Get Complete Project Material File(s) Now! »
Measurements and Testing
The test setup for this work incorporates the custom designed three-phase voltage source inverter and induction motor. Both were designed for electric vehicle traction drive applications. The units were tested on an in-house dynamometer consisting of a Yaskawa Verispeed 626VM3 drive along with a Himmelstein 28003TNS torque sensor. The dynamometer is rated at 22 kW and 6000 RPM, and uses closed loop speed for its control algorithm. Figure 10 shows the motor setup for testing.
To measure all the necessary data for this study, several pieces of laboratory equipment where used. The equipment list as well as a basic overview of the equipment, Figure 11, is given below.
Along with the dynamometer listed above, the equipment used in this testing include the following:
o Tektronix TDS7054 Oscilloscope o Yokogawa 2533EAC Power Meter o Himmelstein 66032 Dynamometer o 2 HP E3631A Power Supplies
o Tektronix A6304XL Current Probe o Tektronix AM503B Probe Amplifier o 2 Fluke 45 Digital Multimeters
o 3 Tektronix P5205 HV Differential Probes o Analog Devices ADMC401 DSP
The instrument applications in the setup are typical. The digital multimeters were used for measuring DC bus voltage and DC bus current through a current shunt. The two differential probes and the current probe were used for measuring voltage and current waveforms for oscilloscope FFT calculations. The remaining equipment use is clear.
Temperature measurements were handled with the use of two J-type thermocouples. The inverter was mounted on a standard aluminum heat sink with three fans placed at one end to provide the cooling. The thermocouple was placed next to the IGBT that was furthest away from the cooling fans. Figure 12 below illustrates the position of both the fan and thermocouple placement. This placement gave a good approximation of the worst-case condition with respect to cooling and maximum temperature.
The motor proved to be much more difficult to cool than the inverter. The GE standard industrial housing is not specifically designed for forced convection cooling. The ideal cooling would be the use of circulated oil in a sealed housing. To help alleviate this problem, a large blower with ducting was placed directly behind the motor. The use of the blower gave much better airflow around the stator as well as through the air-gap of the motor. All in all, this allowed for better cooling, but was still not sufficient enough for heavier load testing.
To track the motor temperature, a thermocouple was placed on the opposing end of the motor to the blower on the stator windings. Similar to the inverter, the end away from the blower was the most probable hot spot. Another benefit to placing the thermocouple in this position is the fact that the stator end turn windings are not touching the motor housing, which can act as a heat sink. Therefore, the most probable hot spot is monitored. Figure 13 below shows the thermocouple placement on the motor.
The inverter was developed in house  with a large amount of integration. It consists of three phases with three half-bridge IGBT modules as the switching devices. Figure 14 shows the basic schematic of the three-phase inverter.
For this work, SKM 300 GB 063D IGBTs were used. Separate laminated bus bars were replaced by integrating them into a PC board design. The PCB consists of heavy 14-ounce copper for the DC bus with four layers. The top layer is the positive DC bus, and the bottom layer the negative DC bus. The middle layers are used for the IGBT’s gate signals. The manufacturing was handled by UPE, Inc. The PCB also consists of an integrated common mode choke. The inductor is a one turn E and I core set clamped through the board, and the Y-capacitors are standard Electronic Concepts stock. The fully assembled inverter is shown in Figure 16.
The ratings for the inverter were designed for up to 400 VDC bus voltage, and 300 A capability. The voltage rating is based on a typical battery pack for an electric traction drive. The current rating also explains the need for the heavy copper bus bars.
Another feature of the inverter is the use of fiber optic isolation. The gate drive boards, shown in Figure 15, use optical transmitters and receivers for the gate signals from the DSP as well as the fault signals from the gate drive circuits. There are fault signals for each drive circuit, but for each set of boards they are combined into a single signal that is then sent back to the control board.
Another unique design feature in the inverter is the use of an Electronic Concepts film capacitor for the DC bus. The capacitor is custom designed, has a slim profile, and takes the place of a larger electrolytic capacitor with a high current rating. Overall, this allows the inverter to be much slimmer without giving up DC capacitor performance. There are also small Electronic Concept capacitors across each half-bridge device, but according to previous work, these are not necessary .
Induction Motor Overview
As mentioned above, the motor used in this work is a high frequency three-phase two-pole squirrel-cage induction motor. This type of motor is very robust, and widely used in industrial applications. The motor was designed by Virginia Power Technologies (VPT) and built by General Electric (GE). The motor, shown in Figure 17, uses a standard GE 180-frame industrial housing.
The motor, which was designed for automotive traction drive applications, needs to have a high power density or power to weight ratio. This is the main difference between this motor and a typical industrial induction motor. Therefore, it was designed as a smaller high-speed 20,000 RPM motor. This allows for reduced weight and size as well as lower rotor inertia. The lower inertia benefits the need for quick acceleration and deceleration. Acceleration is a common automotive requirement, but in this case, with such a high-speed motor and the use of a transmission, the need for the motor to accelerate and decelerate quickly is even more important. The torque production does suffer because of the motor’s smaller size, but with the use of the previously mentioned transmission, the torque is recovered.
For this type of testing, with multiple measurement equipment, a reasonable amount of accuracy is required. In this case, the inverter is run with open loop control of the line frequency and modulation index. Both were controlled via a potentiometer feeding a voltage signal into the DSP A/D converter, and scaled internally to the appropriate value. Changing the switching frequency was accomplished by changing the value directly in the DSP code. While this allowed for a reasonable amount of control over the testing, it also posed the problem that the resolution is limited on all three variables. The switching frequency resolution was limited because of the original coding of the DSP. The code was not intended for use with a variable switching frequency. Therefore, the program would only accept increments of 1000 Hz. This was not as much of a problem with accuracy in that it just limited the number of experimental data points. The potentiometer controlled variables, line frequency and modulation index, where limited in resolution by the onboard A/D.
While the above resolution problems offered some challenges, the larger problem came from the way the DSP code was originally written with respect to the line and switching frequency. As stated before, the code was not intended for use in variable switching frequency applications. Therefore, the problem arises in that for a given line frequency and switching frequency, the program calculates the total number of switchings in one line cycle. This would not be as much a problem except for the fact that the program has limited resolution. Thus, for a given test condition, the actual line frequency can vary slightly with the change in switching frequency, and because line frequency is directly related to slip of the induction motor, the loading could also change slightly. Therefore, the data points taken for each test are not shown to be linear, but there is a visible trend in the data. To show that the data trends make sense, temperature was also recorded for each individual test on the inverter and motor. The testing was all run at a steady state temperature. In recording the temperatures, the expected trends were confirmed. The testing routine is given below.
The first step was to set both the line frequency of the inverter and the speed of the dynamometer. To begin both speeds were set close to 1000, 2000, and 3000 RPM depending on the test. After the speed was achieved, the DC bus voltage is applied and set to approximately 300 V. This voltage approximates the use of a battery pack in an electric vehicle. Next, the speed of the dynamometer is lowered to produce the desired slip. The actual value of slip was not as important as keeping it the same throughout the testing because this work is focusing on general efficiency improvements.
When the desired slip is reached the modulation index is increased or decreased depending on what load is being run. By adjusting the modulation index up or down, the line-to-line voltage on the motor is changed. Ideally the DC bus voltage would not change, but as discussed before the code’s precision is not as high as desired. Therefore, the DC bus voltage would be adjusted slightly to set the load precisely.
Once the conditions above are met, the temperature is monitored. During the thermal transient the modulation index or DC bus voltage will have to be adjusted to account for the thermal losses. As the temperature rises, the conduction losses increase causing the load to drop. Once thermal steady state is reached the measurement data can be recorded. The data recorded would be for a particular switching frequency. Thus, once the data has been recorded the power supply is turned off, and the DSP code is changed to the next switching frequency. Once the code has been loaded into the DSP the power supply can be brought back on-line. From here, the temperature must reach steady state again before any measurements are taken. It is worthwhile to note that because there is a large thermal mass to both the induction motor and inverter, the time to steady state is reduced quite a bit. The above procedure is repeated for all switching frequencies at that particular load. For the next load, the slip is adjusted and the process is repeated.
Once the test plan and test setup were complete, the testing was run at three speeds and at each of those speeds, three loads. These data points where chosen to get an approximate data trend. The raw data collected for all test results are located in Appendix I. The speeds selected for all the testing are: 1000, 2000, and 3000 RPM. Then, for each speed, the following load-torques were run: 60.5, 120.3, and 180 in-lb. Since the main scope of this work is the effect of switching frequency on the motor and inverter, the switching frequency is varied from 5 kHz to 10 kHz in 1000 Hz increments.
1000 RPM Operation
The first tests run were at 1000 RPM with the three different load torques and switching frequencies. Below, Figure 18 shows the overall efficiency of the system for the three separate loads. A quick look at the data points in Figure 18 show the DSP resolution problem. The data jumps around as the switching frequency is changed. Later, the temperature plots will show a much more linear trend. A side note to the problem is that at 7 kHz the before mentioned DSP precision error was so bad that for all tests this data point was excluded.
Examining the data further shows a trend where as the switching frequency is decreased the overall system efficiency increases. To support this trend, the temperature was also recorded at each data point.
As stated in the procedure, the data was recorded when the temperature had reached steady state. Figure 19 and Figure 20 below show the steady state temperatures of both the inverter and motor respectively. The results support the efficiency trend illustrated in Figure 18. Comparing the temperature plot of the inverter to that of the motor shows that the inverter temperature decreases at a faster rate than the motor temperature increases as the switching frequency is reduced.
The power loss plots for both the motor and inverter are shown in Figure 21 and Figure 22 respectively. These also illustrate the trends shown in the efficiency and temperature graphs. As in the case of temperature, the inverter power loss decreases more than the motor losses increase with the reduced switching frequency.
Another interesting result of lowering the switching frequency is the fact that the improvements increase as the load increases. Basically, the slope of a trend line for the inverter temperature and power loss increases as the load is increased while the motor slopes are essentially the same. As a result, the overall system efficiency improves with the reduced switching frequency. This is also the case for all three load torque conditions.
Since the results shown in the previous section show an improvement in overall efficiency, the frequency spectrum is examined for some explanation of the improvements. To accomplish this, a FFT was run on the line-to-line motor voltage as well as the motor phase current using the Tektronix oscilloscope. Since the results show a somewhat linear trend, the FFT was run at 5 kHz and 10 kHz switching frequencies for each load. This should then give the largest differential in harmonic content.
Figure 23 and Figure 24 below show the difference between the line-to-line motor voltage frequency spectrums at the 60.5 in-lb load. Since both plots are on the same scale, the difference is easily discerned. The lines placed on the plots also show the difference. For this case, the 10 kHz operating condition shows a larger magnitude for the switching frequency harmonic as well as multiples of the switching frequency.
Next, the current frequency spectrum needs to be examined. Figure 25 and Figure 26 show the current spectrum at the same load of 60.5 in-lb. Looking closely at the spectrum shows that the sidebands, marked with the oval, of the fundamental have increased with the lower switching frequency. This can be expected because of the increased ripple current introduced into the motor by essentially leaving the IGBTs on and off longer causing larger peaks or ripple. These differences on voltage and current can help explain the improvements shown before because of an improved distortion factor, which is directly related to THD (3.1). The current distortion factor would appear to worsen as the voltage distortion factor improves.
The results for the 120.3 in-lb loads are given below. Figure 27 and Figure 28 show the voltage spectrum and Figure 29 and Figure 30 show the current spectrums. Examining these results also show the same results as the lighter 60.5 in-lb load. The voltage switching frequency harmonics are higher at higher switching frequency while the fundamental frequency side bands of current are worse at lower switching frequency.
As mentioned before, the improvements in efficiency are greater as the load is increased. Therefore, by examining the data closer, there is a more noticeable difference in the harmonic content for both voltage and current for the increased load. This same trend can also be seen in the heaviest load condition. The results for the 180 in-lb. load are given in Figure 31 and Figure 32 for voltage, and Figure 33 and Figure 34 for current. These results again show the same trends as in the two previous load conditions.
The remaining plot of power factor, Figure 35, shows that the power factor increases as the switching frequency decreases. This helps to explain why the improvements occur at lower switching frequencies. For this work, power factor is defined as the displacement power factor, DPF, multiplied by the distortion factors, DF, for voltage and current:
PF = DPF x DFV x DFI (3.1)
Since the harmonic content is directly related to THD, and the harmonic content increases with the lower switching frequency, the distortion factor improves at lower switching frequency.
2000 RPM Operation
The second tests were run at 2000 RPM with the same conditions. Below, Figure 36 shows the overall system efficiency for 2000 RPM. The same trends can be seen as in the previous experiments with the exception that at the lightest load, 60.5 in-lb, the overall system efficiency actually decreases with lowered switching frequency.
If you examine the slopes of the trend lines for the data at 1000 RPM and 2000 RPM, they show that the slopes at the same loads in 2000 RPM are less than those in 1000 RPM. For the 60.5 in-lb load this is obvious because of the decrease in efficiency. The heavier loads also show the difference.
As in previous experiments, the temperature data also shows the same trend. The motor temperature data does not appear to be much different from those at 1000 RPM, but the inverter temperature decreases less than those at 1000 RPM. Figure 37 and Figure 38 show the temperature plots for 2000 RPM operation.
Comparing the motor and inverter temperature graphs again show that the motor temperature increases less than the inverter temperature decreases, but only for the two heavier loads. For the lightest load the inverter temperature is almost constant throughout the frequency range. Therefore, any temperature increase in the motor will give an indication of a decrease of the overall system efficiency.
The power loss plots for both motor and inverter are shown in Figure 39 and Figure 40. Analyzing the power loss plots shows the same outcome as the temperature plots. Once again the inverter power loss is essentially constant at the lightest load, and consequently the system power loss increases at the lightest load.
For the two heavier loads, the results are identical to 1000 RPM operation. The inverter power loss still decreases more than the motor power loss increases. Therefore the overall system losses actually decrease with the switching frequency.
The harmonic content is once again analyzed to show the relationship between the switching frequencies. The same results as in the 1000 RPM test are to be expected with the exception of the lightest load condition. Figure 41 and Figure 42 show the voltage frequency spectrum for the 60.5 in-lb load.
Once again, the higher 10 kHz FFT shows larger switching frequency harmonics. The difference in the magnitude for both 5 kHz and 10 kHz is approximately the same as before. To explain the decreased efficiency at this load the current frequency spectrum should be analyzed.
Figure 43 and Figure 44 below show the current frequency spectrum. Comparing these results to the 1000 RPM case shows that the current harmonics, the fundamental frequency side bands, are slightly worse than those before. This helps explain the difference in the light load efficiency. Basically, the voltage harmonic improvement no longer offsets the increased current harmonics.
For the 120.3 in-lb load, the voltage frequency spectrums are given in Figure 45 and Figure 46. As expected, the harmonic content decreases with the lower switching frequency at 5 kHz. The difference is still very similar to the data from the 1000 RPM test. Once again the current frequency spectrum should be examined to determine any benefits.
The current frequency spectrum scope captures are given in Figure 47 and Figure 48. As with the data from the 1000 RPM testing the fundamental frequency sidebands are worse at 5 kHz. Comparing the content from the 1000 RPM data, the side bands are worse for the 2000 RPM data. Once again the improvement in the voltage harmonics for lower switching frequency offsets the increase in current harmonics at the lower switching frequency.
Table of Contents
Table of Contents
List of Figures
List of Tables
1.2. Previous Work
2. Analytical Model
2.1. Inverter Model Derivation
2.2. Induction Motor Model
2.3. Analytical Results
3. Experimental Results
3.1. Measurements and Testing
3.2. Inverter Overview
3.3. Induction Motor Overview
3.4. Testing Methodology
4. Analytical and Experimental Comparison
5.1. Future Work
GET THE COMPLETE PROJECT