the loop the code looks like this. /* S- Function Block : <Root >/ C28x From Memory ( smemsrc ) */ { /* Memory Mapped Input */ } readoutmemory_b. C28xFromMemory = *( int16_t *)( memind ++); /* S- Function Block : <Root >/ C28x From Memory1 ( smemsrc ) */ { /* Memory Mapped Input */