Professional Documents
Culture Documents
Cogging Torque Measurement, Moment of Inertia Determination and Sensitivity Analysis of An Axial Flux Permanent Magnet AC Motor
Cogging Torque Measurement, Moment of Inertia Determination and Sensitivity Analysis of An Axial Flux Permanent Magnet AC Motor
Traineeship report Traineeship performed at the Charles Darwin University, Darwin, Australia
Coach(es):
Supervisor:
dr.ir. M. Steinbuch
Technische Universiteit Eindhoven Department Mechanical Engineering Control Systems Technology Group
ii
Abstract
Many technical applications require a smooth torque. An axial ux permanent magnet (PM) AC motor is used to achieve this with control methods. Required are motor parameters, such as the moment of inertia. This parameter is determined by calculation with help of the CAD drawings. To verify the result, an experimental setup is designed. The resulting dierence of 6.6% between the calculated and experimental determinded value of the moment of inertia is explained with the help of a sensitivity analysis. One of the properties of the type of motor used to achieve smooth torque is the presence of cogging torque. To compensate for cogging torque, this parameter needs to be measured. To be able to do this, a measurement method is designed. To explain the resulting RM Serror a sensitivity analysis of the calibration method is made. This is done theoretically and veried experimentally. The remaining RM Serror of 0.8 1.5% is caused by the current sensor and control errors.
iii
iv
Samenvatting
Voor vele technische toepassingen is een constant koppel vereist. Door enkele regel methodes toe te passen op een axiale ux permanente magneet (PM) AC motor wordt dit bereikt. Hiervoor is het nodig om de motor parameters te weten, zoals het massatraagheidsmoment. Met behulp van de CAD tekeningen wordt deze parameter berekend. Met een experiment wordt de berekende waarde geverieerd. Het verschil van 6.6% tussen deze twee waarden wordt verklaard aan de hand van een foutenanalyse. Een van de eigenschappen van het type motor dat gebruikt wordt is cogging. Door deze te meten kan hiervoor gecompenseerd worden. Voor deze meting is een opstelling bedacht. Om de resulterende RM Sf out te verklaren wordt een foutenanalyse van de kalibratie methode gemaakt. Allereerst gebeurt dit theoretisch. Hierna zijn de antwoorden experimenteel geverieerd. De overgebleven RM Sf out van 0.8 1.5% wordt veroorzaakt door de stroomsensor en regelfouten.
vi
Contents
Abstract Samenvatting Table of contents Nomenclature 1 Introduction 1.1 1.2 Motor Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Report overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii v vii ix 1 2 2 5 5 7 7 8 10 13 13 18 19 19 20 21
2 Determination of the Moment of Inertia 2.1 Determination of the moment from the CAD drawings . . . . . . . . . . .
2.2 Experimental determination of the Moment of Inertia . . . . . . . . . . . 2.2.1 2.2.2 2.2.3 2.3 Method of determining the Moment of Inertia experimentally . . . Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . Results of the experiment . . . . . . . . . . . . . . . . . . . . . . .
Comparison and sensitivity analysis of the results . . . . . . . . . . . . . . 2.3.1 2.3.2 Sensitivity analysis of the experiment . . . . . . . . . . . . . . . . Improvements of the experiment . . . . . . . . . . . . . . . . . . .
Measurement of the Cogging Torque 3.1 3.2 3.3 Dening problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Measurement method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Contents
4 Sensitivity Analysis of the Calibration 4.1 Calibration Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1.1 4.1.2 4.1.3 Determination of y and z over operating range . . . . . . . . . . .
Determination of , and cog . . . . . . . . . . . . . . . . . .
25 25 28 28 28 29 30 34 39
4.2 Theoretical Sensitivity Analysis . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Measurements for verication of the calibration method . . . . . . . . . .
Bibliography
39
viii
Nomenclature
Symbols from chapter "Determination of the Moment of Inertia" F J K L M m T t x x1 x5 = Logarithmic increment = Rotation of the electric motor (rad) = Damping ratio = Force (N m) = Moment of inertia (kg m2 ) = Spring stiness (N m) = Distance to the rotation point (m) = Mass attached to a sping (kg) = Mass of a spring (kg) = Driving torque of the electric motor (N m) = (Period) time (s) = Elongation of the spring (m) = Maximum of the rst oscillation = Maximum of the fth oscillation
Symbols from chapter "Measurement of the Cogging Torque" 2p cog = Number of motor poles = Cogging torque (N m) ix
Nomenclature
HCF = Highest Common Factor Np Q = Number of periods of the cogging torque in a slot pitch rotation = Number of stator slots
Symbols from chapter "Sensitivity Analysis of the Calibration" m = 1 vector of the mean torque = torque sensor scaling error = current inverter scaling error rated = Rated torque of the electric motor (N m) N = Number of measurements points
RM Serror = Root Mean Square error of the motor torque p p I I i,p K k,p = 3 1 vector of current scaling error estimate = current scaling error in phase p = 3 1 vector of current scaling error = 3 1 vector of current oset error estimate = current oset error in phase p = 3 1 vector of current oset error = element-wise multiplication operator = 3 matrix of the current estimate (A) = 3 matrix of the current (A) (NOT the identity matrix) = current in phase p at encoder point (A) = 3 matrix of the back EMF (V s/rad) = normalised back EMF for phase p at encoder point (V s/rad) x
Nomenclature
cog cog
= 1 vector of the cogging torque estimate (N m) = cogging torque at encoder point (N m) = 1 vector of the cogging torque (N m) = 1 vector of the electro-magnetic torque (N m) = 1 vector of the estimate of motor torque (N m) = estimate of motor torque at encoder point (N m) = 1 vector of the pulsating torque estimate (N m) = 1 vector of the estimated mean torque (N m) = 6 matrix of torque and back EMF = 9 1 vector of scaling and oset errors = 1 vector of residuals
cog em
m m, p r
X y z
xi
Nomenclature
xii
Chapter 1
Introduction
Many electric motor applications require a constant torque, especially applications that require precise tracking. These processes are for instance laser cutting and numerically controlled machining. Pulsating torque (any kind of variation in the torque output of the motor) can have a negative eect on, for example, the surface nish when using rotary machine tools. Also pulsating torque can excite resonances in the drive-train of the application. This produces acoustic noise as well. A smooth torque output can be achieved by using a programmed reference current waveform. This method has the ability to work at dierent speed and torque set points. This minimizes restrictions on motor design and manufacture. When limiting pulsating torque mechanically, accurate manufacturing is required. This limits the practicality for low-cost, high volume production. Research on this subject is performed at the Charles Darwin University (CDU) in Darwin, Australia. The goal of the CDU electric motor research program is to create an output torque with a maximum RM Serror of 1%. The contribution to this research explained in this report consists of three parts: For control purposes, the moment of inertia of the motor has to be known. With the help of CAD drawings of the electric motor, the moment of inertia is calculated. To verify this result, the moment of inertia is determined experimentally. One of the properties of the electric motor used, is the presence of cogging torque. To achieve a smooth output torque, compensation for the cogging torque is added to the control scheme. Therefore the cogging torque is measured. To improve the result of a programmed reference current waveform, a calibration 1
Chapter 1. Introduction
method is designed. The sensitivity of this calibration method is analyzed theoretically and veried practically. In this analysis also the cause for the remaining torque ripple is explained.
1.1
Motor Setup
The type of motor used for this research is a Permanent Magnet Synchronous AC motor. The rst natural frequency of the motor mounted on a force table is 700 Hz. The stator consists of 48 slots and the rotor has 16 poles. The motor has a rated torque of 6N m and a rated voltage of 24V . The position of the axle is measured with a 12-bit (4096 states), gray code, absolute encoder. The torque is measured with piezo electric reaction torque sensors. An eddy current brake is installed to apply dierent torque set points. The dened set points are 1,2,3,4 and 5 N m and 0.5, 0.6, 0.7, 0.8, 0.9 and 1.0 Hz. This provides a mesh which consists of 30 measurements. Data acquisition is done with the help of Labview. In gure 1.1 a picture of the electric motor can be seen. In this picture the magnets of the eddy current brake have been removed.
1.2
Report overview
The contribution to the research consist of three parts, which are divided into three chapters. The determination of the moment of inertia is described in chapter 2. In chapter 3 the cogging torque is measured. The sensitivity analysis of the calibration method is explained in chapter 4. The conclusions and recommendations of the three parts are combined in chapter 5.
Figure 1.1: The electric motor with the dierent components named. The magnets of the Eddy Current brake have been removed.
Chapter 1. Introduction
Chapter 2
2.1
A solid model of the motor is available in Pro/ENGINEER. The moment of inertia of the assembly of the rotating parts (gure 2.1) can be calculated in Pro/ENGINEER. The rotation frequency of the bearing cage assembly is 42% of the inner race and the attached parts [12]. The acceleration is proportional to the moment of inertia. Only 42% of the moment of inertia of the cage assembly contributes to the total moment of inertia. The calculated moment of inertia is 0.01056kg m2 . 5
Figure 2.1: CAD assembly of the rotating components of the electric motor.
2.2
2.2.1
The moment of inertia can be determined experimentally by an acceleration or an oscillation method. Both methods are discussed for this particular case. This discussion is based on work done by Genta en Delprete [4]. Acceleration Method The rotating parts of the electric motor are constrained. Rotating motion is only possible about one axis. Somehow, the body is subject to a driving torque T . During the test the time t and the accompanying rotation are measured. The moment of inertia can be calculated with equation (2.1). The presence of damping is neglected. J= T t2 2 (2.1)
The acceleration method has a non-periodic motion. To decrease measurement errors in the time and position, long tests need to be performed. This however increases the error caused by neglecting damping. Due to the presence of the bearings and the eddy current brake, the inuence of damping on the experiment is rather large. Oscillation Method Rotating motion is only possible about one axis. To create an oscillating motion, an elastic spring with stiness K is attached. Measuring the period time T of an oscillation, in combination with the damping ratio, the moment of inertia can be calculated: J= KT 2 1 2 4 2 (2.2)
The oscillation method is periodic. Measuring a number of oscillations reduces measurement errors. Comparison Acceleration and Oscillation Method Because of the following reasons there is chosen to use the oscillation method to determine the moment of inertia: By measuring a large number of oscillations, the relative error for time and position measurements is reduced. 7
With the oscillation method, the measurement can be started after a slowly decaying motion has been reached. Errors due to initial transients are not present in the measurements. From a number of oscillations the damping ratio can be calculated. This is possible because the position can be measured with the encoder.
2.2.2
Experimental Setup
To get an oscillating motion around the rotation point of the electric motor a lever (an aluminum bar) is attached to the rotor. This is done with two, already available, screws. On the end of the aluminum bar the springs are xed. The other end of the springs is attached to the solid world. The springs are prestressed. Only the linear part of the springs is used. To determine the position of the rotating parts of the electric motor during the oscillation, the encoder is used. During the oscillation experiment the stator of the electric motor is removed. Otherwise oscillating motion is not possible due to the cogging torque. The magnets of the eddy current break are positioned in a way in which they minimize the force exerted on the disc attached to the axle. Removing the eddy current brake is dicult.
Equation (2.2) is valid for one spring attached to the bar. To get an oscillating motion two springs, in opposite direction, are attached to the bar. Also the springs are placed at a distance L of the rotation point. Adapting equation (2.2) gives: J= 2KL2 2 T 1 2 4 2 (2.3)
The length of de bar (L) and the stiness of the springs (K) has to be chosen. The chosen length of the bar is 0.5m. An estimation of the J can be made based on the CAD answer (chapter 2.1) with added the J of the bar. The last one can be calculated with standard formulas. When the damping is neglected, the period time can be plot as a function of the spring stiness, gure 2.2. A spring constant of about 100N/m is chosen. The inuence of a small error in the spring constant on the oscillation time is small. Nevertheless the oscillation time is reasonable. Due to limitations of the available springs, on each side of the bar three springs with a total constant of about 100N/m are used. The constant of every spring is determined experimentally by using F = Kx. Deriving the equation for 8
Oscillation time for varying spring constants, L = 0.5m 0.5 0.45 0.4 Oscillation time t [s] 0.35 0.3 0.25 0.2 0.15 0.1
20
40
140
160
180
Figure 2.2: Oscillation time for various spring constants with L = 0.5m.
the case with three springs on both sides gives: (K1 + K4 ) L2 + (K2 + K5 ) L2 + (K3 + K6 ) L2 2 1 2 3 T 1 2 4 2
J=
(2.4)
A schematic drawing of the setup with the variables can be seen in gure 2.3.
2.2.3
The result of 30 oscillation measurements (and the steady state) is shown in gure 2.4. As can be seen, the zero crossings for each measurement is nearly identical, only the amplitude varies. The period time is determined by averaging the time needed for 5 oscillations. 10
2350
2300
2250
2200
0.1
0.2
0.3
0.5
0.6
0.7
Figure 2.4: The oscillation of 30 measurements and the steady state The logarithmic increment [13] is used to determine the factor : 1 x1 ln 4 x5 = 2 + 2 4 = (2.5) (2.6)
The maximum of the rst and fth oscillation are used to minimize errors. In gure 2.5 a measurement for the last oscillation is shown. There are sucient measurement points for an accurate determination of the maximum. The moment of inertia determined from the measurements includes the moment of inertia of the motor, bar and the springs. The moment of inertia of the springs can be neglected. This is not possible for the mass of the accelerated part of the springs. According to Thomson [13] the mass which should be included (also called "eective mass") is
1 3
to be point masses. Due to the large distance to the rotation point, the inuence is significant. Final answer for the experimentally determined moment of inertia is 0.009902kg m2 . The standard deviation of the 30 moments of inertia determined, is 0.0038%. 11
2320
2310
2300
2290
2280
0.6
0.62
0.64
0.68
0.7
0.72
Figure 2.5: Measurement points of last oscillation of a measurement. Also the steady state is shown.
12
2.3
The dierence between the moment of inertia calculated from the CAD drawings and the moment of inertia experimentally determined is 6.6%. Assuming that the calculation performed in Pro/ENGINEER is perfect, the dierence is caused by errors in the experimental method. Therefore a sensitivity analysis of the experimental method is made.
2.3.1
Errors can be caused by the lengths (L1 ,L2 and L3 ), the determined spring constants (K1 - K6 ) and a dierent weight of the springs. The eective mass of the springs as taken into account is incorrect. The percentage errors, as showed in the gures, are related to a realistic error in the determined variable. On the y-axis the percentage error in the moment of inertia is shown. An error of 6.6% explains the dierence between the calculation and the experiment.
13
Deviation in the spring constants In gure 2.6 the resulting error in the moment of inertia of an error in the spring constant can be seen. An error of 3.0% is equal to a change in the spring constant of 1N m. The six lines correspond to an error in one or more (up to 6) spring constants. An error in all springs can result in an error in the moment of inertia of 6.6%, however this is not very likely.
Moment of inertia error caused by an error in the spring constants 8 6 Moment of inertia error [%] 4 2 0 2 4 6 8 3 K1 K1,2 K1,2,3 K1,2,3,4 K1,2,3,4,5 K1,2,3,4,5,6 2 1 0 1 2 Percentage error in the spring constants [%] 3
Figure 2.6: Moment of inertia error for a percentage error in the measured spring constants.
14
Deviation in the measured lengths The lengths (L1 ,L2 and L3 ) in gure 2.3 are measured. A percentage error of 0.5% is a measurement error of 2mm. The resulting change in the moment of inertia of this error can be seen in 2.7. An error in all three distances (red line) explains max 2% of the total dierence between calculation and experiment. This particular case, an error in all three distances, can be caused by an incorrect determination of the rotation point
Moment of inertia error caused by an error in the lengths 2.5 2 1.5 Moment of inertia error [%] 1 0.5 0 0.5 1 1.5 2 2.5 0.5 0 Percentage error in the lengths [%] L1 L1,2 L1,2,3 0.5
Figure 2.7: Moment of inertia error for a percentage error in the measured lengths.
15
Mass of the springs The accuracy of the balance used to weigh the springs is 1g. The mass of the springs is determined by weighing all of the springs and dividing by six. An error of 1g in the total mass, is about 0.2g in the mass of one spring. This is 3%. The result of this error on the error in the moment of inertia can be seen in gure 2.8.
Moment of inertia error caused by an error in the weighted spring mass 1 0.8 0.6 Moment of inertia error [%] 0.4 0.2 0 0.2 0.4 0.6 0.8 1 3 2 1 0 1 Percentage error in the spring mass [%] 2 3
Figure 2.8: Moment of inertia error for a percentage error in the weighted spring mass.
16
1 3
Fox and Mahanty [3] claim that this is wrong. The eective mass depends on the mass attached to the spring. Also Armstrong [1] and Sears [11] claim this. Speaking of a mass M attached to a spring with mass m. When proaches
m 3. M m
On the other hand, when there is no mass attached to the spring (M = 0) the
4m . 2
eective mass is
shown in gure 2.9. Calculation of the exact eective mass in this case is rather dicult. This is caused by the nonlinearity of the equation and the diculty of determining the attached mass M .
0.34
0.35
0.39
0.4
Figure 2.9: Moment of inertia error for dierent factors for the eective mass of the springs. 17
2.3.2
A combination of the uncertainties in the lengths, spring constants, spring mass and effective mass factor can cause the moment of inertia error of 6.6%. Therefore the accuracy of the experiment should be improved to get a more precise answer. The largest uncertainties are caused by the spring constants and the measured lengths. By taking only one spring on each side, instead of three springs, the possible error is lowered. Only 1 length has to be determined instead of 3. This also reduces the importance of the eective mass factor and the weight of the springs. The length of the bar, currently 0.5m, should be reduced. The error caused by a measurement mistake is reduced quadratically. By reducing the length of the bar, the measured moment of inertia is dominated by the electric motor and not by the bar used. Finally the Eddy Current brake should be removed. This reduces the eect of damping on the experiment.
18
Chapter 3
3.1
Dening problem
It is dicult to measure cogging torque. If measured incorrectly dynamic eects of speed variations of the stator are present. Determination of a correct measurement method is dicult. Many publications ([7],[14],[9] and [10] ) present measurement results. However the explanation about how the measurement is performed is very summary.
Cogging torque measurements can be performed by static, quasi static and dynamic methods. More explanation about these methods can be found in Heins [5]. All measurement equipment of the electric motor is optimized for measuring during rotation. For this reason a dynamic cogging torque measurement method is chosen. 19
3.2
Measurement method
To measure the cogging torque the rotor should turn without any input current. To accomplish this, the rotor is turned with the help of an external motor. On the rotor a pulley is mounted, which is connected to the pulley of the external motor with a exible rubber belt. The force which is caused by the rotation of the rotor is measured by the reaction torque sensor which is mounted underneath the housing of the motor. The position is measured by the encoder. The measurement setup can be seen in gure 3.1. At the dened speeds (see page 2) data from 24 revolutions is saved. The number of revolutions is bounded by the maximum amount of data which can be sampled in one trial. Due to the reaction torque sensor compensation for inertial forces is not necessary [5].
20
3.3. Results
3.3
Results
The measurement is started at an arbitrary point of the rotation. To get complete revolutions starting at the "zero" point, the 24 revolutions are reduced to 23. By averaging the noise is reduced. The Discrete Fourier transform (DFT) of the averaged revolution is showed in gure 3.2. All speed set points are shown.
10 Amplitude
10
10
10
50
0 50 Harmonics
100
150
200
250
Figure 3.2: DFT of the averaged revolution for all speeds The number of cog periods in a slot pitch rotation (Np ) can be calculated [2]: Np = 2p HCF {Q, 2p} (3.1)
In this equation, 2p and Q are the number of motor poles and the number of stator slots respectively. The Highest Common Factor (HCF) of Q and 2p is the denominator. Calculating the Np with the motor data from page 2 gives 1. The stator consists of 48 slots. The main harmonic of the cogging torque is therefore the 48th. This corresponds with the DFT shown. Other harmonics are also present. The signicant harmonics are multiples of 16. The cause of these harmonics is probably a dislocated slot. In combination with the 16 permanent magnets on the rotor this dislocated slot gives a 16th harmonic. The only exception to this is the 0th harmonic. This harmonic is caused by the torque sensor. 21
By extracting the multiples of 16 and converting the signal from the frequency domain back to the time domain, the cogging torque is constructed. Averaging over the speed range gives gure 3.3.
0.4
0.2
0.2
0.4
0.6
500
1000
3000
3500
4000
Figure 3.3: Measured cogging torque averaged over the speed range
22
3.3. Results
The percentage error of the averaged cogging torque can be seen in gure 3.4. The averaged cogging torque is compared to the cogging torque of every speed. As can be seen, the amplitude of the error is speed dependent. This is caused by the exibility of the rubber belt used to drive the rotor of the motor. This causes the errors for the lower speeds. In the higher speeds slip did occur. The cogging torque measurements could be improved by using a stier transmission between the two pullies.
23
24
Chapter 4
4.1
Calibration Method
The calibration method is designed by Heins [5]. A global approach is shown here.
If the "ideal" currents have been correctly calculated, pulsating torque will only come from an error in either: 1. the cogging torque; and/or 2. an unbalance in the current, caused by an oset or gain error in the current sensors. These are shown in pink in gure 4.1, and their determination will be the focus of this section. By decoupling the pulsating torque into the components created from each of these errors, it is possible to determine where the errors lie and compensate accordingly. To do 25
this, it is important to note that the cogging torque will be independent of current input. The cogging torque is redened as the residual resulting from a least squares minimisation matching the electro-magnetic torque to the measured torque.
The torque of the electric motor with the current oset and scaling factors is:
m, = p=a,b,c cog (i,p p + p ) k,p +
(4.1)
where:
m, = estimate of motor torque at encoder point (N m)
= torque sensor scaling error i,p = current in phase p at encoder point (A) p = current scaling error in phase p = current inverter scaling error p = current oset error in phase p k,p = normalised back EMF for phase p at encoder point (V s/rad)
cog = cogging torque at encoder point (N m)
(4.2)
where:
m = 1 vector of the estimate of motor torque (N m)
= 3 1 vector of current scaling error = 3 1 vector of current oset error K = 3 matrix of the back EMF (V s/rad) cog = 1 vector of the cogging torque (N m) A dynamic torque sensor is used which does not measure the average component of the torque. The estimate of the motor torque is therefore:
m = p + r
(4.3)
where:
p = 1 vector of the pulsating torque estimate (N m) r = 1 vector of the estimated mean torque (N m) r is eectively the torque created without , and cog : r = (I K)
(4.4)
(4.5)
If we concatenate the matrices to let: X = IK K and concatenate the vectors to let: y= and let:
z = cog = cog
(4.6)
( 1)
(4.7)
(4.8)
then:
m = Xy + z
(4.9)
where y and z are unknown. Using the Moore-Penrose pseudo inverse is a convenient way conducting a least squares minimisation. By using this inverse and assuming that z will be the residual, y can be found. The residual (z) can then be found by rearranging equation 4.9:
z = m Xy
(4.10)
27
4.1.1
y is attributed to errors in the current sensor system, so regardless of speed and torque set-point it should be constant. z is attributed to cogging torque so should also be independent of operating point. This method will only be valid if y and z are the same over the entire operating range. One method to ensure that the same y is determined for all operating points is to combine all tests at dierent operating points into one long X (6 number of trials). Though this gives only one y for all trials, it does give a dierent z for every trial.
4.1.2
Consideration of y and z and Equation 4.8 suggests the best estimates for the parameters responsible for pulsating torque are: = = y 1,2,3 +1 (4.11) (4.12) (4.13) (4.14)
=
cog = em
Overall system gains These expressions suggest that while an estimate of and cog is possible, an estimate of requires a knowledge of the product of and . A method for nding this product and compensating to ensure that the estimate of the overall system gain is correct is available. Assuming this is possible, analysis will continue with: =1 If = 1 then = , and em however are still eected by the unknown . (4.15)
4.1.3
Once , and cog have been determined from an uncompensated set of measure-
ments over the operating range, they can be used to pre-compensate I to cancel their aect for future operation. This is done as shown in gure 4.2. 28
4.2
The values for , and cog can be found from a set of measurements. Some variation exists in these measurements, therefore the determined vectors are not perfect. The goal of this research is minimizing the torque ripple(page 2). Therefore only the RM Serror (equation (4.16)) of the motor torque caused by an error in one, or a combination, of these three vectors is considered. The performance of the motor is expressed in a root-mean-square (RM S) value of the motor torque according to: m m 2 RM Serror = 100% N rated (4.16)
If there are no scaling errors ( = 1), no oset errors ( = 0) and cog is known perfectly, the RM Serror (equation (4.16) ) is zero. Following the denition of and one will nd: a b c = 1 a + b + c = 0 (4.17) (4.18)
From formula (4.17) follows the value of b and c after inducing an error in a , if dened that b and c are equal . In a similar way the values of b and c are calculated after inducing an error in a . To induce an error in cog , the measured cogging torque (chapter 3) is multiplied with a factor to change the magnitude. The intervals in which the values of , and cog are varied: The resulting RM Serror due to variation of the variables in their interval for torque set point 3 is shown in gure 4.3. The rst row of three gures is the change of the 29
Table 4.1: Intervals of the variables RM Serror which is caused by a change in a and cog for three values (the two extremes and the mean) of the interval of a . The two other possible combinations are shown in the second and third row. As can be seen in the rst row, a change of a with 20% positive, results in a dierent RM Serror than a change of 20% negative. This is caused by the nonlinearity in the determination of b (equation(4.17)) after a has been dened. The change of a and cog is symmetric. The inuence of a 35% change in cog is much smaller than a 20% change in a . The chosen intervals give an error which is larger than the maximum allowed RM Serror of 1%. The maximum variation of the variables a , a and cog to stay below a maximum RM Serror of 1% is shown in gure 4.4. After the calibration all the variables should have their ideal value, the RM Serror is equal to zero. Due to uctuations in temperature the current oset and scaling have small variations.
4.3
If the proposed method of compensating for scaling and oset errors and cogging torque works, after calibration, the RM Serror should be minimized. Inducing an error in the scaling, this induced value should be found after a new calibration. This should also be valid for an induced oset error. The results for these two experiments are shown in table 4.2. The error is the dierence between the calculated,and induced value, from a new set of measurements. Also the resulting RM Serror is calculated. As can be seen, the error between the induced and calculated is small. Besides the error in phase c, this is also the case for the oset compensations. The large error in phase c can be caused by an error in the rst calibration. Though the error is relatively large, the resulting RM Serror is still within the allowed interval. 30
Figure 4.3: Resulting RM Serror due to the variation of the variables for torque set point 3 N m.
Table 4.2: Induced scaling and oset errors and the found compensating values
31
Figure 4.4: Limiting interval for a , a and cog with maximum 1% RM Serror .
32
A similar approach for cogging torque is not possible. The induced change in magnitude is not found after a new calibration. This is caused by small changes in phase. Instead the cogging torque found by calibrating is compared with the cogging torque which has been measured (chapter 3). The error in the time domain is showed in gure 4.5. The maximum error is 8%, which is still within the allowed interval (gure 4.4).
500
1000
3000
3500
4000
Figure 4.5: Percentage error between the cogging torque measured and the cogging torque determined by calibration. The error shown is in the time signal.
Besides the comparison of the time signals, the DFTs of the measured and calibrated cogging torque are compared. Only the harmonics which are multiples of 16 are available in the signals. The error in the magnitude of the signals is approximately 0.5%. The phase causes the largest part of the error, but never excedes 0.5 of an encoder point. 33
4.4
To check the theoretical sensitivity, errors are induced with values which are in the interval which has also been used for the theoretical analysis (table 4.1). These errors are induced after the motor has been calibrated. This has been done for torque set point 3 with speed 0.7 Hz. These set points are chosen because they are in the middle of the operating range. The resulting practical RM Serror caused by an induced error in a is showed in gure 4.6. In the same gure the theoretical error is shown. To compare the measurements and the theory, two lines are tted by using the measurements in a way as shown in the legend. The 5th measurement point is not used because, as can be seen from the lines, it is not part of the two slopes. In a similar way the theoretical models for , cog and are veried. The deection of the middle measurement points from the slopes is caused by an error in the current sensors. As per denition, should compensate for current oset values. The dierence in height between the theoretical an practical slopes is caused by the part from the signal from which no information can be extracted, the so called "residual". This signal consists mainly of measurement noise. Some averaging errors remain in this signal as well. The slopes of the theoretical and practical induced error are similar and therefore there can be concluded that the theoretical model is correct. The ideal a is higher than the actual a calculated from the calibration. This is also the case for cog . The calibration method nds the best value for all set points and speeds. For the used set point the calibrated values are not the best.
34
RMS
4.5 1 4 3.5 2 3 RMSerror 2.5 2 1.5 1 0.5 0 0.8 0.85 0.9 3
error
due to change in
7 4 5 6
0.95
1 a
1.05
1.1
1.15
1.2
11 10
35
2 RMSerror
0.5
0.7
0.8
0.9
1 factor of
1.1
1.2
1.3
cog
Figure 4.8: The rmserror caused by an error in cog practical and theoretical
36
Performing a theoretical analysis on the sensitivity of is complicated due to the interactions with other variables. The results of the practical inuence is showed in gure 4.9. To nd the cause of the remaining measured RM Serror , as shown in gure 4.10,
0.7
0.8
0.9
1.1
1.2
1.3
1.4
Figure 4.9: Practical sensitivity due to an error in the reconstructed torque is calculated. This is done by multiplying the feedback current with the back EMF. The reconstructed torque consist of the error in the current sensors and the error in the control loop. The result can be seen in gure 4.11. As can be seen the remaining measured RM Serror of 0.81.5% is in the same range as the reconstructed torque. To reduce the RM Serror more accurate current sensors are neccessary. The current sensors used have an error of 0.7% [8] in the current. Also the control loop needs to be more accurate.
37
RMSerror of m
Speed [Hz]
1.3 1.2 RMSerror of Ifb 1.1 1 0.9 0.8 5 4 3 2 Torque [Nm] 1 0.5 0.6 0.7 0.8 0.9 1
Speed [Hz]
Figure 4.11: Error due to the current sensor and control loop
38
Chapter 5
Conclusion
The report consists of three parts, moment of inertia determination, cogging torque measurement and a sensitivity analysis of the calibration method of the electric motor.
The moment of inertia is calculated in Pro/ENGINEER with help of the CAD drawings. To verify this calculation the moment of inertia is determined experimentally. The experimentally determined moment of inertia is 6.6% lower than the value calculated. The calculated value is assumed to be correct. To nd a cause for the lower experimentally determined value a sensitivity analysis from the experiment is made. There are determined 4 parameters of the experimental setup which can cause conjointly the deection in the determined moment of inertia. Based on these results, recommendations for an improved experimental setup are made.
The measurement devices of the electric motor are optimized for measuring during rotation. Therefore the cogging torque is measured dynamically. The measurement result is satisfying; nevertheless some improvements to the experimental setup are suggested. This should result in a more accurate answer.
To improve the performance, a calibration method for the electric motor is designed. To verify the calibration method and to explain the remaining RM Serror of 0.8 1.5% a sensitivity analysis is made. This is done theoretically and veried experimentally. The resulting dierences are explained. The remaining RM Serror is a combination of the error in the current sensors and a control error.
39
Chapter 5. Conclusion
40
Bibliography
[1] H.L. Armstrong. The oscillating spring and weight - an experiment often misinterpreted. American Journal of Physics, 37(4):447449, 1969. [2] Nicola Bianchi and Silverio Bolognani. Design techniques for reducing the cogging torque in surface-mounted pm motors. IEEE Transactions on Industry Applications, 38(5):12591265, 2002. [3] J.G. Fox and J. Mahanty. The eective mass of an oscillating spring. American Journal of Physics, 38(1):98100, 1970. [4] G. Genta and C. Delprete. Some considerations on the experimental determination of moments of inertia. Meccanica, 29:125141, 1994. [5] Greg Heins. Control Methods for Smooth Operation of Permanent Magnet Synchronous AC Motors. Ph.d. thesis, School of Engineering and Logistics, Charles Darwin University, Australia, 2007. [6] Thomas M. Jahns and Wen L. Soong. Pulsating torque minimization techniques for permanent magnet ac motor drives-a review. Industrial Electronics, IEEE Transactions on, 43(2):321330, 1996. [7] Jiabin Wang Kais Atallah and David Howe. Torque-ripple minimization in modular permanent-magnet brushless machines. IEEE Transactions on Industry Applications, 39(6):16891695, 2003. [8] LEM. Datasheet lem current transducer lts 25-np. http://www.lem.com/docs/ products/LTS%2025-NP%20E.pdf, Accessed on 01-12-2007. [9] Touzhu Li and G. Slemon. Reduction of cogging torque in permanent magnet motors. IEEE Transactions on magnetics, 24(6):29012903, 1988. 41
Bibliography
[10] S. Mir M.S. Islam and T. Sebastian. Issues in reducing the cogging torque of massproduces permanent-magnet brushless dc motor. IEEE Transactions on Industry Applications, 40(3):813820, 2004. [11] Francis W. Sears. A demonstration of the spring-mass correction. American Journal of Physics, 37(6):645648, 1969. [12] SKF. Bearing frequencies. http://www.skf.com/skf/productcatalogue/
calculationsFilter?lang=en&newlink=&prodid=&action=Calc6, Accessed on 01-12-2007. [13] William T. Thomson and Marie Dillon Dahleh. Theory of Vibrations with Applications. Prentice Hall, Upper Saddle River, New Jersey, fth edition, 1998. [14] S. Ruangsinchaiwanich Z.Q. Zhu and D. Howe. Synthesis of cogging torque waveform from analysis of a single stator slot. 2005 IEEE International Conference on Electric Machines and Drives, pages 125130.
42