Beruflich Dokumente
Kultur Dokumente
1.1 Introduction
In contrast to grid connected ac motor drives, hardly variable in speed, power
electronic devices (e.g. inverter), providing voltage supply variable in both
frequency and magnitude, are used to operate ac motors at frequencies other than the
supply frequency. Developments in this direction have taken place long ago, but a
techno-economical solution could not be found until the late 1980s because of
stringent space requirements, non-availability of high power devices and prohibitive
cost of electronic devices and components.
Rapid developments in the field of power electronics (inverter grade thyristor, GTO
thyristor, IGBT etc.) and miniaturization/mass production of control electronics
(development of VLSI technology and microprocessor based digital control
systems) have reached such a stage that variable ac inverter drives are becoming
increasingly popular in todays motor drives. Presently, inverter drives meet not
only weight and space constraints, but also are economically viable.
In general, two basic types of inverters exist: Voltage-source inverter (VSI),
employing a dc link capacitor and providing a switched voltage waveform, and
current-source inverter (CSI), employing a dc link inductance and providing a
switched current waveform at the motor terminals. CS-inverters are robust in
operation and reliable due to the insensitivity to short circuits and noisy
environment. VS-inverters are more common compared to CS-inverter since the use
of Pulse Width Modulation (PWM) allows efficient and smooth operation, free from
torque pulsations and cogging [Bose 97]. Furthermore, the frequency range of VSI is
higher and they are usually more inexpensive when compared to CSI drives of the
same rating [Dub 89].
In this chapter, only voltage-source inverters are considered. Although the power
flow through the device is reversible, it is called an inverter because the predominant
power flow is from the dc bus to the three-phase ac motor load. Bi-directional power
flow is an important feature for motor drives as it allows regenerative breaking, i.e.
the kinetic energy of the motor and its load is recovered and returned to the grid
when the motor slows down. In electric vehicle application, the dc bus energy is
supplied directly from primary energy sources, e.g. batteries.
Chapter 1
Rectifier
DC bus
Inverter
T1
T3
D1
power
supply
T5
D3
D5
AC
motor
Udc
T4
T6
T2
D4
D6
D2
Switching logic
Figure 1.1: Basic three-phase voltage-source converter circuit.
Neglecting the voltage drop of the inductances (current depending) and diodes
(Ud 1V if i > 0), the positive potential of the dc bus voltage equals the highest
potential of the three phases and the negative potential equals the lowest potential of
the three phases. Since each phase owns one negative and one positive maximum
potential during one period of the net frequency, the rectifier input voltage equals
the maximum of the positive and negative line voltages, respectively. Thus, the
rectifier input voltage traces six pulses as shown in figure 1.2 by the thick line.
500
400
uab -uca
0
ubc
-uab uca
10
t [ms]
15
-ubc
10
t [ms]
uab -uca
0
40
0.5
Udc
500
400
20
iB6 [A]
iB6 [A]
600
Udc
U dc [V]
U dc [V]
600
15
-ubc
-uab
uca
10
15
20
10
15
20
10
15
20
t [ms]
20
20
ubc
40
t [ms]
ia [A]
ia [A]
20
0
-1
10
t [ms]
15
20
0
-20
-40
t [ms]
Figure 1.2: Line voltages (uab, ubc, uca), dc bus voltage Udc, line current
of the first phase ia and output current iB6 of a B6-diode bridge.
Left: No inverter output power (inverter losses 10 W).
Right: Inverter output power Pout 5,5 kW.
Figure 1.2 presents typical voltage and current waveforms of a B6-diode bridge
supplied by a stiff grid. As indicated by the dashed lines, the rectifier current iB6
increases, if the absolute value of a line voltage is higher than the dc voltage.
Consequently, the dc voltage increases slightly. A dc voltage higher than the current
voltage supply causes a reduction of the rectifier input current until the current
Chapter 1
equals zero and the diode bridge blocks the supply voltage. The rectifier current iB6
is identically reflected by the line currents. The sign of each line current depends on
the two non-blocking diodes each conducting the positive and negative rectifier
current, respectively.
During the conducting period, the difference of line and dc voltage is active as
voltage drop over the line inductances and resistances. The higher the line
inductances, the smaller the line current peaks. However, the value of the line
inductances is limited due to economic and efficiency reasons. Furthermore, the
average dc voltage depends on the line inductances and the inverter output power.
The maximum dc voltage (no load) is equal to the maximum amplitude of the line
voltages. Due to voltage drops of line inductances, resistances and rectifier diodes,
the dc voltage slightly decreases with increasing load. For more details concerning
the rectifier, see [Bose 97], [Dub 89] et al.
According to figure 1.1, the dc voltage is switched in a three-phase PWM inverter
by six semiconductor switches in order to obtain pulses, forming three-phase ac
voltage with the required frequency and amplitude for motor supply. The switching
devices must be capable of being turned on as well as turned off. During the last
years, major progress has been made in the development of new power
semiconductor devices. The simpler requirement driving the power switches and the
higher maximum switching-frequency, enabling higher operating frequencies
(higher motor speed), provide continually rising output power. The new generation
of switching devices is capable of conducting more current and blocking higher
voltages. The alternatives at present are gate turn-off thyristor (GTO), MOS
controlled thyristor (MCT), bipolar junction transistor (BJT), MOS field effect
transistor (MOSFET) and insulated gate bipolar transistor (IGBT).
The IGBT is a combination of power MOSFET and bipolar transistor technology
and combines the advantages of both. In the same way as a MOSFET, the gate of the
IGBT is isolated and its driving power is very low. However, the conducting voltage
is similar to that of a bipolar transistor. Presently, IGBTs dominate the mediumpower range of variable speed drives. Since the maximal current rating of IGBT
modules is around 1 kA and the voltage rating is approximately 3 kV, they will
gradually replace GTOs at higher power levels [Vas 99].
Parallel to the power switches, reverse recovery diodes are placed conducting the
current depending on the switching states and current sign. These diodes are
required, since switching off an inductive load current generates high voltage peaks
probably destroying the power switch. Exemplary for one inverter leg, figure 1.3
presents the basic configuration and the inverter output voltage depending on the
switching state and current sign. The basic configuration of one inverter output
phase consists of upper and lower power devices T1 and T4, and reverse recovery
diodes D1 and D4.
When transistor T1 is on, a voltage Udc is applied to the load. Considering an
inductive load, the current increases subsequently. If the load draws positive current,
it will flow through T1 and supply energy to the load. To the contrary, if the load
current ia is negative, the current flows back through D1 and returns energy to the dc
source.
T1 on
Udc
C/2
T1
D1
ia > 0
Udc
Udc
C/2
C/2
T1
D4
T1 off
ua0
T4 off
Udc
ua0
T4
D1
drop
T1
drop
ua0
D1
ia < 0
Udc
ua0
T4
C/2
ia
T1 on
T4 on
t
dead
- Udc
D4
D4
drop
T4
drop
T4 on
Figure 1.3: Basic configuration of a half-bridge inverter and center-tapped inverter output voltage.
Left: Switching states and current direction. Right: Output voltage and line current.
Similarly if T4 is on, which is equal to T1 off, a voltage - Udc is applied to the load
and the current decreases. If ia is positive, the current flows through D4 returning
energy to the dc source. A negative current yields T4 conducting and supplying
energy to the load.
According to figure 1.3, with T1 on and drawing positive load current ia, the output
voltage ua0 will be less than Udc by the on-state voltage drop of T1. When the load
current reverses, the output voltage will be higher than Udc by the voltage drop
across D1. Similarly, the output voltage is slightly changed by the voltage drop of
the lower devices T4 and D4.
Normally, the on-state voltage and diode drops (1 V) are ignored and the centertapped inverter is represented as generating the voltage Udc and - Udc,
respectively. Neglecting additionally the dead-time interval dead, the behavior of the
power devices together with the reverse recovery diode is equally described by ideal
two-position switches.
Chapter 1
1.3.1
Hysteresis current control is a PWM technique, very simple to implement and taking
care directly for the current control. The switching logic is realized by three
hysteresis controllers, one for each phase (figure 1.4). The hysteresis PWM current
control, also known as bang-bang control, is done in the three phases separately.
Each controller determines the switching-state of one inverter half-bridge in such a
way that the corresponding current is maintained within a hysteresis band i.
Switching
logic
Hysteresis band i
ia
Current reference
Real current
ia*
ia
ib
ib*
ua0
1/2 Udc
ic*
ic
Output
voltage
-1/2 Udc
To increase a phase current, the affiliated phase to neutral voltage is equal to the half
dc bus voltage until the upper band-range is reached. Then, the negative dc bus
voltage - Udc applied as long as the lower limit is reached &c. More complicated
hysteresis PWM current control techniques also exist in practice, e.g. adaptive
hysteresis current vector control is based on controlling the current phasor in a /reference frame. These modified techniques take care especially for the interaction
of the three phases [Jen 95].
Obviously, the dynamic performance of such an approach is excellent since the
maximum voltage is applied until the current error is within predetermined
boundaries (bang-bang control). Due to the elimination of an additional current
controller, the motor parameter dependence is vastly reduced. However, there are
some inherent drawbacks [Brod 85]:
Chapter 1
Hysteresis current control is used for operation at higher switching frequency, as this
compensates for their inferior quality of modulation. The switching losses restrict its
application to lower power levels. Due to the independence of motor parameters,
hysteresis current control is often preferred for stepper motors and other variablereluctance motors.
A carrier-based modulation technique, as described in the next subsection,
eliminates the basic shortcomings of the hysteresis PWM controller [Bose 97].
However, when being compared to the hysteresis PWM, an additional current
control loop, calculating the reference voltages, is required when subsequent
modulation schemes are applied to high-performance motion control systems.
1.3.2
d,q
ud*
id*
id
comparator
ub*
Current
controller
uq*
iq*
iq
Current
controller
Switching
logic
ua*
comparator
uc*
a,b,c
comparator
Carrier wave
Phase a
Phase b
Phase c
Carrier wave
Uref
ua0
Upper switch on
Udc/2
t
-Udc/2
Lower switch on
ub0
t
uc0
uab
Since the PWM frequency, equal to the frequency of the carrier wave, is usually
much higher than the frequency of the reference voltage, the reference voltage is
nearly constant during one PWM period TPWM. This approximation is especially true
considering the sampled data structure within a digital control system. Depending on
the switching states, the positive or negative half dc bus voltage is applied to each
phase. At the modulation stage, the reference voltage is multiplied by the inverse
half dc bus voltage compensating the final inverter amplification of the switching
logic into real power supply.
According to figure 1.7, the mean value of the output voltage, resulting from a
reference voltage being constant within one PWM-period, depends on the on- and
off-states of the affiliated switch:
u ao =
1
TPWM
1
u a 0 dt =
(t1 t 2 ) U dc
2
T
PWM
T
PWM
(1.1)
10
Chapter 1
u a* 0
U dc 2
t
Saw-tooth
carrier wave
-1
t1
t2
u a*0
U dc 2
-1
t1/2
ua0
t2 t1/2
Triangular
carrier wave
ua0
ua0
Udc /2
ua0
Udc /2
u a 0 = u a*0
u a 0 = u a* 0
-Udc /2
-Udc /2
TPWM
TPWM
Figure 1.7: Sinusoidal modulation at constant or sampled reference voltage for one phase.
Left: Saw-tooth shaped carrier wave. Right: Triangular-shaped carrier wave.
The switch on- and off-times (t1 and t2) are calculated according to figure 1.7 by
setting the carrier wave equal to the reference voltage related to the dc bus voltage:
1+
TPWM
t1 =
t1 =
TPWM
2
u a* 0
U dc 2
(1.2)
u*
1 + a 0
U 2
dc
t 2 = TPWM t1 =
TPWM
2
(1.3)
u*
1 a 0
U 2
dc
(1.4)
Applying (1.3)-(1.4) on (1.1) shows the mean value of the output voltage ua0 being
equal to the reference voltage u*a0:
u ao =
1
TPWM
u ao = u a*0
U dc
2
TPWM
u* T
1 + a 0 PWM
U 2
2
dc
u*
1 a 0
U 2
dc
(1.5)
(1.6)
11
3
3 U phase
m,
=
2 2
2 U dc
m1
(1.7)
m>1
(1.8)
12
Chapter 1
0.8
0.7
[]
0.6
/U
dc
0.5
2 2
line
0.4
0.3
0.2
0.1
0
m[]
When m is made sufficiently large, the phase voltage becomes a square wave and the
line voltage becomes a 6-step waveform.
ub*0
U dc 2
1
-1
u a* 0
U dc 2
t
carrier wave
ua0
4 U dc
2
ua0
fundamental
U dc
2
uab
U dc
U dc
2
U dc
2
U dc
ub0
Figure 1.9: Strong overmodulation and square-wave shaped output voltage with affiliated fundamental.
Top: Reference voltages (u*a0, u*b0) and carrier wave. Middle: Phase-to-neutral output voltage ua0 and
affiliated fundamental. Bottom: Phase-to-neutral output voltage ub0 and line voltage uab.
4 U dc
2
n 2n 1 sin[(2n 1) t ]
(1.9)
13
2
U phase,max = U dc
(1.10)
m=1
ia [A]
m = 1,33
-2
-4
0.01
0.02
t [s]
0.03
0.04
0.05
Basic drawbacks of the sinusoidal PWM are the not ideal use of the dc bus voltage
and the non-existent interaction between the three phases resulting in superfluous
changes of switching states, increasing semiconductor losses and introducing a
higher harmonic content at the motor terminals.
1.3.2.1
According to (1.9), also multiple of third harmonics are present in the voltage
spectrum. However, the third harmonics are eliminated and not existent in the
current spectrum since the sum of the phase current of a three-phase ac machine
equals zero. As shown in figure 1.11, the range of the sinusoidal PWM can be
increased by adding third harmonics to the reference voltages. The same third
harmonic is added to each of the three reference voltages. Adding third harmonics
agrees with a simultaneous variation of the potential in all phases, thus not
recognized at the terminals of an ac motor with isolated neutral point:
!
(1.11)
Therefore, the introduction of a third harmonic does not distort the line voltages
since third harmonic components in the phase voltages are cancelled.
14
Chapter 1
A geometrical calculation yields the maximum possible increase of the linear area
with the harmonic amplitude being 1/6 of the reference voltage amplitude. Such an
injection of a third harmonic results in a 15,5% higher maximum output voltage
without overmodulation. According to [Jen 95], the harmonic content of the
resulting current spectrum of ac motor drives is minimal at injection of a third
harmonic with the amplitude being 1/4 of the reference voltage amplitude, still
increasing the maximum output voltage without overmodulation by 12%.
u*a0
u a*0
U dc 2
reference plus
third harmonic
third harmonic
t
-1
ua0
carrier wave
U dc
2
U dc
2
Obviously, also multiple of third harmonics do not disturb the current spectrum and
are suitable injection signals. As can be shown [Jen 95], the subsequently described
space vector modulation is equal to the sinusoidal PWM with injection of a suitable
triangular-shaped signal containing all existing multiple of third harmonics.
1.3.3
Space-vector pulse width modulation has become a popular PWM technique for
three-phase voltage-source inverters in applications such as control of induction and
permanent magnet synchronous motors. The mentioned drawbacks of the sinusoidal
PWM are vastly reduced by this technique. Instead of using a separate modulator for
each of the three phases, the complex reference voltage phasor is processed as a
whole. Therefore, the interaction between the three motor phases is exploited. It has
been shown, that SVM generates less harmonic distortion in both output voltage and
current applied to the phases of an ac motor and provides a more efficient use of the
supply voltage in comparison with direct sinusoidal modulation techniques [Jen 95].
15
As shown in table 1.1, there are eight possible combinations of on and off patterns
for the three upper electronic switches feeding the three-phase power inverter
(figure 1.1). Notice that the on and off states of the lower power switches are
opposite to the upper ones and so completely determined once the states of the upper
power electronic switches are known. The phase voltages corresponding to the eight
combinations of switching patterns can be mapped into the / frame through /transformations [Hen 92]. This transformation results in six non-zero voltage vectors
and two zero vectors. The non-zero vectors form the axes of a hexagonal containing
six sectors (S1 S6) as shown in figure 1.12. The angle between any adjacent two
non-zero vectors is 60 electrical degrees. The zero vectors are at the origin and apply
a zero voltage vector to the motor. The derived / voltages in terms of the dc bus
voltage Udc are summarized in table 1.1.
Table 1.1: Switching table and / transformation of affiliated state voltage vectors.
/-transformation of the states
Switch no.
State
S1
S3
S5
Ux,
Ux,
|Ux|
000
OFF
OFF
OFF
100
ON
OFF
OFF
2 U dc 3
2 U dc 3
U dc
2 U dc 3
U dc
2 U dc 3
110
ON
ON
OFF
U dc 3
010
OFF
ON
OFF
U dc 3
011
OFF
ON
ON
2 U dc 3
001
OFF
OFF
ON
U dc 3
U dc
2 U dc 3
101
ON
OFF
ON
U dc 3
U dc
2 U dc 3
111
ON
ON
ON
2 U dc 3
U
010
110
S2
S3
S1
000
111
011
S4
001
S5
S6
100
101
Figure 1.12: Hexagon, formed by the basic space vectors and sector definition (S1 S6).
Using the transformation of the three phase voltages to the / reference frame, the
voltage phasor Uref represents the spatial phasor sum of the three phase voltages.
When the desired output voltages are three-phase sinusoidal voltages with 120
16
Chapter 1
phase shift, Uref becomes a revolving phasor with the same frequency and a
magnitude equal to the corresponding line-to-line rms voltage.
The objective of the space vector PWM technique is to approximate the reference
voltage phasor Uref by a combination of the eight switching patterns. Practically,
only the two adjacent states (Ux and Ux+60) of the reference voltage phasor and the
zero states should be used [Jen 95] as demonstrated by the example in figure 1.13.
The reference voltage Uref can be approximated by having the inverter in switching
states Ux and Ux+60 for t1 and t2 duration of time respectively.
U ref =
1
TPWM
(t1 U x + t 2 U x+60 )
(1.12)
Of course, the affiliated sector must be known first. The sector identification and the
calculation of t1 and t2 are presented in the next subsection. Since the sum of t1 and t2
should be less than or equal to TPWM, the inverter has to stay in zero state for the rest
of the period. The remaining time t0 is assigned to one or both zero-voltage phasors.
t 0 = TPWM t1 t 2
(1.13)
Applying only one of the two zero-voltage states during a PWM period, results in an
asymmetric edge-aligned PWM signal. This is often undesired (higher harmonics)
but reduces the required switching number by 33% since one inverter leg does not
switch during that particular PWM period. Here, the remaining time t0 is equally
assigned to both states. As illustrated in figure 1.13, all state changes are obtained in
each case by switching only one inverter leg.
000
111
000
100 110
110 100
40% 100
ua0
5% 000
50% 110
Uref = U ejt
ub0
5% 111
uc0
TPWM
As mentioned above, the reference voltage is actually equal to the desired threephase output voltages mapped to the / frame. The envelope of the hexagonal
formed by the basic space vectors, as shown in figure 1.12, is the locus of the
maximum output voltage. In order to avoid overmodulation, the magnitude of Uref
must be limited to the shortest radius of this envelope. This gives a maximum rms
value of the line-to-line and phase output voltages of
U line , max =
3
2
17
U
U phase , max = dc
2
(1.14)
being approximately 15% higher when compared to the original sinusoidal PWM.
1.3.3.1
Presently in industry, the SVM is often applied as inverter control strategy because
of its advantages when compared to other PWM techniques: SVM provides efficient
use of the supply voltage and low harmonic distortion in both output voltage and
current. Furthermore, it can easily be implemented with modern DSP-based control
systems. Even recent developments of the DTC-algorithm are modified in regard to
exploit the advantages of the SVM.
As shown in table 1.1, the reference voltage Uref, usually represented by its /
components U* and U*, can be approximated easily by a linear combination of the
two adjacent states and the zero states, i.e. no trigonometric functions are required to
calculate the duty cycles. First, the sector must be identified to determine the
appropriate states. This is performed, as illustrated in figure 1.14, by a comparison
of the / components specifying the position in the /-plane. For instance, if the
reference voltage U* is positive, the sector of the reference voltage is in the upper
half of figure 1.12 (sector S1, S2 or S3). Otherwise, the sector is in the lower half.
Further sector splitting/identification is done by comparison (geometrical
calculation) of the - and -components. The applied normalization at the beginning
eliminates the dc bus voltage dependence of the output voltages. The resulting duty
ratios (a*, b* and c*), as required for PWM generation using e.g. TIs TMSM320P14
DSP, are calculated according to following flowchart. A duty ratio a* = 1 indicates a
continuously closed upper switch of the first inverter leg. At a duty ratio a* = 0, the
turn-on time during each PWM period is equally distributed to the lower and upper
switch and the resulting mean value of the phase voltage ua0 is zero. At a duty ratio
a* = -1, the lower switch is continuously closed, etc.
The relation between the duty cycles of the three phases in percent (relation of the
switch-on to the switch-off times of the three inverter legs within one PWM period)
and the given duty ratios a*, b* and c* is:
a* + 1
;
duty cycles a ; b; c =
2
b* + 1
;
2
c* + 1
100 %
2
(1.15)
Usually, the presented algorithm is easily incorporated into the initialization part of
the real-time program, e.g. by including handwritten C-code. Then, the duty ratios
are directly mapped by a DSP into signals driving the inverter switching logic. As
illustrated in figure 1.14, a final data processing and transmission is required, when
18
Chapter 1
additionally a slave DSP generating the PWM pulses, e.g. TIs TMSM320P14, is
used. To avoid overflow of the fixed-point slave DSP, all duty ratios must be limited
to 1. Since the P14 DSP uses 16-bit compare registers for the PWM generation,
the calculated values are adjusted by the given multiplication before they are finally
transmitted to the slave DSP generating the PWM signals. As illustrated in a
subsequent chapter (e.g. figure 3.2), each two PWM channels are employed to
generate the correct pulses for the inverter.
Voltage
reference
U*, U *
3 1
2 U dc
normalization
u 0
*
No
Yes
u*
1
3
No
u *
u*
Sector 1
No
1
3
Sector 2
Sector 1 & 4:
1 *
a * = u* +
u
3
b * = u* +
3
3
u *
No
No
Sector 3
Sector 4
u*
1
3
Sector 5
Sector 2 & 5:
c * = a *
b* =
2
3
u *
u *
u *
Sector 6
c * = u*
15
2 -1
|u| 1
Overflow
protection
b * = a *
c * = b*
duty ratios:
(a*; b*; c*)
Sector 3 & 6:
1 *
a * = u*
u
3
a * = 2 u*
u *
P14
DSP
3
3
u *
PWM 16
16 bit
compare
register
The turn-on times t0, t1 and t2 of the applied switching states during each PWM
period, as introduced in (1.12)-(1.13) for illustration purpose, are not required for
implementation of the SVM. However, they are easily calculated by the duty cycles
of the three phases. For instance, the zero states 000 and 111 are each equal to
the minimum of the duty cycles given in (1.15) multiplied by the PWM period TPWM.
19
T1 on
Udc
T1
C/2
D1
Udc
T1
C/2
D1
ia < 0
ia > 0
Udc
ua0
C/2
T4
Udc
D4
ua0
C/2
T4
D4
T4 on
T1 off
Ideal gating
pulse pattern
T1
T4
pulse pattern
with dead time
T1
T4
T1
T4
pulse pattern
with dead time
T1
T4
UT1
dead
dead fPWM Udc
Udc
UD1
ua0
ua0
Udc
dead
Ideal gating
pulse pattern
- Udc
- Udc
UD4
UT4
Considering the no-load case, the storage time of the semiconductors is very small
when compared to the dead time: Switching off a power device, the current
20
Chapter 1
commutates directly to the diodes. This condition results in the desired voltage,
which is applied to the motor terminals. In contrast to this, switching on a power
device is delayed by the dead time. During the dead-time interval, the diode
continues conducting until the dead time elapses and the opposite power device is
switched on. This condition results in a loss of voltage at the motor terminals
indicated by the gray marked area in figure 1.15. With a positive current, the duty
cycles are shorter and with negative currents longer than required. Hence, the actual
duty cycle of a bridge is always different from the one of the reference voltage. It is
either increased or decreased, depending on the load current polarity. Furthermore,
the voltage drops of the power switches UT, respectively the voltage drop of the
reverse recovery diode UD, are considered. Summarizing, the voltage distortion can
be described by an error voltage U
U
UT + U D
+ dead f PWM U dc
2
(1.16)
depending on the dead time dead, the dc bus voltage Udc, the PWM frequency fPWM
and the voltage drops UD and UT of IGBT and diode [Bose 97]. This error voltage
and the resistances RT and RD of the switch changes the inverter output voltage ua0
from its intended value uref to:
u a 0 u ref i
RT + RD
U sign(i ) ,
2
(1.17)
The dead time reduces the effective turn-on time and produces the undesired fifthand seventh-order harmonics in the inverter output voltage [Dod 90]. Furthermore, it
generates sub-harmonics, resulting in torque pulsation and possible instability at
low-speed and light-load operation [Leg 97], [Mur 92]. The resulting speed
oscillation and the voltage distortion are illustrated in figure 1.16 showing the deadtime effect on a 1,5 kW induction motor drive in open loop (scalar) control at low
speed and light-load operation. Considering the given drive setup (dead = 2,5 s;
fPWM = 10 kHz) and according to (1.16), the error voltage amounts to U = 12,5 V
(equal to 15,3 V in the alpha/beta reference frame). A reduction of the average
voltages occurs according to (1.17) when one of the phase currents changes its sign.
The motor currents have the tendency to maintain their values after a zero crossing.
In generator mode, the behavior of the motor current is contrary resulting in a
steeper rise of the current after zero crossing. In any case, the motor torque is
influenced as it can be observed by speed oscillations at six times of the fundamental
frequency.
The dead-time problem is more serious in high-power gate-turn-off thyristor (GTO)
inverter systems than in the case of IGBT or MOSFET inverters, since the GTO
requires a longer dead time. However, the use of fast switching devices using high
carrier frequencies (5-20 kHz) with lower dead-time values (1-5 s), will not free
the system of the described distortion. Higher PWM frequencies improve the
21
waveform quality by raising the order of theoretical harmonics, but low frequency
sub-harmonics persist due to the dead time. Furthermore, to avoid unnecessary
switching losses and short-term overheating of a switching device, minimum time
duration in the switching states must be forced. If the commanded voltage value is
less than the required minimum, the affiliated switching state must be either
extended in time or skipped. This causes additional distortion of the inverter output
voltages. Therefore, a compromise must be made by choosing a suitable PWM
frequency: a high PWM frequency improves the theoretical quality of the waveform,
but may increase simultaneously the voltage distortion due to the dead-time effect.
4
60
55
n [rpm]
I [A]
-2
-4
50
45
40
0.25
0.5
0.75
35
0.25
0.5
0.75
t [s]
40
20
20
U [V]
40
uref
u (u)
U [V]
t [s]
-20
-40
-20
uref
0
0.25
0.5
0.75
t [s]
-40
-40
-20
20
40
U [V]
Figure 1.16: Open-loop control of an induction motor & dead-time effect (Udc =500 V, no load).
Left: Measured current I, measured voltage U and reference voltage Uref.
Right: Measured/reference speed and voltage trajectories.
1.4.1
Dead-Time Compensation
Remarkable efforts have been made to compensate the voltage distortion due to the
dead-time effect [Choi 96], [Leg 97], [Sep 94]. Most dead-time compensation
methods are based on an average value theory: the lost voltage is averaged over an
operating cycle and added vectorially to the command voltage [Mur 87], [Jeo 91].
Dead-time compensation can be implemented in hardware or software.
The hardware compensator operates by closed loop control [Mur 87]. Previous
commutation times are measured and used to control the next duty cycles. However,
a potential-free measurement of the inverter output voltages is required. Software
compensators are mostly designed in feed-forward mode. Depending on the sign of
the respective phase current, a fixed time delay is either added to or subtracted from
the command voltage.
22
Chapter 1
60
55
n [rpm]
I [A]
-2
-4
50
45
40
0.25
0.5
0.75
35
0.25
20
20
U [V ]
40
0.75
uref
u (u)
-20
-40
0.5
t [s]
40
U [V ]
t [s]
-20
uref
0
0.25
0.5
t [s]
0.75
-40
-40
-20
20
40
U [V ]
Figure 1.17: Field-oriented control of an induction motor & dead-time effect (Udc =500 V, no load).
Left: Measured current I, measured voltage U and reference voltage Uref.
Right: Measured/reference speed and voltage trajectories.
The influence of the dead time on the current/torque is vastly reduced by the speed
and current control loop. Of course, the falsification of the motor terminal voltages
is the same, but the harmonic distortion of the fundamental voltage is transmitted to
the reference voltages. Due to the arguable compensation by the current controller,
common industrial drives are not always equipped with an additional dead-time
compensation.
Note, that permanent magnet synchronous motor drives behave more sensitive to the
dead-time effect than induction motor drives: Due to the absence of a magnetizing
component in the stator current and the low main reactance, they tend to operate
partly in discontinuous current mode at light load. These machines require an
advanced compensation scheme when applied to high-performance motion control
systems or, alternatively, an additional d-axis current to bridge the discontinuous
current time intervals [Bose 97].
1.4.2
23
Dead-Time Generation
The switching transitions of real switches, especially the transition from current
conducting to voltage blocking, are not infinitely fast. After conducting, a finite time
is required, mainly to remove the space charge, before a semiconductor switch is
able to block the supply voltage. Switching off a power device, the current
commutates to the opposite recovery diode (constant current direction) and the
power switch starts to block the dc voltage. If a switch of one inverter leg is turned
on before the opposite switch blocks the dc bus voltage, the whole dc bus voltage is
shorten across this leg (figure 1.1) resulting in a very high short-circuit current only
limited by the resistances of the power switches. Obviously, such a high short-circuit
current may destroy the power switches as well as the drive system and the dc link
capacitor.
To avoid such short-circuit conditions, a dead-time interval is added between the
turn-off signal of a switch and the turn-on signal controlling the opposite switch.
Dead time control prevents any cross-conduction or shoot-through current from
flowing through the main power switches during switching transitions by controlling
the turn-on times of the semiconductor drivers. The high-side driver is not allowed
to turn on until the voltage at the junction of the opposite power switch is low and
vice versa. During the dead-time interval, recovery diodes continue conducting until
the dead time elapses and the opposite power device is switched on.
In modern DSP systems, the dead time generation is usually programmable, e.g.
added as extra time in a compare register/timer. Considering analog circuits, the
fixed dead-time generation of one half-bridge is easily generated by a RC-circuit
coupled to two optocouplers, each controlling the opposite switches of one inverter
half-bridge as described in figure 1.18. Additionally, such a hardware realization
takes care for galvanic isolation of the digital control system and the power
electronics. The resistance R is calculated by the resistance voltage drop divided by
the operating current of the optocoupler IP:
R=
Us Ud
IP
(1.18)
!
U c (t = dead ) = (U s + U d ) e R C 1 + U d = U d
(1.19)
24
Chapter 1
dead
2 Ud
R ln1
Us + Ud
(1.20)
Switching logic
US
Optocoupler 1
IP1
-Us
UC
UC
-Ud
UC
C
PWM logic:
US = 12V
IP2
Optocoupler 2
-Ud
IP
IP1
IP2
dead
IP1
dead
Figure 1.18: Analog dead-time generation. Left: Exemplary hardware circuit for one inverter leg.
Right: Switching logic, voltage and affiliated current of an optocoupler driving the power switch.
25
For supply voltages over 500V ac, a motor with an enhanced winding
insulation system is required. Alternatively, additional components can be
added to limit the voltage stresses to acceptable levels.
Where the drive spends a large part of its operating time in braking mode,
the effect is similar to increasing the supply voltage by up to 20%.
For drives with PWM active front ends (regenerative and/or unity power
factor), the effective supply voltage is increased by around 15%.
1.6 Conclusions
Controlled power supply for electric drives is obtained usually by converting the
mains ac supply. A typical converter consists of power electronic circuits,
employing switching devices such as thyristors, transistors, GTOs, MOSFETSs,
IGBTs and diodes as well as a host of associated control and interfacing circuits.
The conversion process allows fast control of voltage, current or power to the motor
via the gate circuits of the converter switches. In this way, the required dynamic
response requirements of high-performance ac motor drives can be met.
This chapter provides a detailed survey of voltage-source PWM inverter drives with
emphasis on the modulators and control methods. The most common three-phase
inverter topology is that of a switch mode voltage source inverter. VS-inverters
consist of two main sections, a controller to set the operating frequency and a threephase inverter to generate the required sinusoidal three-phase voltage from a dc bus
voltage.
The basic concepts of pulse width modulation are illustrated. PWM is the process of
modifying the width of the pulses in a pulse train in direct proportion to a small
control signal. The greater the control voltage, the wider the resulting pulses
become. By using a sinusoid of the desired frequency as control voltage for a PWM
circuit, it is possible to produce a high-power waveform whose average voltage
varies sinusoidally in a manner suitable for driving ac motors. Due to the significant
flexibility in controlling the inverter switches, a large number of switching
algorithms were introduced and some of these have gained wide acceptance and are
fully developed.
Usually, the behavior of the power devices together with the reverse recovery diode
is described by ideal two-position switches. In practice, a dead-time interval is
required to prevent the shoot-through effect of a half-bridge during a change of the
switching states. Although the dead time is short, it causes deviations from the
desired fundamental inverter output voltage. Issues of the resulting phase voltage
distortion due to the inverter non-linearity as well as compensation methods are
discussed in detail.
2.1 Introduction
Voltage dips and sags of short duration constitute a serious problem for many
applications and especially for variable speed drives (frequency converters) in
industry. Early types of frequency converter for motor drives were notoriously
sensitive to supply disturbances and often had to perform a full stop and restart to
resume operation. The economic impact, of what actually is a mere incident,
therefore could turn out to be quite substantial.
Usually, voltage source PWM inverter drives are equipped with an under-voltage
protection mechanism, causing the system to shut down within a few milliseconds
after a power interruption in the regular grid. This shut down mechanism can be
associated with a total loss of system control since the control electronics are usually
powered by the (in this case discharged) dc-link capacitor. Particularly in multimotor drives, a loss of mutual synchronization may be critical. This may entail
damage or loss of material in sensitive applications as the production of textile
fibers, paper mills, or extrusion drives. Generally, it is required to wait until the
machine has come to a complete standstill to enable restarting [Baa 89]. Braking to
zero speed and restarting obviously is not an adequate solution. Many continuous
production processes in industry are sensitive to a larger variation in speed or losing
control at worst. In addition, time and additional workload required to get a plant
ready for restart may be considerable.
This chapter discusses a design concept avoiding the standstill/restarting interval at
power interruptions. The proposed solution to the problem is to recover some of the
mechanical energy stored in the rotating masses by kinetic buffering. When the
power supply is interrupted, a dc link voltage control is applied to force an
immediate transition into the regenerative mode. During the interruption interval, the
drive system continues to operate at almost zero electromagnetic torque, just
regenerating a minor amount of power to cover the electrical losses in the inverter.
This maintains the dc link capacitor well charged, keeping the electronic control
circuits active, since they are supplied from the dc link through a switched mode
converter. In this way, the drive remains controllable even at power interruptions of
several seconds. Of course, the (still controlled) braking of the drive depends on the
28
Chapter 2
actual load torque. Since drive control is never lost, the voltage control scheme can
be applied to multi motor drives as well.
The implemented regenerative braking scheme allows the inverter to keep its dc bus
voltage at a predetermined minimum level as long as possible, expanding the time in
which supply voltage can be reapplied without the time-consuming dc-link capacitor
recharging cycle. The temporary speed dip is generally tolerable, since the most
frequent power interruptions last only for a few milliseconds. The implemented
voltage control scheme is derived from a torque controlled dc bus voltage.
Considering realistic conditions, the ride-through capability at short-time power
interruptions is discussed. Measured results are presented and evaluated to
demonstrate the performance and the stability of the system.
29
30
Chapter 2
energy of the moving masses of motor and driven system is substantially higher.
This reservoir can be tapped for bridging the time interval of power interruptions.
Energy is fed back from the rotating masses to the dc link circuit to maintain the dc
link voltage at a predetermined level. This is possible also in the presence of
additional loads connected to the dc link.
The principle of forcing a fast reversal of power flow at a breakdown of the supply
voltage is explained by the trace of the dc bus voltage according to figure 2.1.
Normally, the dc voltage changes within certain limits as indicated by the (shaded)
regular voltage band. The lower limit allows for voltage sags due to load variations,
fluctuations of the supply voltage or single-phase voltage dips. The upper voltage
limit may be reached at fast decelerations of the drive. Normally, a rising dc voltage
forms no problem since the generated kinetic energy can be conducted using a
brake-resistance within the dc link, a common dc bus or a two-way PWM inverter.
Nevertheless, the proposed ride-through scheme can be adopted allowing a
controlled deceleration within a maximum predetermined dc link voltage. This can
be used to save energy rather than a fast deceleration with power dissipation of, e.g.,
a brake-resistance. First, only a low voltage ride-through scheme bridging the time
of a three-phase power interruption is considered. The latter approach is presented at
the end of this chapter.
With reference to figure 2.1, the power supply is interrupted at t = t1. The power
interruption is detected at t2 when the dc bus voltage reaches the predetermined level
UKB causing the system to switch automatically to voltage control mode. Thereafter,
the voltage is maintained by a closed loop control forcing the drive system to
operate at almost zero electromagnetic torque. The motor regenerates just a minor
amount of power by kinetic buffering to cover the electrical losses in the inverter
and motor until the return of the power supply at t4. The return of the power supply
results in a fast rise of the dc link voltage. This reactivates the regular speed control
of the drive at t5 and the motor accelerates to the set value.
Udc
Regular voltage band
Udc,N
UKB
Under-voltage
protection
Dip
detection
Umin
t1
t2 t3
t4
t5
1
0
The lower trace of figure 2.1 shows a logic signal indicating the detected event of a
power interruption. This signal is used in order to switch between voltage and speed
31
control mode. If the inverter control did not react on this signal, the dc bus voltage
continues falling as indicated by the dashed line. The inverter would shut down at t3
by the under-voltage protection at the voltage level Umin. Without kinetic buffering,
the maximum acceptable duration of a power interruption can be determined by
3
1
2
C U dc2 , N U min
= (Ploss + Tload ) dt
2
t1
(2.1)
where Ploss is the power dissipation of motor and inverter. In speed control mode, the
motor speed and consequently the losses as well as the load torque are usually
constant. Typical values of this time interval, mainly depending on the prevailing
mechanical power at the motor shaft, are of the order of 1-50 ms. Of course, the
voltage control must become active before this time has been elapsed. The
maximum time interval tmax of bridging power interruptions by kinetic buffering
can be appraised by solving:
1
1
2
2
C U dc2 , N U min
+ J ref
= (Ploss + Tload ) dt
2
2
t max
(2.2)
In contrast to (2.1), losses and load torque are now speed dependent. The stored
kinetic energy is obtained by the inertia of the moving masses and the actual speed
at power interruption, normally equal to the reference speed ref. Using kinetic
buffering, a maintained and controlled operation of several seconds is possible.
32
Chapter 2
higher than the output. The voltage control must immediately accelerate the PMSM
to stay in the MPP of the PV array. With decreasing irradiance, the power of the PV
array is smaller than the output in the dc bus. The difference comes from the
capacitor, being discharged. This is the most critical condition. The dc bus collapses,
if this condition remains resulting in a voltage drop beyond given limits. Hence, the
inverter must slow down the PMSM to a new stable operating point. Therefore, the
voltage controller has to accelerate/decelerate the motor very quickly guaranteeing a
balanced input/output power ratio in the dc bus. Figure 2.2 shows the energy flow
within the system without loss considerations.
IPV
IInv
Solar
generator
motor-pump
system
Idc
Pkinetic
PPV
Udc
Ppump
The energy generated by the PV array is used to drive the motor/pump system.
Depending on the difference between energy generation and consumption, the dc
bus capacitor is charged or discharged:
U dc =
1
I dc dt
C
(2.3)
The dynamic behavior of the voltage control is determined by energy equations. The
electromagnetic power developed by the motor can be divided in kinetic power
Pkinetic accelerating the motor-pump system and pumping power Ppump. Only the
kinetic power can be used to feed back energy to the dc bus and to control the
voltage.
Tel = J
d
+ Tload
dt
(2.4)
d
+ Tload
dt
(2.5)
Subsequently, the drive efficiency is not taken into account, because of the opposite
influence at acceleration and braking. The losses are small compared to the
mechanical energy consumption. Furthermore, the loss fluctuation is almost as
slowly as the variation of the pumping power. Therefore, they are as being a part of
33
the load. Without considering the drive efficiency, the input power of the inverter
matches the electromagnetic output power generated by the motor.
Pel U dc I Inv = U dc (I PV I dc )
(2.6)
In steady state, the voltage Udc and motor speed are constant. The energy
generated by the PV array is completely used to pump water:
2.4.1
U dc = const I dc = 0
= const
Pkinetic = 0
Ppump U dc I PV
(2.7)
(2.8)
Normally, the motor speed of a conventional drive supplied by a regular grid via a
diode rectifier is completely independent of the dc bus voltage. Here, a PV array is
the source and a water pump acts as load. A relation between motor speed and dc
bus voltage can be obtained by linearization of the dynamic behavior. The
electromagnetic torque of the motor can be controlled very fast given the bandwidth
of the current control loop (960 Hz), whereas the load torque varies slowly with
speed. The speed can be controlled beyond current/torque limitation with a
bandwidth of approximately 26 Hz. Therefore, also the kinetic power Pkinetic can be
varied faster than the pumping power Ppump. Due to similar considerations, the dc
current Idc can be controlled faster than the dc bus voltage Udc. Therefore, the
following equation is valid during transients:
J
dT
d
d 2
>> load
2
d dt
dt
(2.9)
Using (2.5)-(2.6) and assuming constant pumping power and constant current IPV of
the PV array for a short time, the linearized relation between dc voltage and motor
speed is described by:
U dc I PV U dc I dc = J
d
+ Tload
dt
d
U dc I dc
dt
(2.10)
(2.11)
(2.12)
34
Chapter 2
U dc C
dU dc
d
= J
dt
dt
(2.13)
With the transfer function of the closed loop speed control beyond current/torque
limitations
(s)
1
=
* ( s ) s speed + 1
(2.14)
and using (2.13), the resulting linearized transfer function with the reference speed
* as input and the dc bus voltage Udc as output can be written as
U dc ( s )
=
* (s)
J
1
1
,
C s speed + 1 s Vf + 1
(2.15)
where speed is the equivalent time constant of the speed control loop and Vf the time
constant of the voltage measurement including all other smaller time constants.
In fact, the loop to be controlled covers a dominant time constant and a smaller time
constant. Using a PI controller, the dominant time constant can be equalized. The
cut-off frequency of the control loop is calculated by setting the time constant of the
PI controller equal to the largest open loop time constant and choosing a phase
margin guaranteeing a stable system:
u = speed
(2.16)
R ( c ) = arctan ( Vf c )
c = tan(
(2.17)
R ) / Vf
(2.18)
C u
c 20 log Vf c
A( ) = = 20 log
J K pu
K pu = u
C
c
J
( )
Vf
+1
+ 1 = 0
(2.19)
(2.20)
35
During practical investigations, the best results have been obtained using a common
PI controller for the voltage control and choosing a phase margin R = 85. The
input of this inner control loop is the voltage error, calculated from the measured
and filtered dc bus voltage and a reference voltage given by the main control loop.
The PI controller used is equipped with an anti-windup system limiting the
maximum allowed speed of the drive (figure 2.3).
Udc*
Kpu
Udc
T s / u
z
-1
|| < max
The dc bus voltage controlled by the speed of the motor has significant drawbacks.
Choosing a phase margin R = 85, the voltage control loop has a very low
bandwidth fB = 14 Hz. Decreasing the phase margin leads to involuntary speed
oscillations. By no means, the voltage can be controlled faster than the underlying
speed, if such a cascaded structure is proposed. The speed control loop has a
bandwidth fB 26 Hz. Some approaches described in literature suffer also from such
oscillation effects [Mul 97].
Subsequently, the described MPPT is performed by varying the dc voltage
triangularly. However, applying a ramp (U/s2) as a reference voltage and using
(2.15) results in a steady state voltage error Uerror:
U error =
U u
K pu
C
J
(2.21)
The implemented speed based voltage control turned out to malfunction at very
quickly changing irradiance power. However, no undesired crash of the entire
system due to a completely discharged capacitor has been detected during the
practical tests. Nevertheless, the voltage error between optimum and measured
voltage amounts to 10% (~20 V) during such power transients (e.g. passing clouds),
what is absolutely not acceptable for a good working MPPT and for the claim to
pump as much water as possible. Therefore, the voltage has to be controlled in
another way as described in the next subsection.
2.4.2
36
Chapter 2
Tel ( s )
*
Tel ( s )
(2.22)
s eq,i + 1
Neglecting the load torque, the following relation between motor speed and
electromagnetic torque is valid:
(s)
1
,
=
Tel ( s ) J s
Tload = 0
(2.23)
In fact, the load torque is presently handled as a system disturbance, being true
considering pumping and PV power to be equal in steady state.
Replacing the speed in (2.14)-(2.15) by the electromagnetic torque Tel defined in
(2.22)-(2.23), results in a linearized transfer function with the reference torque Tel* as
input and the dc bus voltage Udc as output:
U dc ( s )
*
Tel ( s )
1
1
1
,
s
s
+
1
s
JC
eq ,i
Vf + 1
(2.24)
The voltage can be controlled directly by the electromagnetic torque of the motor. A
PI controller equipped with an anti-windup system limiting the maximum allowed
torque/current is used to calculate the reference torque. The parameters of the PI
controller are determined by choosing the time constant u larger than the sum of the
two open loop time constants and setting the gain Kpu in order to get a maximum
possible phase margin R, guaranteeing a stable system:
Tu = k eq ,i + Vf = k ,
K pu =
with: k > 1
kJC
(2.26)
R ( c ) = arctan ( k ) arctan (
(2.25)
(2.27)
Best results are obtained by choosing 10 < k <40, corresponding to a phase margin
of 55 < R < 72.
The practically implemented torque controlled voltage loop has a bandwidth
fB 235 Hz, being 16 times larger than the other approach. The calculation of the
bandwidth takes no current/torque limitation into account.
37
(2.28)
The multiplication with the sign of the motor speed is dropped in the PV-powered
control system since the pump is driven only in one (positive) direction.
38
Chapter 2
Power
supply
Udc
Udc
*
PI voltage control
*
Tel Torque
*
Control ub
&
*
uc
|T| < Tmax
EKF
Udc
Enable
voltage
control
Speed
reference
-Kpu
Ts /u
Kpn
Ts /n
Dip logic
ua
z-1
ib
PI speed control
sign()
load
SVM
Inverter
ia
AC
motor
Figure 2.4: Block diagram of the dc bus voltage control integrated into the speed control system.
The entire system is controlled by a digital signal processor (DSP) realizing dc bus
voltage control, speed/torque control of the drive and start-up and shut down
automatism's. All control and measurement units are supplied by the dc bus. A dc
voltage beyond given limits leads inevitably to an undesired crash of the entire
system. In particular, the supply of power to the electronic control circuits of the
inverter must continue without interruption to maintain the system in operation.
39
500
[V ]
400
dc
300
Under-voltage protection
200
100
0
10
12
14
10
12
14
t [s]
1000
n [rpm]
800
600
400
200
0
Out of control
0
t [s]
Figure 2.6 shows the experimental result of the voltage control enabled when a short
time three-phase power interruption is applied. After detecting the voltage dip, the
voltage controller has to decelerate the motor very quickly guaranteeing a balanced
input/output power ratio in the dc bus. Otherwise, the dc bus would be discharged
and the system collapses. However, the implemented regenerative braking scheme
allows the inverter to keep its dc bus voltage at the pre-determined minimum level,
expanding the time in which supply voltage can be reapplied without the timeconsuming dc-link capacitor recharging cycle.
Initially, the drive system is in speed control mode with a reference speed
n* = 1000 rpm. If the capacitor is discharged to a level lower than UKB =340 V, a
voltage dip is detected and the system switches automatically to the voltage control
mode with a predetermined voltage reference of Udc* = 340V. Choosing the reference
voltage Udc* lower than the dip logic switch on point UKB results in a current/torque
peak at the beginning of the voltage control mode: As can be seen in [Ter 00b], the
controller starts then with an involuntary acceleration of the drive. The system
returns to the speed control mode at a voltage level higher than Udc = 360 V or if the
motor speed is higher than the speed reference. The deceleration of the motor during
the power interruption is small, because no load is applied. Figure 2.7 shows the
experimental results of a comparable power interruption but with a load torque
applied to the motor. The applied load amounts to 75% of the rated torque. Due to
the load, the deceleration is much faster. However, the power needed to keep the
voltage at a minimum level is the same, as can be seen at the small negative q-axis
current during the interruption interval. In fact, this power (~20 W) generated by the
kinetic energy of the drive system is nearly constant and almost completely used to
compensate the inverter losses.
40
Chapter 2
420
[V ]
400
voltage dip
dc
380
360
340
320
-1
-0.5
0.5
1.5
2.5
3.5
t [s]
15
800
10
i [A]
20
1000
n [rpm]
1200
600
400
200
-5
-1
-1
t [s]
t [s]
[V ]
400
dc
voltage dip
380
360
340
320
-1
-0.5
0.5
1.5
2.5
3.5
1200
20
1000
15
800
10
i [A]
600
n [rpm]
t [s]
400
200
-5
-1
t [s]
-1
t [s]
Figure 2.7: Ride-through at power interruption with load torque (75% rated torque).
Top: Voltage of the dc link. Bottom: Motor speed and q-axis current.
Without the voltage dip control, the capacitor is completely discharged, considering
the given experiment and according to
1
C U 2 Tload t
2
1 CU 2
t =
2 Tload
(2.29)
41
200
dc
[V ]
300
100
0
-1
Under-voltage protection
0
t [s]
1200
n [rpm]
1000
800
600
400
200
0
-1
t [s]
42
Chapter 2
Power
supply
PWM converter
Common dc bus
Inverter 1
Inverter 2
Brakingresistance
S1
Motor 1
Load 1
Motor 2
Load 2
With reference to the given drive, the system switches automatically to voltage
control mode at a preset level higher than Udc > 600 V. The block diagram of the dc
bus voltage control integrated into the speed control system is equal to the earlier
described structure (figure 2.4). Once in voltage control mode, the system continues
to operate with the predetermined
voltage reference Udc* = 600V. Finally,
Udc
the system turns back to speed control
Switching
logic
mode when the motor speed reaches
U > 600V 1
*
U < 590V 0
the reference speed or at a voltage level
| * | | |
lower than Udc = 590 V. The switching
The design constraint of the absolute value of reference speed being lower than the
absolute value of real speed is very important. Otherwise, the motor would, once in
voltage control mode, brake to zero and wait until the capacitor is discharged by the
inverter losses to the level of Udc = 590 V. Thereafter, the motor accelerates to the
reference speed.
Figure 2.11 shows the experimental result of the implemented drive deceleration
tool using the 3 kW PMSM. Initially, the drive system is in speed control mode with
a reference speed of n* = 1500 rpm. At t = 0 s, the speed reference changes to
n* = 500 rpm and the switch on point (Udc = 600V) of the voltage control is reached
60 ms later. The implemented control scheme enables the inverter to keep its dc bus
voltage at the predetermined level. The braking of the drive is mainly caused by the
load torque. Reaching the reference speed, the system returns to the speed control
mode.
Using a brake-resistance within the dc link, the kinetic energy Wkin
Wkin =
1
J 12 22
2
(2.30)
43
n [rpm]
1500
1000
500
-0.5
0.5
1.5
2.5
t [s]
500
dc
[V ]
600
Speed
control
Speed
control
voltage control
400
-0.5
0.5
1.5
2.5
1.5
2.5
t [s]
20
i [A]
10
-10
-20
-0.5
0.5
t [s]
Figure 2.11: Voltage dip with load torque (speed dependent load). Top: Motor speed.
Middle: Voltage of the dc link. Bottom: Electromagnetic torque producing q-axis current.
2.7 Conclusions
Voltage dips and sags of short duration constitute a serious problem for electrical
drives in industry. Initiated by their under-voltage protection, in general, voltage
source PWM inverter drives shut down even at short interruptions of the power
supply. The resulting shut down of critical applications as a production line may
entail loss or damage of material. Especially multi-motor drives lose mutual
synchronization. Usually, time and additional workload to get a plant ready for
restart is then required. However, this and the resulting economic losses can be
avoided by using the proposed ride-through scheme.
Here, the time interval of the power interruption is bridged by kinetic buffering. A
fast reversal of the machine operation from motor to generator mode is commanded
at the event of a power failure. Energy is fed back from the rotating masses to the dc
44
Chapter 2
link circuit to maintain the dc link voltage at a predetermined level. This is possible
also in the presence of additional loads connected to the dc link. Due to advances in
semiconductor technology, modern electric drives can withstand the high peak
currents occurring when the power supply is restored after a short disturbance.
Keeping the capacitor well charged has the additional advantage of the control
electronics being powered over a longer time span, avoiding a time-consuming
restart of the drive.
In voltage control mode, the dc bus voltage is directly controlled by the
electromagnetic torque of the motor. The proposed voltage control scheme was
primarily developed for the PV-powered water pump system. The drive continues
operating even after a quite long power interruption of several seconds. The
temporary speed dip is generally tolerable, since the most frequent power
interruptions last only for a few milliseconds. Since drive control is never lost, the
voltage control scheme can be applied to multi-motor drives as well.
Finally, the proposed ride-through scheme at power interruptions is transformed into
a special drive deceleration tool for saving energy and simplifying the inverter setup.
Measured results are presented and evaluated to demonstrate the performance and
the system stability. Powerful digital signal processing is used to implement the
proposed regenerative braking schemes, expanding the time in which supply voltage
can be reapplied without the time-consuming dc-link capacitor recharging cycle.
3.1 Introduction
Classically, motor control was designed with analog components as they are easy to
design and can be implemented with relatively inexpensive components.
Nevertheless, there are several drawbacks with analog systems including aging,
temperature drift and reliability due to EMC problems. Regular adjustment is
required in those cases. Furthermore, any upgrade is difficult, as the design is
hardwired. Digital systems, on the other hand, offer improvement over analog
circuits. The mentioned drawbacks as drift and external influences are eliminated
since most functions are performed digitally. DSP technology allows both, a high
level of performance and cost reduction. Upgrades can easily be made in software.
DSPs have the capabilities to concurrently control a system and simultaneously
monitor it. A dynamic control algorithm adapts itself in real time to variations in
system behavior. Furthermore, implementation of complex control approaches is
possible and the drive system reliability can be improved.
For development purpose, a commercially available DSP based environment is used.
The heart of the controller board is a TMS320C31 digital signal processor. A slave
processor is employed to perform the digital input and output and generate the PWM
signals. The controller board can be directly programmed using
MATLAB/SIMULINK. A standard VS-PWM inverter with IGBTs is adapted to be
commanded by the DSP controller board.
This chapter presents the mutual interactions between control design and real-time
implementation. The DSP controller board, code generation, experiment
management and hardware interface including required measurements are explained.
Issues of phase voltage distortion/identification due to the inverter non-linearity are
discussed in detail. Finally, the used inverter and different PWM generation schemes
are evaluated. Optimizing (slimming down) a working control algorithm regarding
required computational effort, code optimization and implementation on a more
inexpensive hardware for the final product is something, to be considered in the final
stage of the development process.
46
Chapter 1
4 analog-to-digital converters
4 digital-to-analog converters
16 bit-selectable digital I/O lines
PWM generation on up to 6 channels
2 incremental encoder interfaces
A connector panel provides easy access to all input and output signals: Analog
signals via BNC connectors, all digital signals via Sub-D connectors. The singleboard hardware (appendix A) is integrated on a standard 16-bit PC/AT card slotted
straight into a PC using the ISA bus as a backplane.
The main DSP of the controller board can be directly programmed using
MATLAB/SIMULINK by The MathWorks. This application software is de-facto
standard in the control community, so no further explanation is given. SIMULINK is
a graphical user interface integrated in MATLAB for modeling and constructing
block diagrams via drag & drop operations. Its large block library is enhanced by
specific dSpace-blocks and own user-defined libraries simplifying automatic code
generation and experiment setup including initialization of the I/O subsystems and
PWM generation. Real-Time Workshop is the code generation extension provided
by The MathWorks. It generates C-code automatically from block diagrams and
state-flow systems. For flexibility, the user can introduce own C-code into the block
diagram by computation-time extensive S-functions or alternatively by special usercodes implying a change of the support software. The own C-code should be
preferred, whenever a part of the control algorithm contains many if-loops (e.g.
space vector modulation) or in very extensive programs, e.g. sensorless speed
control with Kalman filtering. Addressing the TI compiler and automatic download
to the DSP is done via the Real-Time Interface (RTI). For more information on
programming and implementation software, it is referred to the appropriate manuals.
The Total Development Environment (dSpace TDE) bundles a set of tools
supporting seamless transition from theory to simulation of new control algorithms
to real-time implementations. Software tools, such as CONTROLDESK, allow
parameter tuning/changing and data recording during the experiment in real-time
mode. CONTROLDESK is the comprehensive experimental environment software
providing management, control and automation of experiments. This user interface
47
enables access to every variable of the original block diagram. Figure 3.1 gives an
idea on how the experiment management looks like. Controller parameters can be
changed on-line (e.g.: speed reference) while the response is observed/recorded
simultaneously.
Figure 3.1: Screen plot during ac motor control experiment with the DS1102.
Usually, no code for the TMSM320P14 fixed-point DSP is generated, but the
appropriate I/O functions are automatically included by the slave-DSPs EPROM.
However, the support software has been changed in order to implement different
PWM strategies as well as variable PWM frequencies. This is extremely valuable
during the development of high-performance motor control using PWM outputs in
order to drive power switches. The modification of the support software has been
made in assembler code, since no C-compiler for the slave-DSP exists. Compared to
the original three-phase PWM generation performed in 73 s, the computation time
has been significantly reduced to 17 s. This new code is automatically included at
every compilation of the main program. The implemented modifications are
summarized in a separate manual.
48
Chapter 1
Power
supply
Current [A]
MPPTracking
Voltage [V]
Interface
DS1102-Processor Board
PWM 1
PWM 2
PWM 3
PWM 4
PWM 5
PWM 6
enable
C31
P14
ib
Udc
*
EXOR
ua
*
EXOR
ub
PWM
Inverter
EXOR
uc
ia
Incremental
encoder signal
AC
motor
Load
The PWM generation scheme implemented in the slave processor is based on phase
voltage reference values. PWM generation on up to 6 channels is possible. Both
subharmonic PWM generation and space vector modulation have been implemented.
The inverter used is a modified standard VS-PWM inverter with IGBTs. An
interface provides a galvanic isolation between controller board and inverter. The
PWM switching signals are fed directly from the slave processor to the inverter
using a high performance optical link, allowing to keep both inverter and drive
several meters from the PC with the controller board. Therefore, the signal
transmission is unaffected by EMC-problems. An enable signal, using one of the
digital I/O lines together with the same high performance optical link, supervises
both entire control system and inverter.
The TMSM320P14 slave DSP generates duty cycles with 40 ns edge resolution and
160 ns PWM period resolution. In this high precision mode, the P14 always sets the
output to a high level at the beginning of each PWM period, resulting in
49
(3.1)
2
1 + duty cycle phase a
(3.2)
2
PWM 1
EXOR
PWM 2
u*a
TPWM
TPWM
50
Chapter 1
3.4 Measurements
Phase current and dc bus voltage measurements, as described in the following
subsections, are required for most high-performance motion control systems. The
measurement of the motor speed/position may be eliminated by estimation
techniques. The rotor position is measured here for control purpose or for
comparison with sensorless drive schemes. An incremental encoder with 1024 lines
is used. This signal is directly fed to the encoder interface of the controller board.
More details on position measurement and the transformation to a speed signal are
given in chapter 4.
3.4.1
51
Figure 3.4: Principle of current measurement via closed-loop Hall effect current sensor.
In this work, the motor currents are measured by LEM-modules. The bandwidth of
the used magnet sensor devices is 150 kHz and the response time is smaller than
1 s. The secondary (canceling) current is is transformed into a voltage signal ui by
measuring the voltage drop across the sensing resistor RM (figure 3.4). This signal is
fed to an anti-aliasing filter connected to the inputs of the A/D converters.
Subsequently, the measured signals may be filtered by a digital low-pass filter.
However, rather than additionally filtering the current signals, observer-based
techniques can be used in order to reduce phase lags.
3.4.2
52
Chapter 1
oriented control and exact flux estimation, the inverter output voltages are required
to calculate desired state values. The output voltage can be measured or, by using
the information of the dc link voltage, estimated by means of the reference voltages.
However, the inverter output voltages are much distorted when compared to the
reference voltages and the use of the estimation is therefore not obvious.
3.5.1
A phase voltage measurement is difficult since the inverter output voltages are
composed of discrete high-voltage/high-frequency pulses. Therefore, a potential-free
measurement is required. A possible measurement setup and affiliated problems are
described in [Maes 01]. Beyond over-modulation, the frequency spectrum of the
output voltages generated by SVM consists of a fundamental frequency and many
higher harmonics around the PWM frequency. Only the fundamental voltage wave
contains useful information for the digital motion control. Thus, all high-frequency
components should be eliminated by a low-pass filter. Due to the low-pass filter, the
measured voltages suffer from phase delay and are not adequate for use in control
purposes [Choi 96].
Particularly at low-speed and light-load operation, where the undesired phase delay
is negligible, problems due to the accuracy of measurement may arise: The
fundamental phase voltage is very small in these operating points and only a fraction
of the measured pulses with a magnitude equal to the dc link voltage. Nowadays, a
measurement of the phase voltages is seldom used. This is mainly caused by the
complexity and extra costs of the additional measuring devices.
The development platform used here enables a simultaneous measurement of only
four signals, limited by the number of available analog-to-digital converters.
However, three A/D converters are already reserved for the measurement of dc bus
voltage and motor current in two phases. Thus, considering the given control setup,
the motor voltages must be calculated considering the inverters non-linearity.
In a voltage-source PWM inverter several causes distorting the output voltages can
be found. The reasons for this originate from the inherent characteristics of the
power switches such as voltage drop, voltage transition slope, turn on/off time and
delay of the control signals. However, this delay distortion is small when compared
to the dead-time effect [Bose 97] and is therefore usually disregarded.
3.5.2
For some subsequent described applications, such as sensorless speed control and
flux estimation, the exact inverter output voltages are required to calculate desired
state values. However, they are not measured due to the lack of sufficient analog-todigital converters, but calculated by means of the reference voltages with
53
10
ref
-U
10
[V ]
-5
Measurement
-10
Equation (3.11)
-15
-10
-8
-6
-4
-2
10
I [A]
1
u0 =
1
(ua + ub + uc )
3
(3.3)
54
Chapter 1
However, this homopolar system reflected in the line-to neutral voltages, must be
considered in the Park transformation:
u
u =
2 1
3
0
1 ua
2 u
b
3
u
c
2
1
2
3
2
(3.4)
In the case of an ideal inverter, the fundamental voltages at the motor terminals
assume the shape of the reference voltage. The reference voltages Uref are equal to
the duty ratios xref = (a*; b*; c*) calculated by the SVM (figure 1.14) multiplied by
the half dc bus voltage:
U ref =
1
U dc x ref ,
2
(3.5)
All together, using these reference voltages, the required alpha/beta voltages u, u
are calculated according to the block diagram in figure 3.6 considering both the nonlinearity of the inverter and the homopolar system of the SVM.
*
xref = (a ; b ; c )
SVM
u
1/2
Udc
i1
i2
i3
Eq.(3.10)
Equation
(3.13)
sign
Note that the calculation of the alpha/beta voltages described above is only valid
without strong over-modulation. The voltage spectrum in normal operation consists
approximately of one fundamental and many higher harmonics around the PWM
frequency. Over-modulation yields a voltage spectrum consisting of all uneven
harmonics. In fact, a current controller with a special anti-windup system (see
subsection 2.6.3) has been implemented, avoiding these operating points as well as
the low harmonics.
55
Udc
I,
n
enable
01
error
logic
enable
signal
1
reset
u>0
01
-1
-1
Figure 3.7: Safety-related monitoring & enable logic of the drive system.
The enable signal controls both the entire control system and the inverter
(figure 3.2). All gating pulses of the power switches are set to zero in case of an
error. However, special care has to be paid when a PMSM with high motor speed is
operated in flux weakening mode. A disabled inverter causes the return of the
unrestrained permanent magnet flux linkage and the dc bus voltage may reach
unacceptable (dangerous) high values, if no additional power dissipation is
connected in the dc link.
The offset of the current measurement is seldom exact equal to zero, which causes a
summation by the integrators of the controller even when the drive is disabled.
Therefore, all integrator values within the control scheme are multiplied by the
enable signal. This feature resets all registers at a restart and prevents an unwanted
overflow of integrator registers.
3.7 Conclusions
All subsequently described motor control algorithms are implemented using the
DSP-based development platform DS1102 from dSpace. In this chapter, an
overview of the given controller board and the hardware interface between DSP and
drive system has been presented. The main DSP of the controller board can be
directly programmed using MATLAB/SIMULINK. For flexibility, the user can
introduce own C-code into the block diagrams. Software tools allow parameter
tuning/changing and data recording during the experiment in real-time mode. The
laboratory test set-up consists of a host PC for the controller board, an IGBT
inverter, and ac motor drives with variable load, current/voltage sensors and an
56
Chapter 1
4.1 Introduction
Induction motors are relatively cheap and rugged machines. Much attention has been
given to induction motor control for starting, braking, speed reversal, speed change,
etc. When the drive requirements include fast dynamic response and accurate speed
or torque control, it is necessary to operate the motor in a closed loop mode with
feedback of the motor speed. Only a closed loop control of the motor meets the
requirements including fast dynamic response, accurate speed and torque control or
even a higher efficiency by means of flux optimization.
However, the speed sensor has several disadvantages from the viewpoint of drive
cost, reliability and signal noise immunity. Therefore, it is necessary to achieve
precise motor control without using position or speed sensors. This chapter deals
with the speed control of induction motor drives without a shaft sensor. The field
oriented control (FOC) technique is used, together with an estimation of the motor
speed. Both rotor field magnitude and position are estimated by summation of rotor
speed and slip frequency. The structure of the implemented sensorless control is
based on the Extended Kalman Filter theory (EKF).
There are many models of sensorless speed controllers described in literature
dealing with the Extended Kalman Filter theory. They are mostly based on the
models of [Bru 90] or [Vas 94]. Brunsbach estimates four states in a rotor-fixed
reference frame. The model of Vas, using the motor equations in a stator-fixed
reference frame, has shown a more stable behavior, but its disadvantage is its higher
order (5 states are observed). This is a drawback when the EKF algorithm has to be
implemented in real-time. However, the model is much simpler than the first one,
since it does not contain conversions between the stator and field coordinate system,
resulting in comparable execution times for both. This approach has become
commonplace. However, this model also causes some problems, especially at low
motor speed and speed reversals. The estimated states are time-dependent resulting
in an error driven nature of the observer even at steady state. Furthermore, the
estimated speed is lagging the real speed during transients, because the speed is
assumed to be constant during the sampling period.
58
Chapter 5
Here, a new model for speed estimation is proposed. This approach is shown to offer
a significant improvement of the drive performance. Along with the speed, also rotor
flux, flux position and acceleration of the drive are estimated. The speed estimation
does not lag the actual motor speed, both in steady state and during periods of
acceleration or braking.
The discussion starts by selecting a suitable motor model. Two marginal different
models are given; their advantages and drawbacks are briefly discussed. Then, the
design and implementation of the observer are explained in detail. A 1,5 kW
induction motor experimental system has been built to verify this approach. Results
are presented to demonstrate the performance of the system. The discussion ends by
evaluating the influence of motor parameter variations and designing a parameter
adaptation scheme in real-time to track these variations.
1
1
2
di
did
u
+ id = d + 1 iq (1 ) 1
dt
Rs
dt
diq
dt
di
dt
+ iq =
Rs
1 id (1 ) 1 i
+ i = id
= r +
Tel = p
uq
iq
2 i
L2
L1h
rd iq = p 1h i iq
Lr
Lr
(4.1)
(4.2)
(4.3)
(4.4)
(4.5)
59
The torque equation (4.5) clearly shows the required torque control property of
providing a torque proportional to the torque command current iq. The mechanical
equation of the drive is:
Tel Tload = J
d J d r
=
dt
p dt
(4.6)
The choice of input and output vector of the model has been determined by the
structure of the electrical equivalent circuit. The induction motor is supplied by a
voltage source PWM inverter. The voltages are not necessary measured, but can be
calculated by means of the reference voltages. The current has to be measured for
the implementation of the field-oriented control.
According to (4.4), the flux speed can be written as a function of the electrical
rotor speed, q-axis and magnetizing current. This property is neglected in many
speed observers assuming the speed of the rotor flux to be constant during the small
sample time interval Ts [Bru 91]. [Lut 93] uses this approximation even for the
discrete state space control of the induction motor. However, such an approximation
can be the origin of a poor estimation during transients. In fact, the speed of the rotor
flux, illustrated in figure 4.1, changes directly with and as fast as the q-axis current,
i.e. the electromagnetic torque.
250
[rad/s]
200
r
150
100
Load
step
slip
50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
t [s]
Furthermore, the derivative of the magnetizing current is often disregarded [Bru 91].
Neglecting a change of the magnetizing current in (4.1) may be an acceptable
approximation of the d/q-axis current equations, but yields no significant advantage
with regard to the computing effort. Thus, the substitutions in the model matrices
should be made by using (4.3)-(4.4). This eliminates both flux speed and flux
derivative in the stator voltage equations (4.1)-(4.2):
60
Chapter 5
iq
did id
=
+ r +
dt
1
2 i
iq (1 ) id i + u d
2
1 Rs
i
1 (1 )
id + r + q
=
2
2 i
1
diq
dt
iq
r +
1
2 i
iq
iq + (1 ) i + u d
2
1 Rs
i
uq
id (1 ) r i + q +
2 1 Rs
i
1 (1 )
iq r + q
=
2
2 i
1
uq
id (1 ) r i +
1 Rs
(4.7)
(4.8)
Assuming a very small sample time Ts, the transformation from continuous to the
discrete time state space causes a negligible error. This discretization error is usually
disregarded, but might be considered later as a part of the noise covariance matrix.
Consequently, the error is compensated by the filter feedback matrix.
i
1
(1 )
id + Ts r + q
id ,k +1 1 Ts
+
2
2 i
1
iq + Ts (1 ) i + Ts u d
2
1 Rs
i
1
(1 )
iq Ts r + q
iq ,k +1 1 Ts
+
2
2 i
1
uq
id Ts (1 ) r i + Ts
(4.10)
1 Rs
(4.9)
T
id + 1 s i
2
2
Ts
k +1 + Ts = + Ts r + Ts
(4.11)
iq
2 i
(4.12)
4.2.1
61
r ,k +1 r + Ts
p 2 L2h
p
i iq Ts Tload = r + model noise
J Lr
J
(4.13)
The known electromagnetic torque must not be used as part of the speed calculation
in (4.13) when also the load is disregarded. This would lead to a steady state speed
error since the Kalman algorithm assumes a zero mean value of the disturbances,
which is not correct, except at no-load, considering only the load torque as a
disturbance. Thus, both electromagnetic and load torque must be handled as system
disturbances while the speed is treated as a constant.
The selection of the first motor model in discrete time is completed by choosing dand q-axis current id, iq, rotor flux i, flux position and the electrical rotor speed r
as state variable xk and the fundamental voltage as input uk. The resulting system
model and its Kalman filter are referred in following discussions as Model 1:
x k +1 = A k x k + B k u k ;
U s
u k = s ;
U
k
id
iq
x k = i ;
r
k
iq
1
1 T 1 + 1
Ts r +
Ts
0
s
2i
2
1 2
iq
1 1
Ts r + i 1 Ts + Ts r 0
2
2
1
Ak =
Ts
Ts
0
1
0
2
2
0
0
0
1
Ts
Ts
0
0
2i
(4.14)
0
(4.15)
0
1
62
Chapter 5
1 = Ls/Rs
2 = Lr/Rr,
The input matrix Bk describes the weighted transformation from a stator-fixed to the
rotor flux reference frame.
cos ( ) sin ( )
sin ( ) cos ( )
T
Bk = s
0
0 ,
Ls
0
0
0
0
(4.16)
The resulting output vector yk consists of the estimated motor current in a statorfixed reference frame (/-system, indices: s). To avoid double calculations, the
sin/cos-terms of the flux angle should be calculated only once and used in both input
and output matrix.
I s
cos ( ) sin ( ) 0 0
y k = s = Ck x k =
I
sin ( ) cos ( ) 0 0
0
xk
0
(4.17)
A block diagram of the discrete motor model together with the feedback matrix of
the observer is shown in figure 4.2. The motor speed as well as all other states are
considered as both, state and parameter. The model matrices Bk and Ck depend on
the position of the rotor flux , the matrix Ak on q-axis current iq, rotor flux i and
rotor speed r.
Measurement:
US
US
+
+
Is
I s
yk
z-1
xk+1 xk
Bk
yk
+
-
Ck
+
+
k+1
xk+1
Ak
EKF
Figure 4.2: Block diagram of the discrete motor model and EKF.
4.2.2
63
the given states, also the acceleration due to the load torque is estimated. The model
presented in this subsection does not assume the velocity r but the load torque Tload
to be constant in a small time interval (sampling time Ts). This results in an
improved performance during transients of the motor speed. Iron and friction losses
of the induction motor are also part of the estimated load torque. Using the torque,
rather than the speed gives a better handle on the mechanical behavior, as in this
way acceleration is controlled, being the input to the speed variations.
The acceleration of the drive equals the difference between electromagnetic Tel and
load torque Tload related to the drive inertia J. The load torque is generally unknown,
but constant at steady state. It creates a disturbance of the speed control loop, which
is compensated by the controller. In steady state, the acceleration of the drive is zero
by definition. Thus, the differential equation of the acceleration due to the load
torque is:
d l d p
= Tload 0
dt
dt J
(4.18)
p
Tload
J
(4.19)
l ,k +1 l =
Now, the known electromagnetic torque is used as part of the speed calculation,
improving the accuracy of the speed:
r ,k +1 r + Ts
p 2 L2h
i iq Ts l
J Lr
(4.20)
In contrast to the remarks concerning the load torque in (4.13), the inaccuracy of
(4.18)-(4.20) has a zero mean value, being a precondition of the Kalman algorithm.
Only a variation of the load is handled as model inaccuracy. This inaccuracy is
neglected here, but will be taken into account afterwards at the evaluation of the
noise covariance matrix. In addition, erroneous electromagnetic torque calculation
and inertia identification are handled as model noise. However, the influence of both
an incorrect estimation of the electromagnetic torque due to electrical parameter
variations and an incorrect identification of the drive inertia are small compared to a
potential load variations. The discrete form of the second model is:
x k +1 = A k x k + B k u k ;
U s
u k = s ;
U
k
id
iq
i
xk =
r
l k
(4.21)
64
Chapter 5
iq
1 T 1 + 1
Ts r +
s
2i
1 2
iq
1 Ts 1 + 1
Ts r +
2
1 2
Ts
0
Ak =
p 2 L2h
Ts
i
0
J Lr
Ts
2i
0
0
Ts
2
1
r
Ts
Ts
1
Ts
cos ( ) sin ( )
sin ( ) cos ( )
0
0
T
Bk = s
0
Ls 0
0
0
0
0
I s
cos ( ) sin ( ) 0 0
y k = s = Ck x k =
I
sin ( ) cos ( ) 0 0
(4.22)
0
Ts
(4.23)
0 0
xk
0 0
(4.24)
This model has a disadvantage: its order is higher. This is a drawback when the EKF
algorithm has to be implemented in real-time. However, one major advantage of this
model is that it does not assume the speed to be constant during the sample time.
The involuntary lag of the speed signal is avoided by the additional estimation of the
load acceleration. In fact, the estimation of the acceleration is insignificantly lagging
at a continuous load torque variation. Nevertheless, the acceleration is, apart from an
initial change, nearly constant during both changing the speed reference and
applying load torque. This special drive property is caused by the current/torque
limitation within the speed control loop. Thus, the acceleration is almost constant
and can be estimated accurately.
The other advantage originates from the higher accuracy of the speed specification.
This accuracy is considered at the calculation of the noise covariance matrix. A
lower value indicates a more accurate estimation and accordingly results in a
smother speed signal.
Obviously, the performance of the system increases as the information of the known
electromagnetic torque is used. Only the load torque is handled as if it were an
unknown system disturbance, being true for many motor drives. If the load-speed
relation is known, this information can be used for further improvement of the
65
l ,k +1 l ,k + Ts
dT
p p 2 L2h
i iq l ,k load
J J Lr
d r
(4.25)
(4.26)
If the known load-speed relation is applied to the algorithm, also the system model
inaccuracy is lower. The noise covariance Q can be reduced, resulting in a very
smooth steady state speed signal and almost no lag during acceleration or braking
periods.
Both with and without applying the load-speed relation, the performance of the
estimator is only slightly affected by a precise knowledge of the inertia. If the inertia
J is set to infinite, the behavior of the algorithm is like the one neglecting the torque
command inputs and assuming the speed to be constant in a small time interval.
Naturally, the inertia must not be set to zero to guarantee a stable functioning. In all
other cases, a mismatch of the inertia is handled by the EKF as system noise. The
steady state estimation of the load torque becomes erroneous but the speed
estimation remains correct.
66
Chapter 5
ud
uq
Predictor
Pk+1|k
1/z
Pk|k-1
x k +1 k
Filter
h
x
1/z
x k k 1
Pk|k
Kk
Kk Yk
Y k = y
measured
xk k
y
The Kalman algorithm distinguishes between filter and predictor equations. The
predictor equations are responsible for projecting the state to obtain the a priori
estimation of the next time step. The filter equations, also called measurement
update, are responsible for the feedback to obtain an improved a posterior
estimate. The predicted value of the state vector xk+1|k is corrected by adding the
product of filter gain and the difference between estimated and measured output
vector yk to the state vector xk|k. In addition still the equation for the corrected
covariance matrix Pk|k is required.
))
(4.27)
Pk|k = Pk|k 1 K k
h
|x= x
x
Pk |k 1
(4.28)
x k |k = x k |k 1 + K k y k h x k |k 1 , k
k |k 1
The matrix Kk is the feedback matrix of the extended Kalman filter. This matrix
determines how the state vector xk|k is modified after the output of the model yk is
compared to the measured output of the system. The filter gain matrix is defined by:
T
K k = Pk |k 1
h
| x = x | 1
x
kk
T
h
|
P
x x = x | 1 k |k 1 x | x = x | 1 + R
kk
kk
(4.29)
x k +1|k = k + 1, k , x k |k 1 , u k
(4.30)
Pk +1|k =
T
| x = x Pk |k
|x = x + k Q k
x
x
k |k
k |k
(4.31)
67
(
h(x
( )
( )
k + 1, k , x k |k 1 , u k = A k x k |k x k |k + B k x k |k u k |k
k |k 1
, k = Ck x k |k 1 x k |k 1
(4.32)
(4.33)
In addition, the derivatives of system and output vector are required for the EKF
algorithm. The derivative of the system vector of Model 2 results in:
68
Chapter 5
iq
1 T 1 + 1
Ts r + 2
s
i
2
1 2
1 1
iq
id
Ts r + i 1 Ts + + i
2
2
2
1
s
0
=
2
x
p 2 L2h
T
i
0
s
J Lr
Ts
2 i
0
0
0 0 0 0
0 0 0 0
0 0 0 0 +
0 0 0 0
0 0 0 0
0 0 0 0
(4.34)
+ 0
2
1
iq
Ts
Ts
uq
Ts iq
2 2i 2
Ls
iq id 1
T
1
0 Ts
i s ud
r Ts id +
2
2 i
Ls
T
0
1 s
0
0
0
0
0
p 2 L2h
iq
Ts
J Lr
iq
Ts
2i 2
0
Ts
Ts
{4,2} and x {4,3} to zero. In this way, the influence of electromagnetic and
x
h cos ( ) sin ( )
=
x sin ( ) cos ( )
cos ( ) sin ( )
=
sin ( ) cos ( )
0 0
0 0
0 0
0 0
id sin ( ) iq cos ( ) 0
id cos( ) iq sin ( ) 0
i 0
i
0
69
(4.35)
The calculation of the estimated /-current is already executed by the output matrix
C of the system model and should be used in (4.35) to avoid double calculations.
The corresponding derivative for Model 1 is obtained by dropping the last column in
(4.35).
The remaining variables of the algorithm are the noise covariance matrices Q and R
and an initial matrix P0|0 representing the covariance of the known initial conditions.
They consist only of diagonal elements.
4.4.1
One critical step towards the implementation of the extended Kalman filter
algorithm is the search for the best covariance matrices. They have to be set-up
based on the stochastic properties of the corresponding noise. The noise covariance
R accounts for the measurement noise introduced by the current sensors and the
quantization errors of the A/D converters. Increasing R reflects a stronger
disturbance of the current. The noise is weighted less by the filter, causing a more
filtered current but also a slower transient performance of the system. The noise
covariance Q describes the system model inaccuracy, the errors of the parameters
and the noise introduced by the voltage estimation. Q has to be increased at stronger
noise levels driving the system, entailing a more heavily weighting of the measured
current and a faster transient performance. Thus, changing the covariance matrices R
and Q affects both the transient duration and the steady state operation of the filter.
An initial matrix P0|0 represents the matrix of the covariance in knowledge of the
initial conditions. Varying P0|0 affects neither the transient performance nor the
steady state conditions of the system and can be chosen at random.
The covariance matrices R, Q and P0|0 are assumed to be diagonal due to the lack of
sufficient statistical information to evaluate their off-diagonal terms. Furthermore,
the diagonal characteristic holds the possibility of saving a lot of computing time as
shown in the next subsection.
In general, the entries of the covariance matrices are unknown and cannot be
calculated. They are often set to the unity matrix. In order to achieve the optimal
70
Chapter 5
filter performance, the filter parameters R and Q can be obtained by tuning based on
experimental investigations. This describes an iterative process of searching the best
values. It is almost impossible to find a plausible evaluation of these parameters in
literature with regard to the sensorless control of motor drives.
However, it is preferable to have a rational basis for choosing the required
parameters. In either case, whether or not a superior filter performance can be
obtained by an additional tuning process, an initial guess of the values is welcome.
As shown, the value of the different parameters differs a lot. Furthermore, the filter
performance may change dramatically by varying only one value. Without any
previous knowledge and considering the high dimension of the matrices, tuning is
very arduously or can even lead to an unstable behavior of the observer. For
instance, changing the sample time requires a new tuning process. A design equation
has the additional advantage of being independent of the given installation, and it
can easily be assigned to other drive installations without an expert tuning the
parameters.
The measurement covariance R can be measured easily in advanced. Measuring is
generally possible because the current measurement is needed anyway while
operating the filter. Some off-line sample measurements are taken in order to
determine the variance of the measurement error. This is done by applying a
constant line-to-line voltage across two phases containing a current sensor and
measuring the resulting dc current. It must be noted, that the measured current
should not be supplementary filtered, apart from an anti-aliasing filter of course. The
noise on the raw measurements will possibly be non-linearly transformed resulting
in second order terms, which may be significant. The Kalman approach handles
white and uncorrelated measurement noise and produces the minimum variance
estimate. Therefore, this is already an optimal filter. A current measurement
respecting the given installation yields a measurement noise covariance matrix,
being almost proportional to the dc bus voltage within a voltage range
300V < Udc < 600V:
1,8 10 6
0 2
0
A 2 1,5 10 4
A
R =
U
dc
6
V 0
1,5 10 4
1,8 10
0
(4.36)
Ts
voltage inaccuracy
Ls
71
(4.37)
The voltage can be either measured or calculated by means of the reference voltages
being the output of the entire control loop. Here, the phase voltages are calculated.
Therefore, the accuracy is only affected by the non-linearity of the converter. This
non-linearity has its origin in the delayed reaction of the switches at turn-on and off, also called dead-time effect [Bose 97]. Therefore, the accuracy of the voltage
calculation, described by an error voltage U, is simplified dependent on the deadtime dead, the dc bus voltage Udc and the PWM-frequency fPWM of the inverter:
U dead f PWM U dc
(4.38)
(4.39)
Q(2,2) = Q(1,1)
(4.40)
For the given drive, using a sample time Ts = 200 s, a dead-time dead = 2 s and a
PWM-frequency fPWM = 10 kHz, the covariance amounts to:
Q(1,1) = Q(2,2) = 0,0018 A 2
(4.41)
k =0
2
2
kT
(4.42)
72
Chapter 5
i T
Q(3,3) < var max s
2
2
imax
Ts2
=
2,6 10 5 A 2
2
3
(4.43)
Assuming a maximum acceleration of max = 1000 s-2 and a sample time Ts = 200 s,
the variance of the flux position is estimated by:
( k +1)Ts
( k +1)Ts
Q(5,5) = var Ts k (t ) dt < var Ts max k Ts max t dt (4.44)
k =0
kTs
kTs
T 2
Q(5,5) < var max s
2
2
max
Ts4
=
3,3 10 11
48
(4.45)
For Model 2, the inaccuracy of the motor speed calculation is also caused by the
discretization error dependent on the maximum acceleration and its time variation.
Based on common bandwidth of torque control loops, a maximum variation time
constant torque = 1 ms is chosen.
( k +1)Ts
QModel 2 (4,4) = var Ts k (t ) dt
k =0
kTs
T 2
QModel 2 (4,4) < var max s
2 torque
2
max
Ts4
1
=
0,0033 2
2
48
s
torque
(4.46)
(4.47)
These values are very small resulting in smooth signal shapes. They could be set to
the maximum in order to achieve maximum dynamic performance of the drive.
For Model 1, the inaccuracy of the motor speed calculation is higher due to the
simplified specification and can be found from the maximum inertia related torque
variation:
QModel 1 (4,4) =< var(2 Ts max )
QModel 1 (4,4) <
2
Ts2 max
1
0,013 2
3
s
(4.48)
(4.49)
73
the number of pole pairs p and a maximum torque variation time constant
load = 2 ms, the top-limit of the acceleration inaccuracy and of the process variance
is:
Tload ,k 1
T
p Tmax Ts
J
J torque
max
1 p Tmax Ts
Q(6,6) <
12 J torque
(4.50)
1482 1
s4
(4.51)
The dynamic and smoothness of both speed and acceleration estimation is tuned by
Q(6,6). This parameter should be smaller, if the load torque is known very well or a
smooth speed signal is more important than the loop dynamic. A high value
increases the dynamic performance, but also the noise of the estimated signals. With
respect to the given drive setup, Q(6,6) is set to 5% of the value given in (4.51) in
order to obtain a good compromise between dynamic performance and smooth
torque command response. All other coefficients of the system covariance matrix are
set to the given values.
It should be noted, that the calculation of all process covariance matrices is
proportional to the square of the sample time. Thus, the given constants should be
adapted accordingly, if a different sample time is chosen. However, it is a major
advantage of the proposed model, that estimation accuracy and stability of the entire
control system are much less sensitive to tuning the covariance matrices compared
to other models.
4.4.2
Computing requirements
The speed estimation and the entire control of the induction motor are implemented
on a TMS320C31 DSP with 128 K 32-bit RAM. The implemented algorithm
estimates five states for Model 1 and six for Model 2. The computing demand grows
almost with the third power of the state dimension. Furthermore, they contain
conversions between stator and field coordinate system and a computation time
intensive matrix inversion. The algorithm can be implemented with relatively few
instructions using matrix calculation. However, without any modification, the
resulting algorithm leads to a program that is not suitable for real-time
implementation, since it is very complex especially due to the matrix inversion. The
execution time would be higher than 400 s, respectively 700 s, using the given
DSP. In consequence, also the bandwidth of the current/torque controller would be
very small. Furthermore, the performance of the EKF decreases as the sample time
increases.
74
Chapter 5
The turnaround time of the final control system, using Model 2, amounts to 187 s.
Only a few extra calculations are necessary compared to the speed observer based on
Model 1 requiring a turnaround time of 167 s. The used sample time is set to
Ts =220 s. However, the execution time is not that meaningful. The DSP power is
simultaneously used for monitoring and recording the experimental data. Due to
developing reasons of the installation, the remaining field-oriented control is not
optimized regarding the computation requirement: e.g., the implemented FOC
contains three different speed controllers for performance comparison. An overview
of the computing requirement considering the different approaches is summarized in
table 4.1.
Table 4.1: Computation requirement of different EKF approaches.
Model 1
Matrix calc.
Model 2
Matrix calc.
Model 1
optimized
Model 2
optimized
Number of
Summations [ ]
Number of
Multiplications [ ]
Turnaround time of
the entire control [s]
546
662
>400
881
1026
>700
207
254
167
255
320
187
Keeping the size of the program limited is achieved by optimizing the model with
hand calculations and exploiting matrix symmetry. The covariance matrices Q, R
and initial matrix P0|0 are set to be symmetrically. In consequence, also the matrix
Pk|k-1 becomes symmetrical which can be exploited avoiding double calculations and
higher memory demand. The implemented EKF covers no superfluous
multiplications by zero. Several matrix calculations of the EKF algorithms are the
same and can be used in different equations, e.g. in (4.28)-(4.29). Furthermore, the
sine and cosine of the flux angle is calculated only once and used in the EKF as well
as in the Clarke-transformations [Bose 97] of currents and voltages.
4.4.3
Model comparison
Two models are closely examined. The first one is based on an approach, that has
become commonplace in almost every speed observer. They do not recognize the
actual torque command inputs to the system and assume the velocity to be
constant in a small time interval. In effect, such techniques treat the known torque
command input as if it were an unknown disturbance torque. Thus, they generally
lag the actual motor speed during periods of acceleration or braking.
The second motor model uses the additional information on the electromagnetic
motor torque. Additionally, the acceleration due to the load torque is estimated. This
model does not assume the velocity but the load torque to be constant in a small time
interval (sampling time Ts). The inaccuracy of the speed calculation is transferred to
the load, which is usually unknown anyway. This results in an improved
75
performance during transients (figure 4.4) presenting the response to a step of the
speed reference and to a load step. Figure 4.5 shows some important details of
figure 4.4.
1500
n [rpm]
2
1000
500
0
-0.1
-0.05
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.2
0.25
0.3
0.35
0.4
t [s]
T [Nm]
15
Tel
10
5
0
-5
-0.1
Tload
-0.05
0.05
0.1
0.15
t [s]
Figure 4.4: Step of the speed reference and response to a load step. Top: Real and estimated speed with
and without load torque estimation. Bottom: Estimation of electromagnetic and load torque.
n [rpm]
600
With l
estimation
400
Without l
estimation
200
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
0.045
0.05
t [s]
n [rpm]
1050
1000
950
900
0.05
0.1
0.15
0.2
0.25
0.3
t [s]
76
Chapter 5
From now on, only the second model is considered. Model 2 offers a large
improvement of the performance. The price to be paid is only marginally extra
computing effort.
4.4.4
Udc
sin()
Voltage
calculation
ia
ib
i
i
32
u cos()
u
id
iq
i
i
EKF
sin()
cos()
id,error
iq,error
Speed
id*
&
iq*
Flux
ud*
Control decoupling u *
q
Speed
reference
ud*
uq*
current
control
ud*
uq*
Udc
2
Udc
*
SVM
ua
Inverter
*
ub
*
uc
u*
SVM
u*
PWM
generation
inverse
Park Trans.
Load
AC
motor
Figure 4.6: Velocity observer integrated into a field-oriented speed control loop.
The presented observer achieves the objective of eliminating lag of the estimated
motor speed by additional load estimation. It should be noted, that the velocity
estimation described here can easily be extended to allow for further improvement
of the entire drive performance especially at load torque variation by adding
acceleration feedback. The information of load acceleration can be used directly by
compensating for the load torque. Rejecting load disturbances improves the dynamic
stiffness of the drive. Therefore, this feedback causes the disturbance to perceive a
more robust system less sensitive to disturbances.
77
n [rpm]
1000
500
0
-500
-1000
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.3
0.35
0.4
0.45
0.5
t [s]
n [rpm]
50
25
0
-25
-50
n = nest - nm
0
0.05
0.1
0.15
0.2
0.25
t [s]
Figure 4.7: Speed reversal test. Top: Reference, measured and estimated speed.
Bottom: Difference between estimation nest and measurement nm.
The current controller, using also the estimated values of d- and q-axis current, has a
bandwidth of 847 Hz. Figure 4.8 presents the response of the induction motor to a
load step at a motor speed of 1500 rpm. The applied load amounts to 65% of the
rated value.
The behavior at low motor speed is shown in figure 4.9. First, the response to a
square wave shaped speed reference is given. With a sinusoidal speed reference,
there is almost no difference between estimation, measurement and reference.
78
Chapter 5
1600
nm
n [rpm]
1550
1500
nest
1450
1400
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.75
t [s]
15
[Nm]
T
2
0
10
load
i [A]
0.25
0.5
0.75
0.25
t [s]
0.5
t [s]
n [rpm]
50
25
nm
0
-25
-50
nest
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.06
0.07
0.08
0.09
0.1
t [s]
n [rpm]
25
0
-25
0
0.01
n [rpm]
0.02
0.03
0.04
0.05
t [s]
50
nest
25
0
-25
-50
nm
nref
0
0.002
0.01
0.02
t [s]
Figure 4.9: Behavior at low motor speed. Top: Square wave speed reference (20 Hz).
Middle: Sinusoidal speed reference (20 Hz). Bottom: Sinusoidal speed reference (100 Hz).
Even at low motor speed and standstill the proposed control scheme is able to
manage the load torque (figure 4.10-4.11). Load is applied using a dc motor
operating in torque control mode. The arising torque ripple components are typical
of a thyristor converter and are returned to the signals of speed and load torque
estimation respectively.
79
8
6
dc
[A]
4
2
0
-2
0.5
1.5
2.5
2.5
t [s]
60
40
nest
n [rpm]
20
0
-20
nm
-40
-60
0.5
1.5
t [s]
Figure 4.10: Response to a load step at standstill. Top: Armature current of the load machine.
Bottom: Measured speed nm and estimated speed nest.
[A]
6
2
dc
4
0
-2
0.5
1.5
2.5
3.5
2.5
3.5
2.5
3.5
t [s]
n [rpm]
120
60
0
0.5
1.5
10
5
load
[Nm]
t [s]
15
0
0
0.5
1.5
t [s]
Figure 4.11: Response to a load step at low motor speed (nref = 60 rpm). Top: Armature current of the
load machine. Middle: Measured and estimated speed. Bottom: Estimation of the load torque.
Also at high motor speed and flux weakening, a good performance of the EKF can
be obtained. Figure 4.12 demonstrates the behavior of the speed and flux estimation
in this speed range. The flux is inversely proportional to the motor speed. In
consequence, the applied fundamental motor voltage remains nearly constant. Only
a small transient is needed to adapt the required d- and q-axis current.
This feature of the EKF makes the proposed system also suitable for applications
with flux optimization increasing the drive efficiency. However, a minimum flux is
required to guarantee a stable operation of the EKF. The minimum flux for the given
induction motor drive amounts to approximately 10% of the rated value.
80
Chapter 5
2000
300
nest
nm
1500
nref
1000
n [rpm]
2500
|U + U | [V]
3000
500
0
0.05
0. 1
0.15
0.2
250
200
150
50
0
0.25
Flux
weakening
100
0.05
0.1
t [s]
3.5
0.25
200
u [V]
2.5
2
100
0
i [A]
0.2
300
Flux
weakening
1.5
1
0.15
t [s]
-100
-200
0.05
0.1
0.15
t [s]
0.2
0.25
-300
-300 -200 -100
u [V]
Figure 4.12: Behavior at high motor speed and flux weakening. Top: Speed response
and applied voltage amplitude. Bottom: Magnetizing current and applied voltage.
= id i = 2
di
dt
=0
(4.52)
81
derived from (4.1)-(4.2), a resistance detuning yields an error of the d-axis current
estimation. Thus, the error is positive, if the resistance is underrated. The same
applies for a positive detuned inductance.
The error value introduced by an inaccurate estimation of the stator resistance
decreases as the supply frequency increases. At high motor speed, stator resistance
detuning causes a negligible speed estimation error [Wang 99]. Therefore, knowing
that there is a finite precision in measurements of stator voltages and currents,
rational stator resistance adaptation is possible only at low motor speed. Due to
similar considerations, stator inductance detuning affects the speed estimation only
at higher motor speed. Practically, the error is used as a feedback signal to adapt
accurately stator resistance at low supply frequencies (|| < 5 Hz) and inductance at
high motor speed and supply frequencies (|| 5 Hz) respectively. Beyond these
boundaries, they are kept constant.
Figure 4.13 presents the experimental result of the stator resistance adaptation at
standstill. Starting with an initial error of 60 %, the adaptation is enabled at t = 0,5 s.
After a short period, magnetizing current matches the d-axis current, confirming the
well-tuned resistance value.
3.8
i [A]
3.6
id
3.4
3.2
3
2.8
0.5
1.5
2.5
3.5
4.5
3.5
4.5
t [s]
6
R [ ]
Real Rs
Estimated Rs
3
2
0.5
1.5
2.5
t [s]
Figure 4.13: Stator resistance adaptation at low speed (n = 0 rpm). Top: Magnetizing
and d-axis current. Bottom: Estimated and real stator resistance value.
Figure 4.14 illustrates the stator inductance adaptation at n = 1000 rpm. A starting
error of the inductance (44%) has been introduced resulting in a poor estimation of
the motor speed. The parameter adaptation scheme, switched on at t = 0,2 s, detects
the steady-state deviation of magnetizing and d-axis current and tunes the
inductance.
It should be noted, that erroneous inductance estimation, directly reflected in both
incorrect torque-current mapping and load estimation, is not compensated by the
speed controller. The implemented speed controller with load torque rejection
consists of a proportional gain and contains no integral-acting part. Thus, parameter
82
Chapter 5
mismatch yields a steady-state error of the speed control loop. However, this steadystate error can be used as an adaptation watchdog or, alternatively to the proposed
algorithm, as a parameter-correcting feedback signal.
3.5
3
0
0.25
0.5
0.75
t [s]
1.25
450
400
350
0.25
1050
1025
975
0.75
t [s]
1.25
0.25
0.5
0.75
t [s]
0.25
0.5
1.25
0.75
t [s]
1.25
1.5
Parameter adaptation
switched ON
250
200
0
0.25
0.5
0.25
0.5
1002
nref
0
id
0
300
1.5
nm
nest
1000
0.5
2.5
350
Parameter adaptation
switched ON
500
1.5
L [mH]
L [mH]
550
n [rpm]
i [A]
id
n [rpm]
i [A]
1.5
0.75
t [s]
1.25
1.5
1.25
1.5
1000
998
996
994
nm
0.75
t [s]
Figure 4.14: Adaptation of the stator inductance. Top: Magnetizing and d-axis current. Middle: Stator
inductance. Bottom: Estimated, real and reference speed. Left: Initial inductance value 44 % overrated.
Right: Initial inductance value 44 % underrated.
Figure 4.15 shows the experimental result of the real-time inductance adaptation at
variable rotor flux. As can be seen, the inductance is clearly dependent on the
saturation level of the machine. Actual and estimated speed are in excellent
agreement, confirming the well-tuned inductance value. Applying load torque, the
obtained results are equivalent. Usually, a flux controller keeps the rotor flux
constant. Rotor flux variations due to both load and electromagnetic torque changes
are small and do not significantly affect the observer performance. However,
according to (4.52), the adaptation must be disabled during fast rotor flux transients
at e.g. initial start, flux weakening and flux optimization.
With respect to the given drive setup, the influence of the leakage coefficient on
the observer performance is very low and hardly measurably. Furthermore, a
mismatch is partially compensated by the inductance adaptation. Therefore, an
adaptation of is not implemented.
No major problem exists in determining the stator frequency . To the contrary, the
estimation error of the rotor frequency slip, directly reflected in the accuracy of the
rotor speed estimation r, depends on the rotor time constant. This error increases
proportionally to the q-axis current and load respectively. The rotor time constant
varies in a fairly wide range during operation. Variations of the rotor inductance are
caused by changes of magnetization. Furthermore, the rotor time constant changes
with the machine temperature. Assuming an equivalent influence of the saturation
level on both the stator and rotor inductance, the stator inductance adaptation
83
450
400
L [mH]
350
i [A]
A promising solution of tuning the rotor time constant in real-time is based on the
evaluation of rotor slot harmonics [Jia 97]. This method permits high speedaccuracy in steady state and allows even position control. However, detection of
rotor slot harmonics should not be used as a stand-alone solution for speed
estimation, since the dynamic performance of such systems is very poor [Ish 82],
[Kre 92]. To the contrary, the dynamic performance of the proposed observer is
excellent. Together with the detection of rotor slot harmonics, compensating slow
temperature variations, a system with both high dynamic performance and high
steady-state accuracy is obtained.
2
1
10
20
30
40
300
250
50
10
20
t [s]
1010
40
50
450
nm
400
L [mH]
1005
1000
350
n [rpm]
30
t [s]
995
990
nest
0
10
20
30
t [s]
40
50
300
250
i [A]
Figure 4.15: Adaptation of the stator inductance. Top: Magnetizing current, d-axis current and stator
inductance. Bottom: Estimated, real, reference speed and flux dependence of the stator inductance.
4.7 Conclusions
This chapter presents the design and the implementation of a field-oriented highperformance motor drive with speed, flux and torque estimation. The speed
controlled induction motor drive requires no shaft sensor measuring speed or
position. The price to be paid is a more extensive and complicated control algorithm.
However, no additional measurements are required. The structure of the
implemented sensorless control is based on the extended Kalman filter theory.
Results of the dynamic and steady-state behavior of a sensorless speed control of an
induction motor are given. After the correct system model is chosen for the
Extended Kalman Filter, the results are satisfactory. Both at very low and at high
84
Chapter 5
motor speed with flux weakening, the proposed control scheme is working very
well. The described control system is a solution without mechanical sensors for a
wide range of applications where good steady state and dynamic properties are
required. Keeping the size of the program reasonable and still reaching a very good
performance is achieved by optimizing the model by hand calculations and
exploiting matrix symmetry.
The performance of the system increases as the information of the known
electromagnetic torque is used. Only the load torque is handled as if it were an
unknown system disturbance. If the load-speed relation is known, this information
can be used for a further improvement of the performance. The speed estimation
does not lag the actual motor speed, both in steady state and during periods of
acceleration or braking. Steady-state errors are used for parameter adaptation.
5.1 Introduction
With the introduction of permanent magnets with a high flux density as well as a
high coercivity in the late eighties, synchronous motors with permanent magnets
became an attractive alternative for applications in high performance variable speed
drives. Significant advantages arise from the simplification in construction, the
reduction in losses and the improvement in efficiency. One of the most active areas
of control development during recent years involving these motor types has been the
evolution of new techniques for eliminating the position and speed sensor.
Elimination of the shaft-mounted sensor is required in many applications since this
device is often one of the most expensive and fragile components in the entire drive
system.
The approaches to sensorless drives vary depending on the rotor flux distribution. A
motor with a trapezoidal rotor flux distribution (BLDM, brushless dc motor)
provides an attractive candidate: Two out of three stator windings are excited at the
same time and the unexcited winding can be used as a sensor. The control scheme as
well as the position detection is relatively simple. The rotor speed and position can
be determined by the electromagnetic field induced in the unexcited winding
[Erd 84], [Mat 90]. This is usually done either by a zero crossing approach of the
back-EMF or by a phase-locked loop technique to lock on to the back-EMF
waveform in the unexcited winding. It is enough to detect the rotor position every
60 to obtain a proper switching sequence.
On the contrary, the permanent magnet synchronous motor (PMSM), having a
sinusoidal flux distribution, excites all three windings at the same time. Both the
control algorithm and the speed/position estimation become more complicated. The
information on the rotor position is required continuously. However, the PMSM is
applicable for fine torque control where a very low level of torque pulsations is
required. Several schemes for position sensorless operation of PMSM have been
reported in literature and are reviewed e.g. in [Raj 94]. The position detection
methods are mainly based on Kalman filtering or Model Reference Adaptive
Systems using the motor parameters and measurements of motor currents and
voltages.
86
Chapter 6
The drive system studied in this chapter is a sensorless control of a PMSM based on
the extended Kalman filter theory using only the measurements of motor current and
dc bus voltage for the estimation of speed and rotor position. On top of the speed,
also the acceleration of the drive is estimated offering a significant improvement of
the drive performance. The applied approach is mainly a transfer of the earlier
described sensorless control of the induction motor to the motor equations of the
PMSM. Therefore, this chapter explains only differences in detail while many still
valid statements of previous chapters are not repeated. Theoretical analyses based on
the physical viewpoint are presented and the associated experimental results are
shown. This chapter also describes the influence on the control design reflected by
the feedback of the estimated values. A torque that at average differs from zero, is
only produced if the excitation is precisely synchronized with the rotor speed and
instantaneous position. The controller has to ensure that the motor never experiences
loss of synchronization. Due to rotor asymmetry, the PMSM is also suitable for
position control. A 3 kW, 4 kW and 45 kW PMSM have been used to verify this
approach. The discussion ends by evaluating a parameter adaptation scheme in realtime to track motor parameter variations.
did
r Lq iq
dt
diq
dt
+ r Ld id + r Md
(5.1)
(5.2)
) )
Tel = p iq Md Lq Ld id
(5.3)
In contrary to the induction machine, the flux angle rotates synchronously with the
rotor speed. With the same simplifications as introduced in the induction motor
study, the mechanical equation is:
Tel Tload = J
87
d J d r
=
dt
p dt
(5.4)
d
= r
dt
(5.5)
R
u
id ,k +1 1 Ts s id + Ts r
iq + Ts d
L
L
Ld
d
d
(5.6)
uq
R
L
iq ,k +1 1 Ts s iq Ts r d id Ts Md + Ts
Lq
Lq
Lq
Lq
(5.7)
r ,k +1 r + Ts
) )
p2
p
iq Md Lq Ld id Ts Tload
J
J
k +1 + Ts r
(5.8)
(5.9)
From the control viewpoint, the PMSM has four electrical parameters: stator
resistance Rs, d- and q-axis inductance Ld and Lq, and permanent magnet flux linkage
Md. The inductances are considered to be constant, which is verified by
measurements [Van 98] and numerical calculations of the given PMSM [Pah 98].
The influence of parameter variations is compensated by real-time adaptation of the
flux linkage Md.
The known electromagnetic torque is used as part of the speed calculation, vastly
increasing the accuracy of the speed specification and the dynamics of the drive.
However, this approach requires information on the load, since the Kalman
algorithm assumes a zero mean value of the disturbances. As mentioned in the
previous chapter, an additional estimation of the load torque increases the observer
performance as well as the performance of the speed control loop. The price to be
paid is a minor extra computing time. Therefore, the acceleration due to the load
torque is estimated additionally:
d l d p
= Tload 0
dt
dt J
(5.10)
p
Tload
J
(5.11)
l ,k +1 l =
88
Chapter 6
The dynamic model for the PMSM, choosing d- and q-axis current id, iq, the
electrical rotor speed r, rotor position and the acceleration l as state variable xk
and the fundamental voltage as input uk, is described by following equations. The
output vector yk consists of the estimated motor current in a stator-fixed reference
frame.
x k +1 = A k x k + B k u k ;
T
1 Rs s
Ld
Ld
r Ts
Lq
Ak =
U s
u k = s ;
U
k
r Ts
id
iq
x k = r ;
l k
Lq
Ld
T
1 Rs s
Lq
Ts
(5.12)
Md
Lq
Ts p 2
Md Lq Ld id
J
0
Ts
1
0
) )
Ts
Ts
cos ( )
sin ( )
L
L
d
d
Ts
Ts
sin ( )
cos ( )
Lq
B k = Lq
0
0
0
0
0
0
Ts
0
1
(5.13)
(5.14)
At each time step, using the previously predicted position and current information,
the current is estimated in two stages to correct the predicted states by the Kalman
feedback matrix.
I s
cos ( ) sin ( ) 0 0
y k = s = C k x k =
I
sin ( ) cos ( ) 0 0
0
x
0 k
(5.15)
The model matrices Bk and Ck depend on the position of the rotor , the matrix Ak
on d-axis current id, flux linkage Md and rotor speed r. The block diagram of the
discrete motor model together with the feedback matrix of the observer is equal to
the one shown in chapter 5. In speed control mode, the flux angle is limited to
89
| | < . In contrary to the induction motor drive, the studied PMSM is also suitable
for position control since the rotor asymmetry can be exploited. Therefore, the
overflow protection has to be disregarded in position control mode. However, a loss
of exact position information is not admissible in any case.
sin ( ) cos ( ) 0 i
(A k x k + B k u k )
k
=
x =
xk
T
L
Ts p 2
1 R s s
Lq Ld iq
r Ts d
Ld
Lq
J
L
T
Ts p 2
r Ts q
Md Lq Ld id
1 Rs s
Ld
Lq
J
L
T
Ts q iq s (Ld id + Md )
1
Lq
Ld
T
T
s uq
s ud
0
Lq
Ld
Ts
0
0
(5.16)
) )
0
0
Ts
1
0
(5.17)
The noise covariance matrices Q and R and an initial matrix P0|0 are evaluated
corresponding to the remarks on the induction motor drive. The matrices R, Q and
P0|0 are diagonal due to the lack of sufficient statistical information to evaluate their
off-diagonal terms. Furthermore, their diagonal nature saves a lot of computing time.
P0|0 affects neither the transient performance nor the steady state conditions of the
system and can be chosen at random. An off-line current measurement, referring to
the 3 kW PMSM drive installation, yields a measurement noise covariance matrix,
which is almost proportional to the dc bus voltage within a voltage range
300V < Udc < 600V:
90
Chapter 6
1,15 0
A 2 1,7 0 3 2
U dc 10 5
10 A
R =
V 0 1,7
0 1,15
(5.18)
The measurement of the noise largely exhibits independence of the motor current.
The values a larger compared to the measurement noise of the induction machine
supplied by the same inverter. This is mainly due to the smaller inductances of the
PMSM smoothing the PWM pulses. For an external field (armature reaction) the
magnets behave as air, introducing a large reluctance and thus a low main
inductance. The coefficients of the system covariance matrix are calculated
according to subsection 4.4.1 taking a sample time Ts = 200 s and the parameter of
the 3 kW PMSM into account:
1 T u
Q(1,1) = s
3 Ld
1 T u
Q(2,2) = s
3 Lq
= 0,021 A 2
(5.19)
= 0,0059 A 2
1
s2
1
1
= k 43,2 4 , with: k 1
6
s
s
(5.20)
(5.21)
(5.22)
(5.23)
The transient performance of the observer is tuned by the factor k in (5.23). All other
coefficients of the system covariance matrix are set to the given values. A high
tuning factor k increases the dynamic performance, but also the noise of the
estimated signals.
Within the implemented speed control, the information on load acceleration is used
as input of the speed controller directly compensating the load torque (figure 5.1).
Rejecting load disturbances inproves the dynamic stiffness of the drive and is
superior compared to common PI controller [Lor 99]. However, a rough torque
command results in increased torque ripples and motor heating by current
harmonics. In order to obtain a good compromise between dynamic performance
and a smooth torque command response and with respect to the given installation,
the tuning factor k is set to 10%. Figure 5.1 shows the structure of the implemented
position and speed controller with load torque rejection. The calculated reference
torque Tel* is mapped into reference commands for d- and q-axis current. The current
commands id* and iq* are extracted according the constraint of maximum torque-perampere operation, being nearly equivalent to maximum drive efficiency [Bose 97].
91
Tel*
1 z 1
Ts / K pn
Kn
Control mode
Position
controller
Position
reference
|Tel|<Tmax
Proportional
gain
Kpp
J
Tload = l
p
Figure 5.1: Position and speed controller with load torque rejection.
Figure 5.2 shows the block diagram of the entire control system with the proposed
observer integrated in the digital motion control loop. The inputs of the control
system are measured motor current in two phases and the dc bus voltage. The
voltages required as input for the EKF are obtained from the reference voltages,
available at the output of the system. Due to the non-linearity of the inverter, a phase
voltage calculation block is added compensating for this non-linearity. The
homopolar component of the phase voltages arising due to the SVM [Leo 85] is also
considered within this block. The measurements should not be additionally filtered
since the Kalman filter handles with white and uncorrelated measurement noise and
produces the minimum variance estimate. Therefore, this is already an optimal filter.
Generally, the estimated states are used as feedback signals of the controller,
because they are less disturbed compared to measured values. Furthermore, the
smoothness of the state signals can be tuned by the system and measurement
covariance matrices.
ib
ia
ia
ib
32
Udc
A/D
AC voltage
calculation
sin()
i cos()
i
id
iq
u
u
EKF &
model
l
r
sin()
cos()
id,error
iq,error
id*
Speed iq*
Control
decoupling
Speed
reference
ud*
uq*
ud*
uq*
current
control
ud*
uq*
u*
SVM
u*
inverse
Park Trans.
ua
*
ub
*
uc
PWM
generation
2
Udc
Figure 5.2: Velocity observer integrated into the digital motion control loop.
The entire speed control system consists of a speed and two current controllers. The
torque of the PMSM is controlled by a reference current, calculated by the speed
controller. Due to the higher q-axis inductance, a negative d-axis current is
impressed to benefit from the reluctance torque. In position control mode, the speed
reference is given by an overlaid position controller (figure 5.1), using the estimated
rotor position as input.
The program code of the EKF is optimized according to the remarks specified in
chapter 5. The computing requirement of the final algorithm takes up 236
multiplications and 178 summations. Compared to the induction machine, both the
EKF program code and the control algorithm are less extensive. The real-time
92
Chapter 6
implementation of the Kalman filter integrated into the motion controller is carried
out using a TMS320C31 DSP in which the turnaround time of the entire control
system amounts to 153 s. Therefore, the filter can operate in a system having a
maximum sampling frequency of 6,5 kHz, or a theoretical system bandwidth of
3,25 kHz. This high bandwidth allows the EKF to be used in high-performance realtime motion systems.
93
the PMSM to a load step (75% rated torque) at a motor speed of 1000 RPM. The
information on load acceleration is directly used to compensate for the load torque.
Rejecting load disturbances increases the dynamic stiffness of the drive. Therefore,
this feedback causes the disturbance to perceive a more robust system responding
less to disturbances. Compared to a common PI speed controller, the overshoot at
steps of both speed reference and load torque is vastly decreased or even vanishes
since the speed controller used (figure 5.1) contains of no integral-acting part.
n [rpm]
500
250
nref
0
-250
(a)
-500
0
0.1
0.2
0.3
0.4
0.5
20
15
i [A]
0.6
0.7
0.8
0.9
1.1
1.2
0.7
0.8
0.9
1.1
1.2
0.7
0.8
0.9
1.1
1.2
0.7
0.8
t [s]
i*q
10
iq
5
0
-5
(b)
0
0.1
0.2
0.3
0.4
0.5
n [rpm]
20
10
0
n = nest - nm
-10
-20
0.1
0.2
0.3
0.4
0.5
0.6
(c)
t [s]
0.2
[rad]
0.6
t [s]
0.1
0
-0.1
-0.2
0.1
0.2
0.3
0.4
0.5
0.9
1.1
1.2
(d)
t [s]
Figure 5.3: Speed reversal test. Top: Speed reference nref, estimated speed nest and measured speed nm.
Middle: Estimated q-axis current (b) and difference between estimated and measured speed (c).
Bottom: Error of the angle estimation and influence of current/voltage signal lag.
94
Chapter 6
n [rpm]
1020
nm
1000
nest
980
0
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
20
1.2
1.4
1.6
1.8
1.2
1.4
1.6
1.8
1.8
t [s]
i [A]
15
10
5
0
-5
T [Nm]
15
Tel
10
Tm
Tload
0
-5
t [s]
0.2
0.4
0.6
0.8
t [s]
Tel
1.2
1.4
1.6
Figure 5.4: Response to a load step (75% rated torque). Top: Measured and estimated speed. Middle:
Estimated q-axis current. Bottom: Estimated load Test, measured load Tm and electromagnetic torque Tel.
In steady state, the estimated load in figure 5.4 equals to the measured load. This is
not obvious since iron and friction losses of the PMSM are part of the estimated but
not of the measured torque. Furthermore, erroneous electromagnetic torque
calculations and inertia identification are directly reflected at the load calculation.
However, this influence is small compared to a potential load variation. In fact, the
torque calculation might be incorrect, but the real load is compensated by the real
torque and the steady state speed error sticks to zero. Only the dynamics of the load
estimation are important for exact speed calculation without any delay. As can be
seen, the delay between estimated and measured load is insignificant. Therefore, the
proposed speed control offers a vast improvement of the drive performance also if
the load is not absolutely known.
At low motor speed (n 0), the equations of the PMSM are simplified, as the
voltage induced by the magnets is very small. Therefore, no prediction can be made
on the position of the magnets and the EKF fails. Since at standstill only dc-values
are given, the necessary flux variation must be forced by impressing a test signal
into the system. A signal, easy to implement, is an additional sinusoidal reference
current in the d-axis of the motor, using the d/q axis-symmetry of the rotor to
estimate the real position. In all experimental results presented the following d-axis
reference current is used:
id ,ref = id* + itest
,
n
1
s
300 rpm
(5.24)
whereby the reference d-axis current id* results from the speed controller calculating
the required torque motor. The amplitude and frequency of the test signal is
95
experimentally chosen regarding observer stability and low acoustic noise level.
Nevertheless, further investigations on optimal shape, frequency and magnitude of
the additional d-axis current have to be made. Figure 5.5 presents the response of the
d- and q-axis current to a step of the speed reference from standstill to 1000 rpm.
The corresponding speed signal is shown in figure 5.6. The unwanted reluctance
torque, generated by the test signal in the d-axis, is compensated by an appropriate
q-axis current. The modification of the q-axis current iq*, calculated by the speed
controller, is obtained by the demand for a constant electromagnetic torque, not
disturbed by the impressed test signal.
) )
iq ,ref = iq*
Md Lq Ld id*
Md Lq Ld id ,ref
itest
*
= i q 1 +
Md
id ,ref
Lq Ld
(5.25)
(5.26)
20
10
i [A]
15
5
0
0
0.2
0.4
0.6
0.8
1.2
1.4
t [s]
id
n = 300 rpm
i [A]
-5
-10
n = 1000 rpm
0
0.2
0.4
0.6
0.8
1.2
1.4
t [s]
Figure 5.5: Current at a speed step (figure 5.6). Top: Reference i*q and q-axis current iq.
Bottom: Reference i*d and d-axis current id.
The developed torque remains nearly constant as can be seen on figure 5.6, showing
the corresponding speed response, marked optimum torque control. The optimal
control of the motor takes advantage of the reluctance torque by introducing a
negative (Ld < Lq) direct axis current component. In the same figure, a comparison is
given of motor control with feedback of the estimated speed and position, optimum
d-axis current and no d-axis current respectively. In spite of identical maximum
current amplitude, the maximum torque using optimum torque control is higher,
yielding a faster acceleration of the motor. The bandwidth of the speed control with
the EKF is comparable to the common control with speed measurement due to the
omission of the filter for speed measurement.
96
Chapter 6
1200
nref
1000
Optimum torque
control
n [rpm]
800
id = 0
600
400
200
0.2
0.4
0.6
0.8
1.2
1.4
t [s]
Figure 5.6: Speed step with feedback of the estimated speed and position (EKF). Comparison of torque
control with optimum d-axis current and no d-axis current respectively.
97
magnet position, the resulting convergence is very smooth. If the initial value of the
estimated position is opposite to the rotor position, the motor temporarily rotates in
the wrong direction. This effect can be avoided by operating the drive in open-loop
control and impressing a test signal in one motor phase. Once the position is
detected, the drive returns to the closed-loop control.
The controller has to ensure that the motor never experiences loss of
synchronization. However, the rotor asymmetry makes the PMSM also suitable for
position control (figure 5.8). For the drive setup with the 3 kW PMSM, the steadystate error of the electrical position angle is smaller than 2,3. This error, as well as
the performance of the algorithm, mainly depends on the quality and accuracy of
voltage and current measurement. It should be remarked, that the proposed
algorithm only identifies the electrical position. The absolute mechanical rotor
position is not detectable.
4
0
-2
-4
-0.2
-0.4
est
0
[rad]
[rad]
0.2
0.5
1.5
est
0
0.5
t [s]
100
10
0
-50
-100
nm
0
1.5
1.5
nm
nest
n [rpm]
n [rpm]
50
t [s]
0.5
t [s]
1.5
0
-5
-10
nest
0
0.5
t [s]
Figure 5.7: Start-up of the sensorless speed control. Top: Estimated position est and measured positionm.
Bottom: Estimated and measured speed. Left: Initial value of the position estimation almost opposite to
the rotor position. Right: Initial value of the position estimation almost aligned to the rotor position.
98
Chapter 6
[rad]
ref
2
1
0
0
0.05
0.1
0.15
0.2
[rad]
0.3
0.35
0.4
0.45
0.5
0.3
0.35
0.4
0.45
0.5
0.3
0.35
0.4
0.45
0.5
0.05
0
= est - m
-0.05
-0.1
0.05
0.1
0.15
0.2
0.25
t [s]
200
nref
150
n [rpm]
0.25
t [s]
0.1
nm
nest
100
50
0
-50
0.05
0.1
0.15
0.2
0.25
t [s]
Figure 5.8: Position control of the 3 kW PMSM. Top: Position reference ref, estimated position est and
measured position m. Middle: Difference between estimated and measured position. Bottom: Speed
reference nref, estimated speed nest and measured speed nm.
99
p (Md
Tel*
( Lq Ld ) id )
(5.27)
According to (2.50), the optimum torque control of the PMSM yields two solutions
for the d-axis current reference id*:
2
Md
Md
+ iq* 2
i =
2 ( Lq Ld )
2 ( Lq Ld )
*
d
(5.28)
The positive sign is valid for PMSM with Ld > Lq. Here, only the case Ld Lq
(PMSM with inset magnets) is considered. Thus, the negative sign in (5.28) must be
used.
Erroneous flux estimation yields incorrect speed estimation. Furthermore, it is
directly reflected in the torque-current mapping. An incorrect torque-current
mapping is not compensated by the speed controller since the implemented speed
controller with load torque rejection consists of a proportional gain and contains no
integral-acting part. Thus, erroneous flux estimation results in a steady-state error of
the speed control loop. However, this error is used for flux adaptation. The structure
of the implemented flux adaptation, the speed controller with load torque rejection
and the modified torque-current mapping is shown in figure (5.9). The
electromagnetic torque is almost proportional to the flux linkage. According to (5.3)
and (5.8), increasing the estimated flux linkage results in a higher absolute value of
100
Chapter 6
sign( *)
Md
Md
Md = K dt
Kn
J/p
Flux
error
Tel*
|Tel| < Tmax
Tload
i*q
Md
x1
Initial value
x2
id
Lq-Ld
x12 + x22
i*d
Initial value
Current mapping
Figure 5.10 presents experimental results of the proposed flux adaptation. An initial
error of the flux linkage (20 %) has been introduced resulting in poor motor speed
estimation. The speed estimation as well as the steady-state error is affected by a
parameter mismatch. The flux adaptation detects the steady-state error and corrects
the initial flux linkage. After a short period, the estimated speed matches the
measured speed, indicating the correct estimation of the flux linkage.
0.26
0.31
0.25
Md
0.28
[Vs]
estimated
flux linkage
0.29
0.27
Md
[Vs]
0.3
real flux
linkage
0.23
0.22
0.21
0.26
0.25
0.24
0.5
1.5
2.5
0.2
0.5
n [rpm]
n [rpm]
10
Parameter adaptation
switched ON
5
0
-5
-10
n = nest - nm
0
0.5
1.5
t [s]
1.5
2.5
t [s]
t [s]
10
2.5
Parameter adaptation
switched ON
5
0
-5
-10
n = nest - nref
0
0.5
1.5
2.5
t [s]
Figure 5.10: Adaptation of the flux linkage. Top: Estimated and real flux linkage. Bottom: Difference
between reference speed (nref = 1000 rpm), estimated speed nest and measured speed nm. Left: Initial flux
linkage 20 % overrated. Right: Initial flux linkage 20 % underrated.
101
Considering PMSMs with magnet placing of the inset-type, the d-axis inductance
Ld is generally independent of the load state [Cha 85]. A slight Ld-mismatch and
variations due to different saturation levels are completely compensated by an
appropriate variation of the flux linkage Md [Van 98]. Figure 5.11 presents the flux
adaptation, based on the structure shown in figure 5.9, at variable d-axis current, no
load and a motor speed n = 1000 rpm. The coincidence of estimated and measured
speed verifies the proposed approach.
15
i [A]
10
5
0
-5
-10
-15
10
15
20
25
[Vs]
Md
35
40
45
50
55
60
35
40
45
50
55
60
55
60
0.26
0.25
0.24
n [rpm]
30
t [s]
0.27
10
15
20
25
n = nest - nm
30
t [s]
0
-2
-4
n = nest - nref
0
10
15
20
25
30
35
40
45
50
t [s]
Figure 5.11: Flux adaptation at variable d-axis current (no load). Top: d-axis current.
Middle: Estimated flux linkage. Bottom: Difference between reference
speed (nref = 1000 rpm),estimated speed nest and measured speed nm.
102
Chapter 6
i [A]
15
10
5
0
10
15
20
25
[Vs]
Md
35
40
45
50
55
60
35
40
45
50
55
60
35
40
0.26
0.255
0.25
n [rpm]
30
t [s]
0.265
10
15
20
25
n = nest - nm
30
t [s]
0
-2
-4
n = nest - nref
0
10
15
20
25
30
45
50
55
60
t [s]
Figure 5.12: Flux adaptation at variable load (i*d = 0 A). Top: q-axis current.
Middle: Estimated flux linkage. Bottom: Difference between reference
speed (nref = 1000 rpm), estimated speed nest and measured speed nm.
5.7 Conclusions
This chapter presents the design and the implementation of sensorless speed control
of permanent magnet synchronous motor drives. The algorithm used is based on the
extended Kalman filter theory. A systematic and analytic approach for developing
the algorithm is given. The discrete extended Kalman filter is well suited to speed
and rotor position estimation of a PMSM. The proposed approach has been validated
by means of real-time experiments using a TMS320C31 DSP. The high bandwidth
allows the EKF to be used in high-performance real-time motion systems.
The known electromagnetic torque is used as part of the speed estimation, vastly
increasing the accuracy and dynamic performance of the drive. The implemented
speed controller with load torque rejection contains no integral-acting part,
providing a system with extremely high stiffness to disturbance inputs. The speed
estimation does not lag the actual motor speed, both in steady state and during
periods of acceleration or braking. A negative d-axis current is impressed to benefit
from the reluctance torque.
The presented sensorless control scheme is self-starting. At low motor speed, the
required flux variation is forced by impressing a test signal in the d-axis. The
unwanted reluctance torque is compensated by a complementary q-axis current. Due
to the rotor asymmetry, the PMSM is also suitable for position control.
103
6.1 Introduction
The use of photovoltaic (PV) energy sources for water pumping and irrigation
applications, especially in remote or rural areas in developing countries, is receiving
considerable attention. Large urban populations in developing countries do not have
access to safe drinking water sources (standpipes or boreholes) or to sanitary
services (sewers, septic tanks or wet latrines). According to statistics of the World
Health Organization, the number of people without access to safe water in 1990 was
1,1 billion [WHO 96]. Human health depends on an adequate supply of potable
water. Therefore, PV-powered water pump systems can improve peoples living
conditions, where power from a utility is not available or too expensive to install.
Furthermore, it is not economically viable to connect such remote areas to the
national electric grid.
While many of the references for residential applications are available in technical
details, it is difficult to locate technical references for the interaction between PV
arrays and an electric machine, especially in water pumping without battery storage
[Mul 97]. This chapter briefly reviews present technology and applications of PV
powered water pump systems and exhibits an extensive description of a new control
approach.
The two basic design approaches of PV arrays for water pumping system
applications are the use of battery, for a backup of the pumping system, and the
other is to pump directly from the PV power without battery. There are advantages
and drawbacks associated with each design. With a battery module, the system
energy generated by the sun can be stored in the battery. With the second approach,
the motor/pump subsystem can be powered either by directly connecting to the PV
array, or by using a maximum power point tracker (MPPT), a dc-dc converter and an
inverter interfacing motor and PV array. In this work, a system is designed, not
requiring a battery. The fact that no battery is required is a key element in the
design. Batteries tend to be very unreliable in the overall framework and
furthermore, they are of interest to people living there for other purposes, too
(read: they are often stolen).
The system analyzed here is a PV powered water pumping system avoiding the use
of the additional dc-dc converter, a battery and its losses. Several new approaches
106
Chapter 8
inverter
PMSM
water
storage
pump
Phase current
for EKF
MPPTracking
Current [A]
PWM
Voltage [V]
control
prototyping
DC bus voltage/current
measurement for MPPT
107
Furthermore, the control system is equipped with a MPPT and a voltage control
guaranteeing a balanced input/output power ratio in the dc bus. The MPPT and
voltage control are realized by feeding back the dc bus voltage and current to the
controller. Additionally, measurements of motor currents in two phases are required
for speed estimation and torque control of the PMSM. The motor voltages required
for the EKF are not measured, but calculated from the reference voltages
determining the PWM output of the controller.
A computer-aided control system is used as a developing platform monitoring and
recording the experimental data. Provisionally, the entire control algorithm, safetyrelated monitoring and the start-up and shut down automatisms are implemented on
a TMS320C31 DSP. The I/O subsystems and the PWM generation are based on
TMS320P14 working as a slave-DSP. However, the final algorithms are intended to
be implemented in a simple microcontroller ensuring an overall low cost system.
6.3 PV Array
The voltage-current characteristic of one PV element at constant cell temperature
and with the irradiance of the sunlight as a parameter is shown in figure 6.2. In the
same figure, also the output power of the PV element is drawn as a function of the
module voltage. The power PPV is calculated by the product of dc bus voltage Udc
and current IPV. The Maximum Power Point (MPP) is characterized by the voltage,
where the PV array generates maximum output power.
135
MPP
120
105
current
90
[W]
60
45
30
15
power
10
15
20
PV
75
PV
[A]
25
U dc [V]
The PV element characteristics are a function of the irradiance of the sunlight and
the cell temperature. In figure6.2, six different levels of insolation are illustrated.
High current curves correspond to high insolation levels while low current curves
correspond to lower insolation levels. With increasing irradiance, the MPP moves
108
Chapter 8
along the marked line. In order to stay at the point of maximum power at rising
irradiance, the current in the dc bus must be increased, while the dc voltage remains
nearly constant.
The voltage at the MPP changes with the array temperature while the current is
almost unaffected. At lower cell temperature, the MPP characteristic is situated in a
higher voltage range. The voltage temperature coefficient of the PV elements used
amounts to 82 mV/C. Therefore, connecting 12 modules in series and a
temperature variation of 10 K results in an optimum voltage shift of 9,84 V on a
rated voltage of 180 V, i.e. 5 %. Thus, the optimum output voltage of the PV array
is not constant and moves as condition varies.
The practically studied PV array consists of 36 modules with a total peak power of
4,32 kW. All wires of the single PV elements are assembled in a modular way using
a switchboard panel, connected in series or parallel. The different experimental
connections are presented in Table 6.1. The experimental results obtained are similar
demonstrating the high flexibility of both the previously and later described control
algorithms. However, to match the requirements of the final inverter and to keep the
presented results clear, all experimental results presented in this chapter are
measurements using 12 modules in series and 3 modules in parallel.
Table 6.1: Various connections of the PV array with a peak power of 4,32 kW.
Number of
modules in
series
6
9
12
18
Number of
modules in
parallel
6
4
3
2
Imax [A]
Umax [V]
44,7
29,8
22,35
14,9
129
193,5
258
387
109
Figure 6.3: Cross sectional view of the 4-pole PMSM rotor geometry.
A centrifugal pump commonly requires a single quadrant drive. The load torque of
the centrifugal pump expressed as a function of speed is
Tload = K 2
(6.1)
where K is the constant of the hydraulic system. Thus, to vary the output of the
water pump, the speed of the motor driving the pump must be varied. The properties
of the PMSM used are summarized in appendix B.3.
6.4.1
110
Chapter 8
calculation of the optimum voltage cannot be avoided due to parameter and load
torque variations as well as measurement errors and the non-linearity of the inverter.
25
20
15
I [A]
Torque
[0->10Nm]
10
0
10
11
12
13
14
U [V]
15
16
17
Due to dead-time effects, the error between reference and output voltage of the
inverter used amounts to U 15 V resulting unaccompaniedly in a current
variation of 22 A for the studied PMSM at a speed of 300 rpm. According to (1.16)(1.17), the voltage error U depends on PWM frequency, dc bus voltage, dead time
and current direction.
This non-linear effect creates a distortion of motor current and torque. Without a
damper cage, the PMSM in open loop is an undamped, oscillating system [Mel 91],
[Hen 91]. Slight variations of the electrical angle
= 0 +
(6.2)
(6.3)
= sin( 2 f e t )
(6.4)
fe =
1
2
Tmax cos0
J p
(6.5)
The frequency fe of the PMSM used varies between 0 Hz < fe <10 Hz for maximum
load (0 = 90) and no load (0 = 0) respectively. This oscillation creates a
111
2 J 12 '
R2 (1 + 1 ) 2
3 p 2 U 12
(6.6)
The modified PMSM used has a damper time constant of Tdamp 60 ms. Considering
the natural frequency, this value is far too large. For a sufficiently damped system,
the time constant and thus the rotor resistance must be smaller. Therefore, a stable
and dynamic open-loop control of the given PMSM is impossible.
A modification of the PMSM geometry, including a completely closed damper cage
[Con 86] and inset magnets, could improve the performance. Nevertheless, it is
much better to consider another control approach. As explained in the next sections,
a high dynamic drive is indispensably for a correct operation of the water pump
system. In fact, this has been the starting motivation for the development of the
earlier described high-performance motor drive with speed, flux and torque
estimation.
Switching losses
Valve losses
Copper and iron losses in the filter coil
In control systems without a battery, the dc bus may collapse when an unbalanced
input/output power ratio occurs at the dc bus. Therefore, systems without battery
require a more complex and complicated control algorithm. The second unit controls
the speed of motor and pump.
Here, a novel control approach is proposed avoiding the use of the additional dc-dc
converter, a battery and its losses. The overall control of the PV-powered water
pump system consists of a current/torque controller, a speed controller without a
shaft sensor and a main control consisting of a dc bus voltage controller and the
112
Chapter 8
MPPT. The voltage control varies the speed/torque of the PMSM in order to stay
within the calculated optimum voltage given by the MPPT. The structure of the
overall control system is shown in figure 6.5.
PV power
supply
IPV
IPV
Tel*
Udc
START-UP
&
Udc*
MPPT
Voltage
control
-1
PI with
anti windup
*
PI with
anti windup
ua*
SVM
Torque * Inverter
u
b
Control
&
uc*
EKF
Tel*
*
Tel
||<max
Udc
ib
PID with
anti windup
pump
ia
AC
motor
In contrast to the very quickly and frequently changing irradiance intensity, the cell
temperature of the PV array and thus the dc voltage in the MPP varies very slowly.
Therefore, the control of the PV-powered water pump system is performed by
varying the dc voltage in a small range, searching the MPP and controlling the
speed/torque characteristic of the motor in order to stay within the calculated
optimal voltage corresponding to the highest efficiency of the system. The adaptive
MPPT algorithm is described in detail in sections 7.7.
The dc bus voltage can be controlled either by the speed of the motor requiring an
additional speed control loop or directly by the electromagnetic torque affecting the
motor speed derivation. The different control approaches are indicated by the switch
in figure 6.5. Advantages and drawbacks of both control approaches are explained in
the next section. The first approach turned out to malfunction, when tested under
extreme but realistic conditions. However, many algorithms described in literature
are based on the variation of the speed reference; e.g.: [Mul 97] varies the power by
changing the frequency output of the inverter stepwise, being even slower than using
an extra speed control loop.
The high-performance speed/torque control of the PMSM is described in previous
chapters. The mechanical position sensor is replaced by an observer requiring no
additional measurements. Only measurements of motor current and dc bus voltage
are necessary. Figure 6.6 shows the experimental results of a speed step with a
centrifugal pump as load and using a regular grid as power supply (Udc = 220V).
With sufficient power generated by a PV array, the results obtained would be the
same. Otherwise, the dc bus would be discharged and the system collapses. The
applied load at n = 2000 rpm amounts 85% of the rated motor torque. The
current/torque controller has a bandwidth of 960 Hz. According to subsection 2.5.1,
113
the optimal torque control of the motor takes advantage of the reluctance torque by
introducing a negative (Ld < Lq) direct-axis current component increasing the
efficiency of the drive.
2500
n [rpm]
2000
Reference
1500
Measurement
1000
500
0
0.25
0.5
0.75
1.25
1.5
1.25
1.5
t [s]
30
i [A]
20
iq
10
id
0
-10
0
0.25
0.5
0.75
t [s]
Figure 6.6: Speed step with a pump load using a regular grid as power supply.
Top: Measured and reference speed. Bottom: d- and q-axis current.
The start-up and shut down logic is based on the motor speed and the open-circuit
voltage. Below a speed of n 180 rpm, the centrifugal pump used is not able to
pump water. A non-productive idle run is not conducive for the durability of the
pump and all other wear. Therefore, the system goes in standby modus and all PWM
pulses are disabled if the PV power supply is definitively too low. The energy
consumption of all control and measurement units and the inverter in standby modus
amounts to 20 W. A start-up procedure requires a pre-determined minimum opencircuit voltage guaranteeing a productive motor speed, being higher than the
minimum speed of the shut down automatism.
The implemented safety-related monitoring consists of detecting over-current and
over-speed, both depending on the motor/pump system, and a pre-determined
voltage window mainly defined by the PV array coupled to the inverter. An
inadmissible failure disables the entire system, requiring a manual reset by an
expert. However, no such failure has been detected during weeks of testing.
114
6.6.1
Chapter 8
In figure 6.7 the response of the voltage, q-axis current and speed for a step of the
voltage reference from 225 to 125 V and back to 225 V is shown. It can be seen, that
the voltage and the torque producing q-axis current are controlled very fast. They are
already in steady state, while the speed still varies. The speed of the motor changes
indirectly, controlled by the electromagnetic torque until the reference voltage is
reached.
250
U dc [V]
200
Reference
150
Measurement
100
0.2
0.4
0.6
0.8
0.6
0.8
0.6
0.8
t [s]
30
i [A]
15
-15
-30
0.2
0.4
t [s]
n [rpm]
900
800
700
600
0.2
0.4
t [s]
Figure 6.7: Step of the voltage reference from 225V to 125V and back to225V.
Top: dc bus voltage. Middle: q-axis current. Bottom: Motor speed.
The MPP of the PV array used is situated between 185 and 195 V. Without voltage
control the voltage area below the MPP (Udc < 185 V) is unstable. Slightly
increasing the motor speed in this unstable area, results in a very quick discharge of
the capacitor leading inevitably to a crash of the entire system. The averaged voltage
error of this inner control loop is smaller than 0.1%. Even at the starting procedure
and under very quickly changing irradiance, the error reaches a maximum of 0.5%.
The averaged voltage error delivers the minimum search range for the later
described main control (MPPT) providing the reference voltage.
An important feature of the system is the independence of the pipe characteristic.
The pumping head or water pressure can be varied by a throttle lever
increasing/decreasing both water pressure and reversely water flow. Thus, also the
pipe characteristic changes. Figure 6.8 demonstrates this independence by varying
the pumping head from a m (= bar) to 10 m (=10 bar) and back to m. The
115
200
195
195
[V]
200
185
180
Reference
190
dc
190
dc
[V]
response of the voltage control is plotted above, below the speed. The reaction time
of the voltage control is quite slow due to the time intensive valve closure. The
experiments are made with a reference voltage of Udc = 191V approximately
agreeing with the MPP of the PV array.
185
180
1500
1500
1400
1400
1300
1200
1100
t [s]
t [s]
n [rpm]
n [rpm]
t [s]
1300
1200
1100
t [s]
One of the most important features of the voltage control is its robustness during
power interruptions, occurring at instantaneous decrease of irradiance (e.g. passing
clouds). This property has been tested using a regular grid as power supply and
applying a complete power interruption on all three phases for a short time agreeing
with the worst-case condition of the system. These tests were done with a smaller,
3 kW prototype PMSM. The implemented regenerative braking scheme allows the
inverter to keep its dc bus voltage at the pre-determined minimum level, expanding
the time in which supply voltage can be reapplied without the time-consuming dclink capacitor recharging cycle. The experimental results of such short time threephase power interruptions are shown in section 8.5.
116
Chapter 8
voltage remains nearly constant. The voltage at the MPP changes with array
temperature and current is almost unaffected. At lower cell temperature, the MPP
characteristic is situated in a higher voltage range. Thus, the optimum output voltage
of the PV array is not constant and moves as conditions vary.
The MPPT is the main control loop, calculating the MPP and the search range of the
dc bus voltage, and delivers a reference quantity to the voltage control loop. First, a
default voltage and search range is given. After the default voltage is reached, it is
varied slightly around this point. The quantity of this variation is given by the search
range. During this variation, the power generated by the PV array is measured and
the voltage linked to the maximum power is stored during the respective searching
procedure. The new optimum voltage and the new search range are calculated from
these actual measurements and in its stored values by an adaptive control algorithm.
With these new quantities, the controller starts again a searching procedure to find
the MPP. Figure 6.9 shows the flow chart of the MPP-Tracking.
Startup with
initial values
Measurement
Calculate Uopt
Calculate U
Decrease Uref
NO
Uref
YES
Reference
voltage Uref
Increase Uref
NO
Uref
Uref
t
YES
The previous values are very important for the calculation of the new optimum
voltage and search range. If, e.g., the new calculated optimum voltage during a
searching procedure with rising voltage is situated higher than the last optimum
voltage, the MPP-voltage seems to change. However, this can also indicate an
increasing irradiance. If the second condition occurs, the controller should not
change the new optimum voltage. Otherwise, the calculated voltage drifts from the
MPP. The same considerations are also valid for decreasing irradiance. Thus, the
adaptive control must be able to distinguish between a changing MPP and changing
conditions. A new optimum voltage is only calculated, when a tendency is indicated
by a searching procedure with both rising and falling voltage reference.
117
Pictorially expressed, the shape of the reference voltage looks like the movement
playing an accordion. Both, search range and speed depends on the variation of the
calculated optimum voltage. If a new calculated MPP is situated in the half of the
past voltage range, the search range and the search speed are reduced, otherwise
both are increased. If the irradiance power changes very often and too fast to track
the real MPP, the MPPT algorithm behaves as a common constant voltage tracking.
PV
[kW]
1
0
50
100
150
200
250
300
200
250
300
t [s]
2000
n [rpm]
1500
1000
500
0
50
100
150
t [s]
Starting with the open-circuit dc voltage, in which the power generated by the PV
array supplies only the control and measurement electronics (~20 W), the voltage
decreases in voltage control mode to a pre-determined reference value. The MPPT is
switched on after reaching this operation point and searches subsequently for the
optimum voltage, where the PV array generates maximum power. The MPP is
reached after approximately 15 s and the voltage is varied from now on slightly
around this point.
During the MPPT, the power generated by the PV array is measured and the voltage,
linked to the maximum power, is calculated. The implemented system tracks
automatically the present conditions; e.g. with increasing insolation, the optimum
voltage is situated in a higher voltage range. As can be seen from the details of
figure 6.11, the variation of the voltage depends on external influences as insolation
118
Chapter 8
Udc [V]
240
220
Box 1
200
180
50
100
150
200
250
300
285
290
295
t [s]
195
Udc,ref
Udc [V]
194
193
192
191
190
265
Udc
270
275
280
t [s]
Due to the lack of storage element in the dc bus, the power of the PV array must be
used immediately to accelerate the PMSM. Measurements during a sunny day of the
implemented MPPT are plotted in figure 6.12 showing the power of the PV array as
a function of the dc voltage Udc.
Three different starting conditions are shown to demonstrate the ability of
reproduction of the MPPT. The direction of the searching procedure is indicated by
the arrows. The characteristic indicated by Start 1 refers to the time exposure of
figures 6.10-6.11. The artificial starting point exhibits the MPPT starting in an
unstable area, where slightly increasing the motor speed results in a very quick
discharge of the capacitor. This starting point is reached by first using the voltage
control mode with a reference Udc* = 120V and then switching over to the MPPT
control mode. However, this artificial starting point is never reached during regular
operation. The measured results demonstrate the ability of reproduction as well as
the stability of the entire system.
119
3.5
MPP
3
2.5
PV
[kW]
Start 2
Artificial
starting point
1.5
Start 1
0.5
0
100
120
140
160
180
200
220
240
U dc [V]
Figure 6.12: Measured results of the MPP-Tracking using a PV array with a peak
power of 4,32 kW and a PMSM driving a centrifugal pump (sunny day).
MPP
200
d
[W]
150
PV
c
b
100
50
a
0
120
130
140
150
160
170
180
190
200
210
U dc [V]
Figure 6.13: Measured results of the MPP-Tracking using a PV array with a peak
power of 1,2 kW and an induction motor driving the pump (6 hours of a cloudy day).
120
Chapter 8
6.9 Conclusions
This chapter presents the design and the implementation for PV-powered water
pump systems using a PMSM without a shaft sensor. In a first approach, the
performance of a PMSM with damper cage in open loop control is evaluated. Due to
the insufficient damper cage, a stable and dynamic open-loop control of the given
PMSM is impossible. Furthermore, a simple U/f control is absolutely inferior
compared to a field-oriented high performance motor drive with speed, flux and
torque estimation. The price to be paid is a more extensive and more complicated
control algorithm. However, no additionally measurements are required.
Additionally, an increased efficiency of the entire system is achieved.
Due to the lack of storage in the dc bus, the power of the PV array is used
immediately to accelerate the motor. Practical investigations are done to
demonstrate the stability of the dc bus voltage control, the independence of the
pump characteristic and its robustness during power interruptions. To optimize the
energy captured by the PV array and to pump as much water as possible, the output
power should always be at its maximum power point. Increasing the efficiency of
the system is very advantageously considering the cost-intensive PV array
installation amounting to 70% of the total system costs. Therefore, a novel MPPT
algorithm, realized by feeding back the dc voltage and current to a controller, has
been developed and implemented. The measured results of the MPPT exhibit the
ability of reproduction and the stability of the entire control system.
The development of a second prototype PMSM for submersible applications has
been stopped for practical and economic reasons. The design constraints due to the
mechanical construction of the motor housing and stator iron, together with the
filling of the motor interior with water yield a very unusual mechanical rotor
construction. The classical rotor design of a permanent magnet synchronous
machine with surface mounted magnets would inherit the fixing of the magnets with
glue and a polymer bandaging. The long-term stability for use directly in water of
both permanent magnets and bandage cannot be guaranteed by the manufacturer.
This applies even for coated magnets. The production costs of such a machine are
enormous. Due to the design constraints, only a marginal efficiency increase can be
expected by replacing the induction motor with the PMSM. However, the
implemented MPPT and voltage control are suitable for both a PMSM and an
induction motor driving the pump.
In the meantime, a request for the installation of the presented PV powered water
pump system has been received from four different countries: Mali, Mauritania,
Senegal and Chad.
7. Conclusions
122
Chapter 9
Conclusions
123
approaches were developed because the algorithms described in literature for this
kind of systems turned out to malfunction. A novel maximum power point tracking
optimizing the energy captured by the PV array has been designed and implemented.
The PV-powered water pump system consists, among other control loops, of a highperformance dc bus voltage control, which constitutes the most important control
function guaranteeing the stability of the drive.
As discussed in chapter 8, the realized dc bus voltage control is also applicable for
ride-through schemes at power interruptions considering inverter-controlled drives
supplied by a regular grid. The proposed solution to the problem is to recover some
of the mechanical energy stored in the rotating masses by kinetic buffering. This
maintains the dc link capacitor well charged keeping the electronic control circuits
active. The temporary speed dip is generally tolerable, since the most frequent
power interruptions last only for a few milliseconds. Furthermore, the proposed ridethrough scheme at power interruptions has been transformed into a special drive
braking tool saving energy and simplifying the inverter setup: the installation of
brake-resistance, power switch and cooler may be eliminated.
124
Chapter 9
The proposed observer together with advanced control techniques can be applied to
the active filter (active front-end) design, which forms nowadays an interesting field
in the area of power quality. Especially a disturbance (current harmonics) rejection
approach, similar to the proposed load torque rejection approach within the
speed/current control loop of drives, promises a vastly increased performance.
Classical control theory suffers from some limitations due to non-linearity, timeinvariance etc. of the controlled system. These problems can be overcome by using
artificial-intelligence-based control techniques. In literature, e.g. [Vas 99], it is
expected that intelligent sensorless instantaneous torque-controlled drives
incorporating some form of intelligence will become the standard in the future.
These drives will not require machine or controller parameters, and all the control
and estimation tasks are performed by a single artificial-intelligence-based system.