6.5K views

Uploaded by abdurrahim_t

- Power Inverter Hard Ware Design
- Tesla Model s (Electric Motor Specifications)
- Current Control Techniques for Three-phase Voltage-source PWM Converters
- Speed control of induction motor
- Modelling, simulation and analysis of low-cost direct torque control of PMSM using hall-effect se
- DC/AC Pure Sine Ware Inverter
- Current Source Inverter
- Final Project Report
- Full Control of a PWM DC AC Converter for AC Voltage Regulation
- Voltage Source Inverter_2
- Dstatcom Project Report
- Sinusoidal Pulse Width Modulation
- CSI vs VSI
- Electrical Machines and Drives
- 36-1 FIELD ORIENTED CONTROL OF A MULTILEVEL PWM INVERTER FED INDUCTION MOTOR
- IEC 61727.pdf
- FULL REPORT
- PSIM Tutorial
- Ansoft Maxwell 3D v11 Userguide
- dvr project

You are on page 1of 125

1.1 Introduction

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 today’s motor drives. Presently, inverter drives meet not

only weight and space constraints, but also are economically viable.

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.

2 Chapter 1

providing a pulsed dc voltage from the mains, is required. Alternatively, a second

ac-to-dc converter, acting as a rectifier during the motoring mode and an inverter

during the breaking mode, is used between drive and utility grid. An additional

benefit of the active front end is enabling unity power factor, (sinusoidal) current

flows to or from the grid.

Although the basic circuit for an inverter may seem simple, accurately switching

these devices provides a number of challenges for the power electronic engineer.

The most common switching technique is called Pulse Width Modulation (PWM).

PWM is a powerful technique for controlling analog circuits with a processor’s

digital outputs. PWM is employed in a wide variety of applications, ranging from

measurement and communications to power control and conversion. In ac motor

drives, PWM inverters make it possible to control both frequency and magnitude of

the voltage and current applied to a motor. As a result, PWM inverter-powered

motor drives are more variable and offer in a wide range better efficiency and higher

performance when compared to fixed frequency motor drives. The energy, which is

delivered by the PWM inverter to the ac motor, is controlled by PWM signals

applied to the gates of the power switches at different times for varying durations to

produce the desired output waveform.

There are several PWM modulation techniques. It is beyond the scope of this book

to describe them all in detail. The following illustration describes the basic three-

phase inverter topology and typical pulse width modulation methods. Furthermore,

issues of phase voltage distortion/identification due to the inverter non-linearity are

discussed in detail.

stages: ac to dc and dc to variable frequency ac. The basic converter design is shown

in figure 1.1. The grid voltage is rectified by the line rectifier usually consisting of a

diode bridge. Presently, attention paid to power quality and improved power factor

has shifted the interest to more supply friendly ac-to-dc converters, e.g. PWM

rectifier. This allows simultaneously active filtering of the line current as well as

regenerative motor braking schemes transferring power back to the mains.

The dc voltage is filtered and smoothed by the capacitor C in the dc bus (figure 1.1).

The capacitor is of appreciable size (2-20 mF) and therefore a major cost item

[Bose 97]. Alternatively, the inverter can be supplied from a fixed dc voltage. The

filtered dc voltage is usually measured for control purpose. Because of the nearly

constant dc bus voltage, a number of PWM inverters with their associated motor

drives can be supplied from one common diode bridge. The inductive reactance L

between rectifier and ac supply is used to reduce commutation dips produced by the

rectifier, to limit fault current and to soften voltages spikes of the mains.

Voltage-Source PWM Inverter 3

T1 T3 T5

L D1 D3 D5

C

power AC

Udc

supply motor

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.

600 600

Udc Udc

U dc [V]

U dc [V]

500 500

uab -uca ubc -uab uca -ubc uab -uca ubc -uab -ubc

uca

400 400

0 5 10 15 20 0 5 10 15 20

t [ms] t [ms]

1 40

iB6 [A]

iB6 [A]

0.5 20

0 0

0 5 10 15 20 0 5 10 15 20

t [ms] t [ms]

1 40

20

ia [A]

ia [A]

0 0

-20

-1 -40

0 5 10 15 20 0 5 10 15 20

t [ms] 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

4 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.

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).

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 medium-

power 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.

inductive load, the current increases subsequently. If the load draws positive current,

Voltage-Source PWM Inverter 5

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

C/2 T1 D1 ia

½ Udc

ia > 0 0

ωt

ua0

C/2

½ Udc T4 D4

T1 off ua0 D1 T1

drop drop

T4 off ½ Udc

ua0

C/2 T1 D1 T4 on

½ Udc 0

ia < 0 τdead ωt

T1 on

ua0

C/2 -½ Udc

½ Udc T4 D4 D4 T4

drop 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 -½ 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 ½ Udc by the on-state voltage drop of T1. When the load

current reverses, the output voltage will be higher than ½ 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 center-

tapped inverter is represented as generating the voltage ½ Udc and -½ Udc,

respectively. Neglecting additionally the dead-time interval τdead, the behavior of the

power devices together with the reverse recovery diode is equally described by ideal

two-position switches.

6 Chapter 1

Usually, the on- and off-states of the power switches in one inverter leg are always

opposite. Therefore, the inverter circuit can be simplified into three 2-position

switches. Either the positive or the negative dc bus voltage is applied to one of the

motor phases for a short time. Pulse width modulation (PWM) is a method whereby

the switched voltage pulses are produced for different output frequencies and

voltages. A typical modulator produces an average voltage value, equal to the

reference voltage within each PWM period. Considering a very short PWM period,

the reference voltage is reflected by the fundamental of the switched pulse pattern.

Apart from the fundamental wave, the voltage spectrum at the motor terminals

consists of many higher harmonics. The interaction between the fundamental motor

flux wave and the 5th and 7th harmonic currents produces a pulsating torque at six

times of the fundamental supply frequency. Similarly, 11th and 13th harmonics

produce a pulsating torque at twelve times the fundamental supply frequency

[Dub 89]. Furthermore, harmonic currents and skin effect increase copper losses

leading to motor derating. However, the motor reactance acts as a low-pass filter and

substantially reduces high-frequency current harmonics. Therefore, the motor flux

(IM & PMSM) is in good approximation sinusoidal and the contribution of

harmonics to the developed torque is negligible. To minimize the effect of

harmonics on the motor performance, the PWM frequency should be as high as

possible. However, the PWM frequency is restricted by the control unit (resolution)

and the switching device capabilities, e.g. due to switching losses and dead time

distorting the output voltage.

There are various PWM schemes. Well-known among these are sinusoidal PWM,

hysteresis PWM, space vector modulation (SVM) and “optimal” PWM techniques

based on the optimization of certain performance criteria, e.g. selective harmonic

elimination, increasing efficiency, and minimization of torque pulsation [Jen 95].

While the sinusoidal pulse-width modulation and the hysteresis PWM can be

implemented using analog techniques, the remaining PWM techniques require the

use of a microprocessor.

A modulation scheme especially developed for drives is the direct flux and torque

control (DTC). A two-level hysteresis controller is used to define the error of the

stator flux. The torque is compared to its reference value and is fed into a three-level

hysteresis comparator. The phase angle of the instantaneous stator flux linkage space

phasor together with the torque and flux error state is used in a switching table for

the selection of an appropriate voltage state applied to the motor [Dam 97],

[Vas 97]. Usually, there is no fixed pattern modulation in process or fixed voltage to

frequency relation in the DTC. The DTC approach is similar to the FOC with

hysteresis PWM. However, it takes the interaction between the three phases into

account. In the following subsections, hysteresis PWM, sinusoidal PWM and SVM

are discussed in more detail.

Voltage-Source PWM Inverter 7

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.

Current reference

Hysteresis band ∆i

ia Real current

Switching

logic

ia*

ia ∆i

ib* 0

ωt

ib ∆i ua0 Output

voltage

1/2 Udc

ic*

ic ∆i 0

ωt

-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 -½ 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 α/β-

reference frame. These modified techniques take care especially for the interaction

of the three phases [Jen 95].

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]:

lower subharmonics.

• The current error is not strictly limited. The signal may leave the hysteresis

band caused by the voltage of the other two phases.

• Usually, there is no interaction between the three phases: No strategy to

generate zero-voltage phasors.

• Increased switching frequency (losses) especially at lower modulation or

motor speed.

• Phase lag of the fundamental current (increasing with the frequency).

8 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 variable-

reluctance motors.

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.

three separate comparators with a common triangular carrier wave of fixed

amplitude and frequency (figure 1.5-1.6). Each comparator output forms the

switching-state of the corresponding inverter leg [Dub 89], [Leo 85]. In torque

controlled ac motor drives using sinusoidal PWM, the reference voltages (u*a, u*b, u*c)

are usually calculated by an additional current control loop (FOC).

Switching

logic

d,q ua*

id* ud*

comparator

id Current ub*

controller

iq Current uc*

controller a,b,c

comparator

Carrier wave

the fixed PWM frequency, is simultaneously used for all three phases. This

modulation technique, also known as PWM with natural sampling, is called

sinusoidal PWM because the pulse width is a sinusoidal function of the angular

position in the reference signal.

Voltage-Source PWM Inverter 9

Uref

ωt

Udc/2

ωt

-Udc/2

Lower switch “on”

ub0

ωt

uc0

ωt

uab

ωt

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

∫ u a 0 dt =

1

(∆t1 − ∆t 2 ) U dc (1.1)

TPWM T

PWM

TPWM 2

10 Chapter 1

1 u a* 0 1 u a*0

U dc 2 U dc 2

0 0

t t

-1 Saw-tooth -1 Triangular

carrier wave carrier wave

∆t1 ∆t2 ∆t1/2 ∆t2 ∆t1/2

ua0 ua0

ua0 ua0

Udc /2 Udc /2

u a 0 = u a*0 u a 0 = u a* 0

t t

-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:

2 ! u a* 0

−1+ ∆t1 = (1.2)

TPWM U dc 2

TPWM u*

⇒ ∆t1 = 1 + a 0 (1.3)

2 U 2

dc

TPWM u*

∆t 2 = TPWM − ∆t1 = 1 − a 0 (1.4)

2 U 2

dc

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:

1 U dc TPWM u* T u*

u ao = 1 + a 0 − PWM 1 − a 0 (1.5)

TPWM 2 2 U 2 2 U 2

dc dc

⇒ u ao = u a*0 (1.6)

value, equal to the reference voltage within each PWM period. Therefore, the

Voltage-Source PWM Inverter 11

voltage.

The modulation technique using a saw-tooth shaped carrier wave always sets the

output to a high level at the beginning of each PWM period, resulting in

asymmetrical PWM pulses. The pulses of an asymmetric edge-aligned PWM signal

always have the same side aligned with one end of each PWM period. On the

contrary, the pulses of a symmetrical PWM signal, e.g. obtained by using a

triangular-shaped carrier wave, are always symmetric with respect to the center of

each PWM period. The symmetrical PWM is often preferred, since it generates less

current and voltage harmonics [Bose 97], [Dub 89].

The sinusoidal PWM is easy to realize in hardware by using analog integrators and

comparators for the generation of the carrier and switching states [Ter 96]. However,

due to the variation of the reference values during a PWM period, the relation

between reference and carrier wave is not fixed. This introduces subharmonics of the

reference voltage causing undesired low-frequency torque and speed pulsations. In

contrary, software implementation provides sampled data during a PWM period

(uniform/ regular sampling) and hence, the pulse widths are proportional to the

reference at uniformly spaced sampling times. Compared to the analog

implementation, the modulation with uniform sampling has lower low-frequency

harmonics. Since the phase relation between reference and carrier wave is fixed,

even for the asynchronous mode, the subharmonics and the associated frequency

beats are not present [Dub 89].

The ratio of the reference magnitude to that of the carrier wave is called modulation

index m. Considering the mean output voltage equal to the reference phase voltage

(1.6) in the linear range (m ≤ 1), the fundamental component of the line voltage is:

U line 3 Uˆ phase 3

= = m, m≤1 (1.7)

U dc 2 U dc 2 2

(figure 1.8). For m > 1, the number of pulses becomes less and the modulation

ceases to be sinusoidal PWM. The modulation is still working, but the output

voltages are no longer sinusoidal: they correspond to the reference values with

limitation to the half dc bus voltage. The fundamental component of the line voltage

then is [Jen 95]:

U line 3 1 1

= m ⋅ arcsin + 1 − 2 , m>1 (1.8)

U dc π 2 m m

12 Chapter 1

0.8

0.7

6

π

0.6

3

[]

0.5

dc

/U

0.4 2 2

line

U

0.3

0.2

0.1

0

0 1 2 3 4

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 a* 0

*

u U dc 2 U dc 2

1

-1

ωt

carrier wave

ua0

fundamental

4 U dc ua0

π 2

U dc ωt

−

2

u

U dc uab

U dc

2

−

U dc ωt

2 ub0

−U dc

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 1

ua0 =

π 2

∑n 2n − 1 sin[(2n − 1) ωt ] (1.9)

limited by the dc bus voltage:

Voltage-Source PWM Inverter 13

2

Uˆ phase,max = U dc (1.10)

π

results in a strong increased spectrum of lower voltage and current harmonics

especially for the 5th, 7th and 11th harmonics. In figure 1.10, the current of an

induction motor (scalar control) in the linear range (m = 1) and at overmodulation

(m = 1,33) is presented to illustrate the involuntary current distortion.

m=1

2

ia [A]

0

m = 1,33

-2

-4

0 0.01 0.02 0.03 0.04 0.05

t [s]

(induction motor in open loop: uref = 200 V sin(ωt); Udc = 400 V and Udc = 300 V resp.)

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.

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:

!

u ab = u a 0 − ub 0 =(u a 0 + uthird ) − (ub 0 + uthird ) (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 a*0

u*a0 U dc 2 third harmonic

t

reference plus

third harmonic

1

t

-1

carrier wave

ua0

U dc

2

U dc t

−

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.

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].

Voltage-Source PWM Inverter 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 α/β frame through α/β-

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 α/β voltages in terms of the dc bus

voltage Udc are summarized in table 1.1.

Table 1.1: Switching table and α/β transformation of affiliated state voltage vectors.

State S1 S3 S5 Ux,α Ux,β |Ux|

000 OFF OFF OFF 0 0 0

110 ON ON OFF U dc 3 U dc 3 2 U dc 3

011 OFF ON ON −2 U dc 3 0 2 U dc 3

101 ON OFF ON U dc 3 − U dc 3 2 U dc 3

111 ON ON ON 0 0 0

Uβ

010 110

S2

S3 S1

000 Uα

011 111

100

S4 S6

S5

001 101

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 α/β 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.

1

U ref = (t1 U x + t 2 U x+60 ) (1.12)

TPWM

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

40% ‘100’ 100 110 110 100

ua0

50% ‘110’ 5% ‘000’

ub0

5% ‘111’

Uref = U ejωt uc0

TPWM

As mentioned above, the reference voltage is actually equal to the desired three-

phase output voltages mapped to the α/β 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

Voltage-Source PWM Inverter 17

3 U

U line , max = Uˆ phase , max = dc (1.14)

2 2

being approximately 15% higher when compared to the original sinusoidal PWM.

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 α/β

components Uα* and Uβ*, 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 α/β components specifying the position in the α/β-plane. For instance, if the

reference voltage Uβ* 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 α- and β-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. TI’s 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 b* + 1 c* + 1

duty cycles a ; b; c = ; ; 100 % (1.15)

2 2 2

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. TI’s 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.

Uα*, Uβ * Voltage

reference

3 1

2 U dc

normalization

uβ ≥ 0

*

No

Yes

1 No No −1

uα* ≥ u *β uα ≥ u *β

3 3

−1 No No 1

uα* ≥ u *β uα* ≥ u *β

3 3

1 * 1 *

a * = uα* + uβ a * = 2 uα* a * = uα* − uβ

3 3

3 2

b * = −uα* + u *β b* = u *β b * = −a *

3 3

3

c * = −a * c * = −b* c * = −uα* − u *β

3

duty ratios:

(a*; b*; c*) PWM 1−6

15 P14

2 -1

DSP

|u| ≤ 1 16 bit

Overflow compare

protection 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.

Voltage-Source PWM Inverter 19

“shoot-through” effect of a half-bridge during a change of the switching states. All

semiconductor-switching devices react delayed to the turn-off signals owing to the

storage time. During this storage time, depending on the operating point, the switch

is not able to block the dc link voltage. Therefore, to avoid a short circuit of the half-

bridge, a dead-time interval must be introduced between the turn-off signal of a

switch and the turn-on signal controlling the opposite switch. The dead time τdead is

usually constant and determined as the maximum value of storage time τst plus an

additional safety margin. The dead times of common IGBT-inverters used in

industry vary between τdead ≈ 1-5 µs.

Although the dead time is short, it causes deviations from the desired fundamental

inverter output voltage. The effects of the dead time on the output voltage will be

described from one half-bridge of the PWM inverter according to figure 1.15. The

basic configuration consists of upper and lower power devices T1 and T4, and

reverse recovery diodes D1 and D4.

T1 on T4 off

C/2 T1 D1 C/2 T1 D1

½ Udc ½ Udc

ia > 0 ia < 0

ua0 ua0

C/2 C/2

½ Udc T4 D4 ½ Udc T4 D4

T1 off T4 on

pulse pattern T4 pulse pattern T4

τdead τdead

pulse pattern T1 pulse pattern T1

with dead time T4 with dead time T4

τdead fPWM Udc τdead fPWM Udc UD1

UT1

½ Udc ½ Udc

ua0

ua0

-½ Udc -½ Udc

UD4 UT4

Left: Positive load current. Right: Negative load current.

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

UT + U D

∆U ≈ + τ dead ⋅ f PWM ⋅ U dc (1.16)

2

depending on the dead time τdead, 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:

RT + RD

u a 0 ≈ u ref − i ⋅ − ∆U ⋅ sign(i ) , (1.17)

2

The dead time reduces the effective turn-on time and produces the undesired fifth-

and 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 dead-

time 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 (τdead = 2,5 µs;

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.

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 µs), will not free

the system of the described distortion. Higher PWM frequencies improve the

Voltage-Source PWM Inverter 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

2 n [rpm]

50

I [A]

0

45

α

-2

40

-4 35

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

t [s] t [s]

40 40

uref

20 20

U [V]

U [V]

0 0 uβ (uα)

β

α

uα

-20 -20

uref

-40 -40

0 0.25 0.5 0.75 1 -40 -20 0 20 40

t [s] U [V]

α

Figure 1.16: Open-loop control of an induction motor & dead-time effect (Udc =500 V, no load).

Left: Measured current Iα, measured voltage Uα and reference voltage Uref.

Right: Measured/reference speed and voltage trajectories.

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

since the actual storage delay is not exactly known. Furthermore, the PWM

generation is a part of a superimposed high-bandwidth current control loop

compensating the involuntary torque/speed distortions to a certain extent. This may

eliminate the need for a separate dead-time compensator. Figure 1.17 illustrates the

dead-time effect on an induction motor drive in field-oriented speed control mode at

low speed and light-load operation. Except the control mode, the conditions are the

same as in figure 1.16.

4 60

55

2

n [rpm]

50

I [A]

0

45

α

-2

40

-4 35

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

t [s] t [s]

40 40

uref

20 20

U [V ]

U [V ]

0 0 uβ (uα)

uα

β

α

-20 -20

uref

-40 -40

0 0.25 0.5 0.75 1 -40 -20 0 20 40

t [s] U [V ]

α

Figure 1.17: Field-oriented control of an induction motor & dead-time effect (Udc =500 V, no load).

Left: Measured current Iα, measured voltage Uα 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].

Voltage-Source PWM Inverter 23

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.

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:

Us −Ud

R= (1.18)

IP

negative voltage (e.g.: Us = ±12V) results in a discharging of the capacitor

depending on the photodiode operating voltage (Ud ≈ 1V if i > 0). While the

photodiode P1 directly blocks, the dead-time τdead passes before the capacitor

voltage equals the voltage -Ud, equal to the on state of photodiode P2 driving the

opposite switch of the inverter leg:

−τ dead !

U c (t = τ dead ) = (U s + U d ) e R C − 1 + U d =− U d (1.19)

24 Chapter 1

τdead:

τ dead

⇒ C≥ (1.20)

2 Ud

R ln1 −

Us + Ud

Switching logic

US

Optocoupler 1

t

IP1 -Us

UC UC

-Ud UC

C

t

PWM logic: IP2 -Ud

US = ±12V R

IP

Optocoupler 2 IP1 IP2 IP1

t

τdead τdead

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.

Variable speed ac drives are used in ever-increasing numbers because of their well-

known benefits for energy efficiency and for flexible control of processes and

machinery using low-cost readily available maintenance-free ac motors. While the

connection of a motor to an inverter supply is straightforward, some basic

considerations are necessary to ensure trouble free long-term operation. Insulation

performance is one of the considerations required in engineering variable speed

drive solutions. Following summary provides basic information to enable the correct

matching of low voltage ac motors and PWM inverters with respect to motor

insulation:

with an inverter than when connected directly to the ac mains supply.

The higher stresses are dependent on the motor cable length and are caused

by the fast rising voltage pulses of the drive and transmission line effects in

the cable.

For supply voltages less than 500V ac, most standard motors are immune to

these higher stresses.

Voltage-Source PWM Inverter 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 three-

phase 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. Regenerative Braking and Ride-

Through at Power Interruptions

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 multi-

motor 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.

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.

A voltage dip is a short-duration reduction in the supply voltage, in many cases due

to network faults somewhere in the energy distribution system. During a voltage dip,

the voltages in the three phases are no longer the same, causing a number of

problems. A major fault more than 100 km away from a customer may still yield a

significant voltage dip. Mains voltage dips and short interruptions are caused by a

wide variety of phenomena. They can be caused by nearby events, such as a faulty

load on an adjacent branch circuit causing a circuit breaker to operate, or perhaps by

a large motor or other large load on the same circuit being switched on. They can

also be caused by far away events such as lightning strokes or downed power lines.

In case of a fault in the power distribution grid, an automatic circuit recloser may

cycle open and close several times within a short period attempting to clear the fault,

thus resulting in a sequence of short interruptions noticed by downstream loads. In

any case, the voltage changes produced can affect the operation of or even damage

nearby electrical equipment as e.g. drives. Therefore, immunity for these types of

events should be available to ensure safe and reliable product operation.

Voltage dips are probably the power quality disturbance with the highest impact on

customers. The voltage drop yields tripping of process control equipment such as

adjustable-speed drives, process computers and switchgears. This in turn leads to

production halts, lasting much longer than the dip itself. Voltage dips of 100 ms

duration can lead to production halts of 24 hours or more. The economic impact per

event may be less than for regular interruptions, but the annual impact is in many

cases higher.

An ac motor directly connected to the regular grid may slow down during such a

power failure. An air-gap flux wave may be still in existence, but its magnitude,

phase angle and speed changes. Then, a return of the voltage with inadequate values

necessarily produces large current/torque transients. As has been reported by

industrial users, these transients generated by the motor may even cause a break of

the drive shaft. However, this problem can be overcome using a simple relay as a

Regenerative Braking and Ride-Through at Power Interruptions 29

special mechanisms may be required.

Concerning motor drives supplied by voltage source inverters, a dip on all three

phases leads to an instantaneous decrease of the dc link voltage, whereas a single-

phase dip may allow continued operating of the drive, albeit at higher rectifier stress.

Rectifier bridges must be properly designed to withstand these high peak currents.

Due to advances in semiconductor technology, modern variable speed drives can

tolerate the high peak currents occurring when the power supply is restored after a

short disturbance. Furthermore, powerful digital signal processors enable drive

manufacturers to implement regenerative braking schemes allowing the inverter to

keep its dc-link voltage at a required minimum level.

The availability of electrical power from the public supply as a function of the down

time at interruptions (in Germany) is given in [Sch 85] indicating that a power

interruption of more than 10 ms is likely to occur every 200 h, on average. Against

this, the mean times between failures due to long time power interruptions are of the

order of several 10 000 h. Short time interruptions of the power supply are therefore

the most frequent cause for inverter failure. A ride-through scheme at these short-

time power interruptions is presented in the next subsection.

dc link to stiffen the dc bus voltage and provide a path for the rapidly changing

currents drawn by the inverter. However, the amount of energy stored in the dc link

capacitor is normally insufficient to maintain the inverter active during a short

power failure interval. When a power interruption occurs, the dc-link energy is

absorbed by the motor within a few milliseconds. Since the electronic control

system loses power as well, the inverter shuts down commanded by an under-

voltage protection in order to avoid possible damage to the electronic or drive

equipment [Baa 89]. It is then required to wait until the machine has come to a

complete standstill to enable restarting. However, time-intensive restarting is

obviously not an adequate solution.

described in [Sei 92]. The control scheme is applicable to general-purpose inverters

with scalar motor control. Although this scheme can catch a running machine, the

time required for synchronization (up to 6 s) is too long for many critical

applications. It becomes even more severe with multi motor drives. Here, a solution

is presented using the high dynamic performance of a field-oriented motor control.

The dc link capacitor is a major cost item in the drive system and an increase of the

capacitor is therefore economically not feasible [Bose 97]. In contrast, the kinetic

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

Umin

t1 t2 t3 t4 t5 t

detection

1

Dip

0

t

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

Regenerative Braking and Ride-Through at Power Interruptions 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

t

1

( )

3

C U dc2 , N − U min

2

= ∫ (Ploss + ω Tload ) dt (2.1)

2 t1

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

2

( )

C U dc2 , N − U min

2 1

+ J ω ref

2

2

= ∫ (Ploss + ω Tload ) dt (2.2)

∆t max

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 ωref. Using kinetic

buffering, a maintained and controlled operation of several seconds is possible.

Primarily, the proposed voltage control scheme at power interruptions has been

developed for a PV-powered water pump system [Ter 02]. There, the voltage control

is designed to withstand abrupt power interruptions, occurring at an instantaneous

decrease of the irradiance intensity (e.g. passing clouds). The total power failure

considered here can be regarded as a worst-case situation.

The most important control loop for the stability of the entire system is the dc bus

voltage control. The system has been set up to work independently in island

operation. All control and measurement units are supplied by the dc bus. A dc

voltage beyond given limits leads inevitably to a crash of the entire system. The

voltage reference is calculated by an overlaid MPP-Tracking and controlled directly

or indirectly by the speed of the motor.

Due to the lack of a major storage element in the dc bus, the power of the PV array

must be used immediately to accelerate the PMSM. As irradiance increases,

resulting in a higher output power of the PV array, the input power of the dc bus is

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.

motor-pump

IPV IInv system

Solar Idc

generator 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:

1

C∫

U dc = I dc dt (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.

dω

Tel = J + Tload (2.4)

dt

dω

Pel = ω Tel = Pkinetic + Ppump = J ω + ω Tload (2.5)

dt

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

Regenerative Braking and Ride-Through at Power Interruptions 33

the load. Without considering the drive efficiency, the input power of the inverter

matches the electromagnetic output power generated by the motor.

In steady state, the voltage Udc and motor speed ω are constant. The energy

generated by the PV array is completely used to pump water:

• U dc = const ⇒ I dc = 0 (2.7)

Pkinetic = 0

• ω = const ⇒ (2.8)

Ppump ≈ U dc I PV

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:

d 2ω dT dω

J >> load (2.9)

dt 2

dω dt

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 ω is described by:

dω

U dc I PV − U dc I dc = J ω + ω Tload (2.10)

dt

Ppump = Tload ω ≈ U dc I PV ≈ const (2.11)

dω

Pkinetic = J ω ≈ − U dc I dc (2.12)

dt

34 Chapter 2

dU dc dω

⇒ U dc C = −J ω (2.13)

dt dt

With the transfer function of the closed loop speed control beyond current/torque

limitations

ω (s) 1

= (2.14)

ω * ( s ) s τ speed + 1

and using (2.13), the resulting linearized transfer function with the reference speed

ω* as input and the dc bus voltage Udc as output can be written as

U dc ( s ) J 1 1

=− ⋅ ⋅ , (2.15)

ω * (s) C s τ speed + 1 s τ Vf + 1

where τspeed is the equivalent time constant of the speed control loop and τVf 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:

τ u = τ speed (2.16)

π

ϕ R (ω c ) = π − arctan (τ Vf ω c ) − (2.17)

2

π

⇒ ω c = tan( − ϕ R ) / τ Vf (2.18)

2

amplification at the cut-off frequency A(ωc) to zero:

C τu

( )

!

A(ω )ω =ω = −20 log − ω c − 20 log τ Vf ω c 2

+ 1 = 0 (2.19)

c J K pu

⇒ K pu = −τ u

C

J

ωc (ω τ )

c Vf

2

+1 (2.20)

Regenerative Braking and Ride-Through at Power Interruptions 35

During practical investigations, the best results have been obtained using a common

PI controller for the voltage control and choosing a phase margin ϕR = 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* ω*

Kpu T s /τ u

Udc -1 |ω| < ωmax

z

The dc bus voltage controlled by the speed of the motor has significant drawbacks.

Choosing a phase margin ϕR = 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].

triangularly. However, applying a ramp (∆U/s2) as a reference voltage and using

(2.15) results in a steady state voltage error Uerror:

∆U τ u C

U error = − (2.21)

K pu J

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.

current and can be controlled very fast with the equivalent time constant τeq,i of the

current control loop.

36 Chapter 2

Tel ( s ) 1

= (2.22)

*

Tel ( s ) s τ eq,i + 1

Neglecting the load torque, the following relation between motor speed and

electromagnetic torque is valid:

ω (s) 1

= , Tload = 0 (2.23)

Tel ( s ) J s

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.

(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 ) 1 1 1 1

=− ⋅ ⋅ ⋅ , (2.24)

*

Tel ( s ) JC s s τ eq ,i + 1 s τ Vf + 1

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 τu 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 ϕR, guaranteeing a stable system:

( )

Tu = k τ eq ,i + τ Vf = k τ σ , with: k > 1 (2.25)

kJC

K pu = − (2.26)

τu

1

⇒ ϕ R (ω c ) = arctan ( k ) − arctan ( ) (2.27)

k

Best results are obtained by choosing 10 < k <40, corresponding to a phase margin

of 55° < ϕR < 72°.

fB ≈ 235 Hz, being 16 times larger than the other approach. The calculation of the

bandwidth takes no current/torque limitation into account.

Regenerative Braking and Ride-Through at Power Interruptions 37

error, being obviously due to the integrating term in (2.24). This property is very

advantageously for the implementation of a MPPT.

Compared to the other approach, the dc bus voltage directly controlled by the torque

has many advantages regarding speed of response, steady-state error and robustness.

Thus, all following experiments are made based on this approach. The voltage

controller (figure 6.5) switches only in speed control mode, if the maximum speed is

reached and the PV array generates sufficient power or if the motor/pump system

pumps too much water for the storage capacity.

speed/position estimation within this PV powered water pump system is not

superfluous. Both, PMSM and induction motor, being part of a high performance

drive, require information of the field position.

The structure of the dc bus voltage control integrated into the speed control system

is presented in figure 2.4. The proposed control algorithm requires a fast torque

control scheme. The well-known principle of field orientation [Leo 85] is employed

here. The high performance speed/torque control of the given ac motor drives are

described in chapter 7 as well as the calculation of the controller parameter. The

mechanical position sensor is replaced by an observer requiring no additional

measurements. Only measurements of motor current and dc bus voltage are

necessary.

Whenever the logic signal (‘dip logic’, figure 2.4) indicates a power interruption, the

torque reference is temporarily switched from the regular speed controller to the

voltage controller. The ‘dip logic’ is obtained using a simple (digital) relay with,

considering the given installation, a switch on point UKB = 340 V and a switch off

point at 360 V. The predetermined reference voltage Udc* should be within these

boundaries to prevent involuntary torque transients or oscillations of the logic

signal: Ideal is the switch on point. The integrator is used for both speed and voltage

control. Of course, the integrator time constant is automatically tuned. This prevents

involuntary torque transients and saves computation time. Note the negative sign of

the voltage controller gain as well as the multiplication by the sign of the motor

speed. During power interruptions, the dc bus voltage can be maintained only when

negative electromagnetic power is generated by the motor. A positive power

decreases the dc voltage. Considering a four-quadrant drive, negative

electromagnetic power is generated by inverse signs of torque and speed:

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 ua SVM

Tel Torque

*

Udc

*

* Inverter

Control ub

Enable -Kpu Ts /τu

& *

voltage uc

Dip logic |T| < Tmax EKF

control

Speed ω* z-1

ib ia

Kpn Ts /τn

reference

ω PI speed control

AC

motor

load

sign(ω) ω

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.

source PWM inverter has been used to verify the proposed approach. Instead of

using a PMSM, the implemented regenerative voltage control is also suitable for an

induction motor driving the load. The dynamic performance of induction motor and

PMSM are similar, only the efficiency of the former is lower especially at partial

load. The used load machine is a dc motor drive with constant electrical excitation

coupled with a variable resistor bench. The performance of the ride-through at

power interruptions has been tested using a regular grid as power supply and

applying an abrupt power interruption on all three phases for a short time

(approximately 2 s). Figure 2.5 presents a measurement of dc bus voltage and motor

speed without applying the proposed voltage control. The speed reference amounts

to ω* = 1000 rpm and no load is applied. The dc link capacitor is discharged to a

critical level within ∆t ≈ 0,1 s. The under-voltage protection switches on and the

drive is out of control. This time is much shorter with applied load torque.

Regenerative Braking and Ride-Through at Power Interruptions 39

500

[V ] 400

300

Under-voltage protection

dc

200

U

100

0

0 2 4 6 8 10 12 14

t [s]

1000

800

n [rpm]

600

400

200

Out of control

0

0 2 4 6 8 10 12 14

t [s]

Top: Voltage of the dc link. Bottom: Motor speed.

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 time-

consuming 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

400

voltage dip

[V ]

380

dc

360

U

340

320

-1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4

t [s]

1200 20

1000 15

800 10

n [rpm]

i [A]

600 5

q

400 0

200 -5

-1 0 1 2 3 4 -1 0 1 2 3 4

t [s] t [s]

Top: Voltage of the dc link. Bottom: Motor speed and q-axis current.

420

400

voltage dip

[V ]

380

dc

360

U

340

320

-1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4

t [s]

1200 20

1000 15

800 10

n [rpm]

i [A]

600 5

q

400 0

200 -5

-1 0 1 2 3 4 -1 0 1 2 3 4

t [s] 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 ω ∆t

2

(2.29)

1 CU 2

⇒ ∆t =

2 Tload ω

Regenerative Braking and Ride-Through at Power Interruptions 41

requiring a time-intensive restart of the converter. With the implemented voltage

control, a kinetic buffering during a total energy drop is possible for several seconds.

The span of time depends on the drive moment of inertia and the actual speed at the

moment of the voltage dip. Figure 2.8 presents a measurement at sustained power

failure. Before the drive gets uncontrolled by the under-voltage protection, the motor

has come to a complete standstill. However, the motor stays controllable during

braking, being important especially for critical applications as multi-motor drives.

400

300

[V ]

200

dc

U

100

0

Under-voltage protection

-1 0 1 2 3 4 5 6 7 8 9

t [s]

1200

1000

800

n [rpm]

600

400

200

0

-1 0 1 2 3 4 5 6 7 8 9

t [s]

Top: Voltage of the dc link. Bottom: Motor speed.

into a special drive-braking tool. The upper voltage limit of the inverter may be

reached at fast braking of the drive. Regularly, the surplus generated kinetic energy

is handled using a braking-resistance within the dc link, a common dc bus or a two-

way PWM inverter (figure 2.9). Nevertheless, the voltage control can be adopted

allowing a controlled braking with a maximum predetermined dc link voltage and

without redirecting the kinetic energy. The kinetic energy is usefully conducted to

the load mainly responsible for the braking during the voltage control mode. Thus,

this deceleration tool can be used, when saving energy is preferable to a fast braking

with power dissipation in, e.g., a brake-resistance. If the dynamic performance is not

crucial, the installation of a brake-resistance, power switch and cooler may be

eliminated. Especially in small motor drives, the economic gain is considerable.

42 Chapter 2

Power Motor 1 Load 1

supply

Inverter 2

Braking- Motor 2 Load 2

S1

resistance

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

mode when the motor speed reaches U > 600V ⇒ 1 dc

logic

U < 590V ⇒ 0

the reference speed or at a voltage level ω * dc

ω

logic’ is obtained using a simple

(digital) relay linked to the required Figure 2.10: Switching logic of the braking tool.

speed information (figure 2.10).

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.

∆Wkin =

1

2

(

J ω12 − ω 22 ) (2.30)

Regenerative Braking and Ride-Through at Power Interruptions 43

tool, this energy is directed to the load. Considering the load torque as useful, energy

is saved.

1500

n [rpm]

1000

500

t [s]

600

[V ]

500

Speed Speed

control control

dc

U

t [s]

20

10

i [A]

0

q

-10

-20

-0.5 0 0.5 1 1.5 2 2.5

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.

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.

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. DSP-based Drive Control and

Measurements

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.

DSP’s 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.

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

Management

The motor control is implemented using a DSP based controller board with

additional I/O features and an encoder interface. The DS1102 single-board system

from dSpace™ (Germany) employs a TMS320C31 digital signal processor

operating at 60 MHz for the main program and a slave subsystem with a

TMSM320P14 fixed-point DSP for the I/O subsystems and PWM generation. In

addition, the used development platform contains a comprehensive selection of I/O

interfaces that meet typical requirements for rapid motor control prototyping:

• 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 single-

board 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 user-

codes 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.

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

Voltage-Source PWM Inverter 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-DSP’s 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 µs, the computation time

has been significantly reduced to 17 µs. This new code is automatically included at

every compilation of the main program. The implemented modifications are

summarized in a separate manual.

The laboratory test drive consists of a host PC for the controller board, an IGBT

inverter, and ac motor drive with variable load, current/voltage sensors and an

incremental encoder. Figure 3.2 shows the control setup with the DS1102 controller

board. Photos of the experimental set-up are presented in appendix A.

48 Chapter 1

and dc bus voltage. Here, the motor currents are measured in two phases using LEM

sensors. The dc bus voltage is measured via a galvanically isolated potentiometer.

These signals are fed to the interface connected to the inputs of the A/D converters.

Due to involuntary parasitic disturbances (EMC-problems), the measured signals

should be filtered in either an analog or digital way. In general, digital filtering is

preferred. Phase shifts introduced by filtering can be corrected (if necessary) by the

transformation angle from the stator to the rotor reference frame. Only low-weighted

analog first order filters with a cut-off frequency 5 kHz are added between

voltage/current measurements and the A/D converters of the controller board. The

rotor position is measured using an incremental encoder and directly fed to the

encoder interface of the controller board.

control Power

supply

MPP-

Tracking

prototyping

Current [A]

Voltage [V]

Interface Udc

DS1102-Processor Board PWM 1 *

ua

PWM 2 EXOR PWM

PWM 3 *

C31 ub Inverter

PWM 4 EXOR

PWM 5 *

P14 uc

PWM 6 EXOR

enable

ib ia

Incremental

encoder signal Θ AC

Load

motor

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

Voltage-Source PWM Inverter 49

always have the same side aligned with one end of each PWM period. On the

contrary, the pulses of a symmetrical PWM signal are always symmetric with

respect to the center of each PWM period. The symmetrical PWM is often preferred,

since it generates less current and voltage harmonics [Bose 97], [Dub 89].

In order to overcome the problem of asymmetrical PWM generated by the P14, each

of the two PWM channels are employed to generate the pulses for one phase as

shown in figure 3.2 and 3.3. By means of an EXOR gate, pulses symmetrical to the

center of the PWM period can be achieved if the switching times of each two

channels, depending on the required duty cycle, are calculated according to the

example reflecting the calculation for the first motor phase:

PWM 1 = (3.1)

2

1 + duty cycle phase a

PWM 2 = (3.2)

2

PWM 1

PWM 2

EXOR

u*a

TPWM TPWM

program. The ‘user-code’ offers the inclusion of handwritten C-code into the

initialization part and the timer-driven task running with the base sample time and is

preferable compared to the use of Simulink C-coded S-function since it saves

computation time. Every S-function block used in a Simulink model introduces an

execution time overhead of about 9 µs in the real-time program due to the associated

function calls. Considering the given development platform (DS1102 60MHz), the

computation requirement of the implemented SVM and the data transmission to the

slave DSP amounts to 22 µs.

50 Chapter 1

3.4 Measurements

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.

high-performance motor control. Measurement accuracy and bandwidth influence

directly the current control loop as well as all overlaid loops. Filtering a feedback

signal additionally decreases the dynamic response time of the loop [Leo 85].

Current is typically measured by one of two methods: voltage drop across a resistor

or magnetic transducer. Resistive shunt sensing has the advantage of a relatively

low-cost sensor. A drawback is the trade-off between sensitivity and power

dissipated in the resistor. Since the actual motor current is the desired value, the

sensing resistor is usually placed in series with the motor phase. This complicates

the measurement, because the signal of interest is a millivolt differential value

across the resistor, but the common-mode voltage of the motor phase is typically

hundreds of volts switching at high frequency with rapid du/dt.

Magnet sensors, on the other hand, are isolated by their very nature. This means that

the motor current can directly be measured without the common-mode voltage

problems mentioned before. They use a ring-type magnetic core with a Hall-effect

semiconductor element placed in an air gap to measure the magnetic flux resulting

from the primary current ip through the center of the core (figure 3.4). In “closed-

loop” Hall effect current sensors, a canceling coil of e.g. 1000 turns is wound around

the magnetic core. A built-in feedback amplifier drives current through the canceling

coil in such a way that the flux, measured by the Hall-effect sensor, is always forced

to be zero. Therefore, dc current can be measured. The output of the current

transducer is the canceling current, equal to the measured current scaled-down by

the ratio of coil turns. The overall bandwidth, accuracy and temperature

independence of these transducers has proven to be sufficient for motor drive

applications.

Voltage-Source PWM Inverter 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 µs. 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.

voltage is required for the exact transformation of the reference voltages into the

duty cycles for the inverter PWM. Even when the inverter is supplied by a constant

voltage (regular grid), the dc voltage varies due to load variations. In some

applications described later (e.g. power interruptions, braking schemes, PV-

systems), the dc bus voltage is the main control variable. Furthermore, knowledge of

the dc bus voltage makes a more complicated measurement of the phase voltages

superfluous.

The measurement of the dc bus voltage is not as crucial as the current measurement

since the dc voltage is filtered and smoothed by a capacitor of appreciable size

present in the dc bus. Usually, one side of the dc bus is grounded eliminating the

common-mode problem already described at the current measurement. In the

applications mentioned, the dc voltage is measured via a resistive potentiometer, a

high-performance galvanic isolation and a first-order analog filter (cut-off frequency

5 kHz) connected to one A/D-converter of the controller board.

The field-oriented control of ac motor drives, e.g. induction motor and PMSM,

demands the measurement of the motor current in two phases and the knowledge of

the dc link voltage. This makes a more complicated measurement of the phase

voltages superfluous. However, in some applications, such as sensorless field-

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.

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.

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 inverter’s 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.

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-to-

digital converters, but calculated by means of the reference voltages with

Voltage-Source PWM Inverter 53

by the SVM. A compensation of the dead-time effect is not implemented since the

actual storage delay, varying depending on the operating point, is not exactly

known. Practical investigations have shown even a deterioration of the observer

performance by using an inadequate compensation approach: If the compensation is

not perfect, a duplication of the dead-time effect at zero crossings of the current may

occur.

Due to the delayed reaction of almost all semiconductor switches at turn-on and

turn-off, the phase voltages strongly deviate from the reference voltages. The

voltage distortion does not depend on the magnitude of the reference voltages and

hence its relative influence is very strong in the lower speed range where the

reference voltage is small. Actually, the dead-time error is one of the major reasons

limiting the performance of sensorless control in low speed operation [Choi 94],

[Lee 96]. Disregarding this distortion yields in the subsequently described

speed/flux observer to large position and speed errors, especially at low motor

speed, where the error voltage becomes a multiple of the reference voltage.

Therefore, the dead-time effect is considered at the estimation of the phase voltages.

Figure 3.5 presents a comparison of the error voltage calculated by (1.17) and the

measured falsification of the fundamental voltages.

15

10

5

[V ]

10

0

-U

ref

U

-5

Measurement

-10

Equation (3.11)

-15

-10 -8 -6 -4 -2 0 2 4 6 8 10

I [A]

1

(Udc =400 V, fPWM = 10 kHz, τd = ± 2,5 µs).

PWM generation is performed by space vector modulation. SVM provides a more

efficient use of the supply voltage in comparison with sinusoidal modulation

methods by imposing a homopolar system u0 in all three phases (multiple of third

harmonics).

1

u0 = (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:

1 1 ua

uα 2 1 − −

2 2 u

u = (3.4)

3

b

β 3 3

0 − u

2

c

2

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:

1

U ref = U dc x ref , |a*| ≤ 1; |b*| ≤ 1; |c*| ≤ 1 (3.5)

2

All together, using these reference voltages, the required alpha/beta voltages uα, uβ

are calculated according to the block diagram in figure 3.6 considering both the non-

linearity of the inverter and the homopolar system of the SVM.

* * *

xref = (a ; b ; c )

SVM

uα

1/2

Udc Equation

(3.13) uβ

Eq.(3.10)

i1

i2

i3

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.

recording the experimental data. Furthermore, the safety-related monitoring and the

start-up and shut down automatisms are implemented in software on the main DSP

Voltage-Source PWM Inverter 55

over-current and over-speed, both depending on the drive system, and a pre-

determined voltage window. The minimum and maximum admissible dc bus voltage

mainly depends on the inverter used. An inadmissible failure disables the entire

system, requiring a manual reset. The reset signal is activated only by the rising edge

of a manual reset protecting the drive/inverter from an everlasting reset while an

error may be still active. In addition, fuses are integrated in the motor current circuit

as well as in the dc bus voltage measurement.

error ≡ 0 enable

no error ≡ 1

Udc 0⇔1 enable

error signal

Iα,β

logic

n

reset u>0

0⇔1

-1 -1

z z

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

transforming required measurements has been added to the experimental set-up.

Support software has been changed in order to implement different PWM strategies

as well as variable PWM frequencies on the TMSM320P14 slave-DSP. Different

PWM generation schemes are evaluated. A standard VS-PWM inverter with IGBTs

is adapted to be commanded by the DSP controller board. The PWM switching

signals are fed directly from the slave processor to the inverter using a high-

performance optical link. Furthermore, the inverter is supervised by an enable

subsystem.

discussed in detail. The required inverter output voltages are not measured but

calculated by means of the reference voltages with consideration of the inverter non-

linearity and the homopolar component generated by the SVM.

4. Sensorless Speed Control of

Induction Motor Drives

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).

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.

estimation via the Kalman filter approach. Choosing a stator flux reference frame

causes time-dependent states resulting in an error driven nature of the observer even

at steady state. Signal lags are inevitably increased. Significant problems arise

especially due to the zero crossing of the states at low motor speed and speed

reversal. Here, the system model of the induction motor used is based on the motor

equations in a rotor flux reference frame [Bla 72], [Hen 92]. The angle of the

transformation from the stator to the rotor reference frame coincides with the rotor

flux angle γ rotating at synchronous speed ωµ. Thus, the rotor flux lies entirely in the

d-axis. At steady state, all values, apart from the flux angle, are constant. The

electrical properties of the induction motor in continuous time are completely

described by two voltage equations of the stator, two rotor equations and a torque

equation:

did u diµ

στ 1 + id = d + στ 1ω µ iq − (1 − σ )τ 1 (4.1)

dt Rs dt

diq uq

στ 1 + iq = − στ 1ω µ id − (1 − σ )τ 1ω µ iµ (4.2)

dt Rs

diµ

τ2 + i µ = id (4.3)

dt

iq

ωµ = ωr + (4.4)

τ 2 iµ

L1h L2

Tel = p ψ rd iq = p 1h iµ iq (4.5)

Lr Lr

Sensorless Speed Control of Induction Motor Drives 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:

dω J dω r

Tel − Tload = J = (4.6)

dt p dt

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 ωµ 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

ωµ

200

ωr

ω [rad/s]

150

100

Load

ωslip step

50

0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t [s]

(simulation of a 0,8 kW induction motor drive)

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

did − id iq

iq − (1 − σ ) id − iµ + u d

dt

= + ωr +

στ 1 τ 2 iµ στ 2

( )

στ 1 Rs

(4.7)

− 1 (1 − σ )

iq + (1 − σ ) iµ + u d

i

= − id + ω r + q

στ 1 στ 2 τ 2 iµ στ 2 στ 1 Rs

− iq i

id − (1 − σ ) ω r iµ + q +

diq iq uq

= − ω r +

dt στ 1 τ 2 iµ σ τ 2 στ 1 Rs

(4.8)

− 1 (1 − σ )

id − (1 − σ ) ω r iµ +

i uq

= − iq − ω r + q

στ 1 στ 2 τ 2 iµ σ στ 1 Rs

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.

1 (1 − σ )

iq + Ts (1 − σ ) iµ + Ts u d

i

id ,k +1 ≈ 1 − Ts + id + Ts ω r + q (4.9)

στ 2

στ 1 τ 2 iµ στ 2 στ 1 Rs

1 (1 − σ )

id − Ts (1 − σ ) ω r iµ + Ts

i uq

iq ,k +1 ≈ 1 − Ts + iq − Ts ω r + q (4.10)

στ στ τ σ στ

1 2 2 iµ 1 Rs

respectively the flux position in the discrete time domain:

Ts T

iµ ,k +1 ≈ id + 1 − s iµ (4.11)

τ2 τ2

iq

γ k +1 ≈ γ + Tsω µ = γ + Tsω r + Ts (4.12)

τ 2 iµ

the rotor in one direction over a long time span. This non-linearity reflects no

negative influence on the EKF.

equations in discrete time and with the rotor speed as a variable. The speed must be

estimated by the filter. Thus, a suitable state equation is required. Because usually

neither the load torque nor its time variation is known, a simplification of the

mechanical equation is necessary.

Sensorless Speed Control of Induction Motor Drives 61

small time interval (sampling time Ts). Information on drive inertia is not required.

Mechanical and electrical model are fully decoupled. Nevertheless, this model

causes some problems. As will be shown, the estimated speed is lagging the real

speed during transients.

p 2 L2h p

ω r ,k +1 ≈ ω r + Ts iµ iq − Ts Tload = ω r + model noise (4.13)

J Lr J

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 d-

and q-axis current id, iq, rotor flux iµ, flux position γ and the electrical rotor speed ωr

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”:

id

iq

U s

x k +1 = A k x k + B k u k ; u k = αs ; x k = iµ ; (4.14)

U β

k ω r

γ k

iq

1 − T 1 + 1 − σ Ts ωr +

1− σ

Ts 0 0

s

στ 1 στ 2 τ 2i µ στ 2

iq 1 1− σ 1− σ

− Ts ωr + τ i 1 − Ts στ + στ − Ts σ ωr 0 0

2 µ 1 2

Ak = (4.15)

Ts Ts

0 1− 0 0

τ2 τ2

0 0 0 1 0

Ts

0 0 Ts 1

τ 2i µ

• Rs, Rr Stator and rotor resistance

• σ = 1-Lh2/(LsLr) Blondel coefficient

62 Chapter 5

• τ2 = Lr/Rr, Rotor time constant

The input matrix Bk describes the weighted transformation from a stator-fixed to the

rotor flux reference frame.

cos (γ ) sin (γ )

− sin (γ ) cos (γ )

T

Bk = s 0 0 , (4.16)

σLs

0 0

0 0

The resulting output vector yk consists of the estimated motor current in a stator-

fixed reference frame (α/β-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.

Iˆ s cos (γ ) − sin (γ ) 0 0 0

y k = ˆαs = Ck x k = xk (4.17)

Iβ

sin (γ ) cos (γ ) 0 0 0

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 γ, the matrix Ak on q-axis current iq, rotor flux iµ and

rotor speed ωr.

Iαs

UαS Measurement:

Iβ s

UβS ∆yk

+ z-1

+ +

xk+1 ⇒ xk γk -

yk

Bk Ck

+

∆x

+

γk+1 xk+1

EKF

Ak

Figure 4.2: Block diagram of the discrete motor model and EKF.

The second motor model, in future referred to as “Model 2”, uses additional

information on the electromagnetic torque generated by the motor. Additionally to

Sensorless Speed Control of Induction Motor Drives 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 ωr 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:

dα l d p

= Tload ≈ 0 (4.18)

dt dt J

p

⇒ α l ,k +1 ≈ α l = Tload (4.19)

J

Now, the known electromagnetic torque is used as part of the speed calculation,

improving the accuracy of the speed:

p 2 L2h

ω r ,k +1 ≈ ω r + Ts iµ iq − Ts α l (4.20)

J Lr

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:

id

iq

U s i

x k +1 = A k x k + B k u k ; u k = αs ; xk = µ (4.21)

U β ω r

k

γ

α

l k

64 Chapter 5

iq

1 − T 1 + 1 − σ Ts ω r +

1−σ

Ts 0 0 0

s

τ 2iµ

στ 1 στ 2 στ 2

iq

− Ts ω r + 1 − Ts 1 + 1 − σ

1−σ

− Ts ωr 0 0 0

τ στ

i

2 µ 1 στ 2 σ

Ts Ts (4.22)

Ak = 0 1− 0 0 0

τ 2 τ2

p 2 L2h

0 Ts iµ 0 1 0 − Ts

J Lr

Ts

0 0 Ts 1 0

τ 2iµ

0 0 0 0 0 1

cos (γ ) sin (γ )

− sin (γ ) cos (γ )

T 0 0

Bk = s (4.23)

σLs 0 0

0 0

0 0

Iˆ s cos (γ ) − sin (γ ) 0 0 0 0

y k = ˆαs = Ck x k = xk (4.24)

Iβ

sin (γ ) cos (γ ) 0 0 0 0

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

Sensorless Speed Control of Induction Motor Drives 65

determined employing equations (4.5), (4.6) and (4.19):

= = ⋅ p el (4.25)

dt dω r dt dω r J

p p 2 L2h dT

⇒ α l ,k +1 ≈ α l ,k + Ts iµ iq − α l ,k load (4.26)

J J Lr dω r

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.

The induction motor torque depends on both air-gap flux and speed, but neither

torque versus flux nor torque versus speed relations are linear. This complicates the

design of control systems and speed estimation for induction machines. Due to the

lack of a system with linear equations, also the state model of the induction motor

used is non-linear. The mechanical speed and position of the flux 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 iµ and rotor speed ωr.

Therefore, the extended Kalman filter (EKF) has to be used to estimate the

parameters of the model matrices, as well. The EKF performs a re-linearization of

the non-linear state model for each new estimation step, as it becomes available.

Furthermore, the EKF provides a solution that directly cares for the effects of

measurement or system noise. The errors concerning the parameters of the system

model are also handled as system noise.

A more complete introduction to the general idea of the Kalman filter can be found

in literature [Bram 94], [May 79], Bro 92]. Here, only the basic equations of the

EKF are repeated. The EKF algorithm used is based on [Bram 94]. The Kalman

filter estimates a process by using a form of feedback control. The signal flow of the

EKF in a recursive manner is shown in figure 4.3.

66 Chapter 5

ud

uq

∂Φ Predictor

∂x Pk+1|k 1/z

Pk|k-1

∂h

Filter

x k +1 k 1/z Pk|k Kk Kk ∆Yk

x k k −1 ∂x ∆ xk k

∆Y k = y −y

measured k

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.

( (

x k |k = x k |k −1 + K k y k − h x k |k −1 , k )) (4.27)

∂h

Pk|k = Pk|k −1 − K k |x= x Pk |k −1 (4.28)

∂x 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:

−1

∂h

T

∂h ∂h

T

K k = Pk |k −1

∂x

| x = x | −1 kk ∂ x x = x | −1 k |k −1 ∂ x | x = x | −1 + R

| Pkk kk

(4.29)

Based on the calculated state vector xk|k, a new value of the state vector can be

predicted. The same applies to the error covariance matrix. The prediction is:

(

x k +1|k = Φ k + 1, k , x k |k −1 , u k ) (4.30)

T

∂Φ ∂Φ T

Pk +1|k = | x = x Pk |k |x = x + Γ k Q Γ k (4.31)

∂x k |k

∂x k |k

Sensorless Speed Control of Induction Motor Drives 67

All equations of the EKF algorithm can be written as a function of a system vector

Φ and an output vector h describing the re-linearized model of the induction motor.

The system and output vector respectively can be derived from the model equations

of the induction motor.

( ) ( ) ( )

Φ 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 (4.32)

h(x k |k −1 ) ( )

, k = Ck x k |k −1 x k |k −1 (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 − σ Ts ω r + 2 0 0 0 0

s

στ 1 στ 2 τ i

2 µ

iq 1 1−σ id

− Ts ω r + τ i 1 − Ts στ + στ + τ i

0 0 0 0

2 µ 1 2 2 µ

∂Φ T s

= 0 0 0 0 0 +

∂x τ2

p 2 L2h

0 T s iµ 0 0 0 0

J Lr

Ts

0 0 0 0 0

τ 2 iµ

0 0 0 0 0 0

(4.34)

1−σ iq

2

Ts

0 0 Ts − Ts iq uq 0

στ 2 τ 2iµ 2 σ Ls

iq id 1 − σ 1−σ T

0 0 Ts − ω r − Ts id + iµ − s ud 0

τ 2 iµ 2

σ σ σ Ls

T

+ 0 0 1− s 0 0 0

τ2

p 2 L2h

0 0 Ts iq 1 0 − Ts

J Lr

iq

0 0 − Ts Ts 1 0

τ 2iµ 2

0 0 0 0 0 1

the product of α/β-voltages and input matrix Bk. Thus, the result can be used to save

computing time. Note that the q-axis voltage influences the linearized specification

of d-axis current and vice-versa.

dropping the last column as well as the last row in (4.34) and setting the elements

∂Φ ∂Φ

∂x

{4,2} and ∂ x {4,3} to zero. In this way, the influence of electromagnetic and

load torque on the speed is canceled.

Sensorless Speed Control of Induction Motor Drives 69

=

∂ x sin (γ ) cos (γ ) 0 0 id cos(γ ) − iq sin (γ ) 0

(4.35)

cos (γ ) − sin (γ ) 0 0 − iˆβ 0

=

sin (γ ) cos (γ ) 0 0 iˆα 0

The calculation of the estimated α/β-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.

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.

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.

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:

R =

−6

U − A (4.36)

1,5 ⋅10 − 4

dc

0 1,8 ⋅10 V 0

an estimation of the matrix elements is possible using some simplifications.

Furthermore, the given assumptions have been examined experimentally. All given

values are calculated using the parameters of the 1,5 kW induction motor drive and a

sample time Ts = 200 µs. The calculated values are valid for both system models;

by the accuracy of the voltage identification being the input of the system.

Sensorless Speed Control of Induction Motor Drives 71

Ts

current model inaccuracy ≈ ⋅ voltage inaccuracy (4.37)

σ Ls

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 dead-

time τdead, the dc bus voltage Udc and the PWM-frequency fPWM of the inverter:

So, the covariance of the current model inaccuracy can be estimated by:

2

1 T ∆U

Q(1,1) ≈ s (4.39)

3 σ Ls

Q(2,2) = Q(1,1) (4.40)

For the given drive, using a sample time Ts = 200 µs, a dead-time τdead = 2 µs and a

PWM-frequency fPWM = 10 kHz, the covariance amounts to:

only caused by the discretization of the continuous equations. In contrast to

Model 1, the speed specification within Model 2 is very accurate. The inaccuracy is

much lower and mainly caused by the discretization error. Considering a very small

sample time, these errors are negligible. Nevertheless, the worst of all

approximations is to set the model inaccuracy to zero. White noise is a much better

approximation than zero. Thus, this discretization error is considered by a very small

value in the noise covariance matrix Q. The maximum discretization error of the

magnetizing current is dependent on the maximum motor current and the rotor time

constant.

∞ id ,k − iµ ,k ( k +1)T id ,k − iµ ,k

s

k =0 τ τ

2 kT s

2

72 Chapter 5

i T imax

2

Ts2

⇒ Q(3,3) < var max s = ≈ 2,6 ⋅ 10 −5 A 2 (4.43)

τ2 τ 2

3 2

Assuming a maximum acceleration of αmax = 1000 s-2 and a sample time Ts = 200 µs,

the variance of the flux position is estimated by:

∞ ( k +1)Ts ( k +1)Ts

Q(5,5) = var ∑ Ts ω k − ∫ ω (t ) dt < var Ts α max k Ts − ∫ α max t dt (4.44)

k =0

kTs kTs

α T 2 α max

2

Ts4

⇒ Q(5,5) < var max s =

≈ 3,3 ⋅10 −11 (4.45)

2 48

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 τtorque = 1 ms is chosen.

∞ ( k +1)Ts

QModel 2 (4,4) = var ∑ Ts α k − ∫ α (t ) dt (4.46)

k =0

kTs

α T 2 α max

2

Ts4 1

⇒ QModel 2 (4,4) < var max s = ≈ 0,0033 2 (4.47)

2 τ torque 48 τ 2

s

torque

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:

Ts2 α max

2

1

⇒ QModel 1 (4,4) < ≈ 0,013 2 (4.49)

3 s

acceleration equation. The variance of the acceleration in system Model 2 equals the

variance of the inertia related load. The load torque is generally unknown. However,

an estimation of the variance top-limit can be obtained by assuming a maximum

torque-inertia relation of the drive. The calculation is based on the considerations

made in chapter 4: A constant relation of Tmax/J = 1000 s-2 is assumed. Considering

Sensorless Speed Control of Induction Motor Drives 73

the number of pole pairs p and a maximum torque variation time constant

τload = 2 ms, the top-limit of the acceleration inaccuracy and of the process variance

is:

T − Tload ,k −1 p Tmax Ts

model inaccuracy < p load ,k ≈ (4.50)

J max J τ torque

2

1 p Tmax Ts

≈ 1482 1

⇒ Q(6,6) < (4.51)

12 J τ torque

s4

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.

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.

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 µs, respectively 700 µs, 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 µs.

Only a few extra calculations are necessary compared to the speed observer based on

Model 1 requiring a turnaround time of 167 µs. The used sample time is set to

Ts =220 µs. 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.

Summations [ ] Multiplications [ ] the entire control [µs]

Model 1 546 662 >400

Matrix calc.

Model 2 881 1026 >700

Matrix calc.

Model 1 207 254 167

optimized

Model 2 255 320 187

optimized

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.

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 ω 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

Sensorless Speed Control of Induction Motor Drives 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

2

n [rpm]

1000

500

1

0

-0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

t [s]

15

Tel

10

T [Nm]

0

Tload

-5

-0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

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.

600

With αl

estimation

n [rpm]

400

Without αl

200

estimation

0

0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05

t [s]

1050

n [rpm]

1000

950

900

0.05 0.1 0.15 0.2 0.25 0.3

t [s]

Bottom: Details indicated by Box 2.

algorithm as feedback to control the motor. With the other algorithm in the control

loop, the obtained results are almost the same but with a higher overshoot due to the

delayed response of speed and current controller. A simultaneous real-time

implementation of both algorithms requires a faster DSP. Nevertheless, the

simulation has the advantage of calculating the real motor speed without any delay

in contrast to a real-time implementation using a filter for the measured speed signal

(see also figure 5.7). The estimated speed signal requires no additional filter. The

smoothness and the transient performance of the signals are adjustable by the noise

covariance matrix Q of the EKF algorithm [Ter 01].

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.

requirement of computation time. Figure 4.6 shows the closed loop observer

integrated into a simplified field-oriented speed control loop of an induction motor

drive. The voltages required as input for the EKF are either measured or obtained

from the reference voltages. Here, they are calculated regarding the inverter non-

linearity as explained in chapter 3. The current is measured in two phases. As

mentioned earlier, the current should not be additionally filtered. Pre-filtering

decreases the performance of the proposed observer. The current measurement

should be offset-free as the Kalman filter assumes a zero mean value of the error. An

offset generates erroneous estimations, especially at very low motor speed.

Power

supply

Udc

sin(γ)

Udc

Voltage uα cos(γ) 2

calculation uβ sin(γ)

id id,error Udc

cos(γ) * SVM

iq iq,error uα* ua

Inverter

*

ub

ia iα iα Speed ud* ud* SVM *

iµ id* uc

& uq* uq* uβ*

ib iβ ωr PWM

iβ Flux iq* ∆ud* generation

Control decoupling ∆u *

3⇒2 EKF

Speed

q

inverse

reference

current

control Park Trans.

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.

Sensorless Speed Control of Induction Motor Drives 77

A 1,5 kW induction motor has been used to verify the applied approach. The inertia

of the whole drive system (motor and load machine) is about 0,008 kgm2. The used

load machine is a dc motor drive with constant excitation. At high motor speed, the

dc motor is coupled with a resistor bench. The experiments at low motor speed are

done with the dc motor supplied by a thyristor converter. The load machine can be

controlled in either torque or speed control mode.

All presented results are obtained with the second model in the loop. Figure 4.7

shows the experimental results of a speed reversal using the estimation of speed,

rotor flux and flux angle as feedback to control the motor. Additionally, the real

speed is measured and compared. It can be seen that there is a very good accordance

between real and estimated speed, without any steady state error. During transients,

the estimation of the speed is even faster thus better than the measured one, because

a filter is used for the speed measurement causing a delay of the signal (incremental

encoder with 1024 lines, cut-off frequency of the used speed filter ≈ 1 kHz).

1000

500

n [rpm]

-500

-1000

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

t [s]

50

25

∆ n [rpm]

-25

∆n = nest - nm

-50

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

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

1550

n [rpm]

1500

nest

1450

1400

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

t [s]

15

6

[Nm]

10

i [A]

load

q

5

2

T

0 0

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

t [s] t [s]

estimated speed nest. Bottom: q-axis current and load torque estimation.

50

25

nm

n [rpm]

-25

nest

-50

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

t [s]

25

n [rpm]

-25

0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1

t [s]

50

nest

25

n [rpm]

-25

nref nm

-50

0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 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.

Sensorless Speed Control of Induction Motor Drives 79

[A]

dc

2

I 0

-2

0 0.5 1 1.5 2 2.5 3

t [s]

60

40

nest

20

n [rpm]

0

-20

-40

nm

-60

0 0.5 1 1.5 2 2.5 3

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.

8

6

[A]

4

2

dc

I

0

-2

0 0.5 1 1.5 2 2.5 3 3.5 4

t [s]

120

n [rpm]

60

0

0 0.5 1 1.5 2 2.5 3 3.5 4

15 t [s]

[Nm]

10

load

5

T

0

0 0.5 1 1.5 2 2.5 3 3.5 4

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

3000 300

nest

2500 250

nref

|U + U | [V]

2000 200

nm

n [rpm]

β

1500 150

1000 100

Flux

weakening

α

500 50

0 0

0 0.05 0. 1 0.15 0.2 0.25 0 0.05 0.1 0.15 0.2 0.25

t [s] t [s]

3.5 300

3

Flux 200

weakening 100

2.5

u [V]

i [A]

0

2

β

µ

-100

1.5 -200

1 -300

0 0.05 0.1 0.15 0.2 0.25 -300 -200 -100 0 100 200 300

t [s] u [V]

α

Figure 4.12: Behavior at high motor speed and flux weakening. Top: Speed response

and applied voltage amplitude. Bottom: Magnetizing current and applied voltage.

The used motor model, as well as the implemented EKF, contains four electrical

motor parameters: stator inductance Ls, stator resistance Rs, rotor time constant τ2

and leakage (Blondel) coefficient σ. All other parameters, as e.g. the stator time

constant τ1, are linked to these parameters. Obviously, the quality of the speed

estimation in the observer depends on the accuracy with which the motor parameters

are known. Inaccurate model parameters lead to misalignment of the field-oriented

coordinate system, impairing the dynamic performance of the drive. Possibly more

important is the steady-state accuracy of the speed control, being poor with detuned

model parameters.

If the machine operates under no-load conditions, the relevant parameters are stator

resistance and stator self-inductance. Particularly at low motor speed, the speed

estimation is sensitive to an inaccurate stator resistance value in the observer model.

Also the leakage inductance value, being the decisive parameter at high motor

speed, should be properly tuned to the actual leakage inductance of the machine. The

implemented real-time adaptation of these parameters is based on monitoring of

magnetizing and d-axis current in steady state. Assuming a constant rotor flux,

equation (4.3) can be simplified:

diµ

ε = id − i µ = τ 2 =0 (4.52)

dt

flux controller of the field-oriented control system. However, the error ε becomes

non-zero at a mismatch of stator resistance and inductance respectively. As can be

Sensorless Speed Control of Induction Motor Drives 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.

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 ε is used as a feedback signal to adapt

accurately stator resistance at low supply frequencies (|ωµ| < 5 Hz) and inductance at

high motor speed and supply frequencies (|ωµ| ≥ 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

3.6

id

3.4

i [A]

3.2

iµ

3

2.8

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

t [s]

6

5

Real Rs

R [Ω ]

4

s

3

Estimated Rs

2

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 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.

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 steady-

state error can be used as an adaptation watchdog or, alternatively to the proposed

algorithm, as a parameter-correcting feedback signal.

3.5

5 iµ

id 3

i [A]

i [A]

4

iµ 2.5

3 id

2

0 0.25 0.5 0.75 1 1.25 1.5 0 0.25 0.5 0.75 1 1.25 1.5

t [s] t [s]

550

350

500 Parameter adaptation

L [mH]

L [mH]

300

switched ON

450

250 Parameter adaptation

s

s

400 switched ON

200

350

0 0.25 0.5 0.75 1 1.25 1.5 0 0.25 0.5 0.75 1 1.25 1.5

t [s] t [s]

1050 1002

nm 1000

n [rpm]

n [rpm]

1025

nest 998

1000

996

nref nm

975 994

0 0.25 0.5 0.75 1 1.25 1.5 0 0.25 0.5 0.75 1 1.25 1.5

t [s] 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 σ on

the observer performance is very low and hardly measurably. Furthermore, a

mismatch is partially compensated by the inductance adaptation. Therefore, an

adaptation of σ is not implemented.

No major problem exists in determining the stator frequency ωµ. To the contrary, the

estimation error of the rotor frequency ωslip, directly reflected in the accuracy of the

rotor speed estimation ωr, 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

Sensorless Speed Control of Induction Motor Drives 83

scheme of the rotor time constant compensating the temperature variations has not

yet been realized. A simulation scheme based on the feedback of the observer state

error ∆x (figure 4.2) turned out to malfunction in practice, since the obtained signals

are too low and noisy to carry suitable information.

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 speed-

accuracy 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.

5 450

4 400

L [mH]

i [A]

3 350

s

2 300

1 250

0 10 20 30 40 50 0 10 20 30 40 50

t [s] t [s]

1010 450

nm

1005 400

L [mH]

n [rpm]

1000 350

s

995 300

nest

990 250

0 10 20 30 40 50 1 2 3 4 5

t [s] iµ [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 high-

performance 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.

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.

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. Sensorless Speed Control of PMSM

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.

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 real-

time to track motor parameter variations.

the rotor. The resulting back-EMF voltage induced in each stator phase winding

during rotation can be modeled quite accurately as a sinusoidal waveform. A

mathematical model describing the PMSM motor dynamics in a rotor flux reference

frame is well known [Jah 86], [Hen 92]. The electrical properties of the PMSM in

continuous time are completely described by two stator voltage equations:

did

u d = Rs id + Ld − ω r Lq iq (5.1)

dt

diq

u q = Rs iq + Lq + ω r Ld id + ω r ΨMd (5.2)

dt

simply, since the instantaneous electromagnetic torque can be expressed similarly to

that of the dc machine as the product of q-axis current iq and magnet flux ΨMd. In

case of interior permanent magnets, the additional reluctance torque can be

exploited:

( (

Tel = p iq ΨMd − Lq − Ld id ) ) (5.3)

In contrary to the induction machine, the flux angle γ rotates synchronously with the

rotor speed. With the same simplifications as introduced in the induction motor

study, the mechanical equation is:

Sensorless Speed Control of PMSM 87

dω J dω r

Tel − Tload = J = (5.4)

dt p dt

dγ

= ωr (5.5)

dt

the discrete time state space is equivalent to:

R Lq u

id ,k +1 ≈ 1 − Ts s id + Ts ω r iq + Ts d (5.6)

L d Ld Ld

R L Ψ uq

iq ,k +1 ≈ 1 − Ts s iq − Ts ω r d id − Ts Md + Ts (5.7)

Lq Lq Lq Lq

ω r ,k +1 ≈ ω r + Ts

p2

J

( ( ) ) p

iq ΨMd − Lq − Ld id − Ts Tload

J

(5.8)

γ k +1 ≈ γ + Ts ω r (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

ΨMd. 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 ΨMd.

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:

dα l d p

= Tload ≈ 0 (5.10)

dt dt J

p

⇒ α l ,k +1 ≈ α l = Tload (5.11)

J

88 Chapter 6

The dynamic model for the PMSM, choosing d- and q-axis current id, iq, the

electrical rotor speed ωr, rotor position γ and the acceleration αl 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.

id

iq

U s

x k +1 = A k x k + B k u k ; u k = αs ; x k = ω r ; (5.12)

U β

k γ

αl k

T Lq

1 − Rs s ω r Ts 0 0 0

Ld Ld

Ld T ΨMd

− ω r Ts 1 − Rs s − Ts 0 0

Lq Lq Lq

Ak = (5.13)

Ts p 2

0

J

( (

ΨMd − Lq − Ld id ) ) 1 0 − Ts

0 0 Ts 1 0

0 0 0 0 1

Ts Ts

cos (γ ) sin (γ )

L

d L d

Ts Ts

− sin (γ ) cos (γ )

B k = Lq Lq (5.14)

0 0

0 0

0 0

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.

Iˆ s cos (γ ) − sin (γ ) 0 0 0

y k = ˆαs = C k x k = x (5.15)

Iβ

sin (γ ) cos (γ ) 0 0 0 k

The model matrices Bk and Ck depend on the position of the rotor γ, the matrix Ak

on d-axis current id, flux linkage ΨMd and rotor speed ωr. 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

Sensorless Speed Control of PMSM 89

|γ | < π. 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.

According to the EKF algorithm described in chapter 5, all equations can be written

as a function of a system vector Φ and an output vector h describing the re-

linearized model of the PMSM. The derivatives of the output and the transposed

system vector are:

= =

∂xk ∂xk sin (γ ) cos (γ ) 0 id cos(γ ) − iq sin (γ ) 0

(5.16)

cos (γ ) − sin (γ ) 0 − iˆβ 0

=

sin (γ ) cos (γ ) 0 iα

ˆ 0

∂ (A k x k + B k u k )

T T

∂Φ k

=

∂x = ∂xk

k

− Ts p 2

1 − R s s

T L

− ω r Ts d ( )

Lq − Ld iq 0 0

Ld Lq J

L Ts p 2

ω r Ts q

T

1 − Rs s ( ( ) )

ΨMd − Lq − Ld id 0 0 (5.17)

Ld Lq J

L T

Ts q iq − s (Ld id + ΨMd ) 1 Ts 0

Ld Lq

T T

s uq − s ud 0 1 0

Ld Lq

0 0 − Ts 0 1

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

R = U dc 10 −5 − 10 A (5.18)

0 1,15 V 0 1,7

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 µs and the parameter of

the 3 kW PMSM into account:

2

1 T ∆u

Q(1,1) = s = 0,021 A 2 (5.19)

3 Ld

2

1 T ∆u

Q(2,2) = s = 0,0059 A 2 (5.20)

3 Lq

1

Q (3,3) = 2 Ts4 ⋅1010 s −6 = 3,2 ⋅10 −5 (5.21)

s2

1 1

Q(5,5) = 1,2 k p 2 Ts2 ⋅108 6

= k ⋅ 43,2 4 , with: k ≤ 1 (5.23)

s s

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-per-

ampere operation, being nearly equivalent to maximum drive efficiency [Bose 97].

Sensorless Speed Control of PMSM 91

ω*

ω⇔Θ ω* Tel*

1 − z −1 Control mode Kn

Ts / K pn ω* ω Proportional |Tel|<Tmax

gain

Θ* Kpp

J

Position Speed controller with Tˆload = α̂ l

Position Θ controller load torque rejection

p

reference

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.

sin(γ)

ib ia iα iα cos(γ)

iβ sin(γ)

ia ib iβ id id,error cos(γ)

uα*

*

iq iq,error ua

3⇒2 *

ub

id* ud* ud* SVM *

Udc AC voltage uα uc

αl Speed iq* uq* uq* uβ*

calculation uβ ωr

PWM

Control ∆ud* generation

A/D

EKF & decoupling ∆uq*

inverse 2

model

Speed current Park Trans. Udc

reference control

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 µs. 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 real-

time motion systems.

The proposed speed sensorless control scheme has been tested using a 3 kW, 4 kW

and 45 kW PMSM (data are given in appendix B). However, to keep the presented

results clear, all experimental results presented in this chapter are measurements

using the 3 kW prototype motor. Some additional experiments regarding the 4 kW

motor, specially designed for PV-powered water pump systems, are presented in

chapter 7.

Since the DSP power is simultaneously used for monitoring purposes and recording

experimental data, the sample time used is fixed to Ts =200 µs. A dc generator with

constant excitation coupled to a variable resistor bench is used to load the PMSM.

Via a power switch a load step can be applied. Additionally, the load torque is

measured by a torque transducer. All experimental results and measurements are

carried out using the estimated states as feedback to a speed controller with load

torque rejection (figures 5.1 and 5.2). The bandwidth of both current controllers

using also the estimated values of d- and q-axis current is about 950 Hz. The

bandwidth of the current loop is not decreased by using the EKF instead of the field-

oriented control with position measurement.

Figure 5.3 shows the experimental results of a speed reversal using the estimated

speed and position as feedback. Additionally, the real speed and position are

measured for comparison. There is a very good agreement between real and

estimated speed and position respectively. Using the information on generated

electromagnetic torque and drive acceleration, the noise as well as the lag in the

estimated speed signal is even lower than the measured and filtered speed signals

during transients.

Furthermore, figure 5.3 exhibits the influence of signal lag due to data transmission.

Without any delay, the required phase voltages, calculated by means of the voltage

references controlling the inverter, are directly available in the control loop. To the

contrary, the affiliated current response is measured not before the next sample

period of the digital control system. Neglecting the current signal lag causes a poor

estimation of the position angle. Therefore, an extra sample delay is added in the

loop of the estimated phase voltages used in the observer algorithm.

The observer presented achieves the objective of eliminating the lag of the estimated

motor speed by additional estimation of the load. Figure 5.4 presents the response of

Sensorless Speed Control of PMSM 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.

500

250

n [rpm]

0

nref

-250

-500 (a)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

t [s]

20

15

10

i*q iq

i [A]

5

q

0

-5 (b)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

t [s]

20

10

∆ n [rpm]

-10 ∆n = nest - nm

-20 (c)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

t [s]

0.2

0.1

∆ γ [rad]

-0.1

Without extra delay With extra delay (d)

-0.2

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

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

1020

nm

n [rpm]

1000

980

nest

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

t [s]

20

15

10

i [A]

5

q

0

-5

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

t [s]

15

10 Tel

T [Nm]

5

Tm

0

Tload

-5

Tel

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

t [s]

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:

1 n , with: |n| ≤ 300 rpm (5.24)

= id* + 3A sin( 2π 100 t ) ⋅ 1 −

s 300 rpm

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

Sensorless Speed Control of PMSM 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.

( ( ) ) ( ( ) )

!

Tel* = p iq* ΨMd − Lq − Ld id* = p iq ,ref ΨMd − Lq − Ld id ,ref (5.25)

⇒ iq ,ref = iq*

ΨMd − Lq − Ld id* ( ) *

= i q 1 +

itest

(5.26)

(

ΨMd − Lq − Ld id ,ref )

ΨMd

− id ,ref

Lq − Ld

20

15

i [A]

10

q

0

0 0.2 0.4 0.6 0.8 1 1.2 1.4

t [s]

5

id

n = 300 rpm

0

i [A]

d

-5

n = 1000 rpm

-10

0 0.2 0.4 0.6 0.8 1 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

800

control

id = 0

n [rpm]

600

400

200

0

0 0.2 0.4 0.6 0.8 1 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.

find the absolute rotor position as indicated by an encoder index pulse. This start-up

procedure is necessary in both position control and speed or torque control mode. A

torque is only produced if the excitation is precisely synchronized with the rotor

speed and instantaneous position. A start-up strategy is executed by impressing a

(assumed) q-axis current and slowly increasing the initial assumption of the rotor

angle until the motor rotates and the index pulse is found. However, the motor has to

rotate up to one mechanical revolution. Once the index is found, all registers are

reset and the drive is ready for normal operation.

initial rotor position. If the rotor position can not be exactly estimated, the starting

torque of the motor decreases and the motor may temporarily rotate in the wrong

direction after start. A starting strategy often proposed is based on energizing two

windings by a large armature current (about rated current) and expecting the rotor to

align with a certain definite position. This method yields the direction of the magnet

axis but cannot distinguish between North and South Pole.

To the contrary, the presented sensorless control scheme is self-starting. Here, the

variation of the inductance as a function of the rotor position is used to obtain the

position. Due to the low permeability of the magnet material, the inductance along

the q-axis of the PMSM with interior permanent magnets is larger than the

inductance along the d-axis. Impressing the test signal (5.24), the difference is

detected by the Kalman algorithm and the estimated position converges

automatically to the real position. Figure 5.7 presents the initial start-up of the

digital control system. Aligning the initial value of the estimated position to the

Sensorless Speed Control of PMSM 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 steady-

state 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

γm

2 0

γm

γ [rad]

γ [rad]

0 -0.2

-2 -0.4

γest γest

-4

0 0.5 1 1.5 0 0.5 1 1.5

t [s] t [s]

100 10

nm

50 5

nest

n [rpm]

n [rpm]

0 0

-50 -5

nm nest

-100 -10

0 0.5 1 1.5 0 0.5 1 1.5

t [s] t [s]

Figure 5.7: Start-up of the sensorless speed control. Top: Estimated position γest and measured positionγm.

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

2

γref

γ [rad] 1

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

t [s]

0.1

0.05

∆ γ [rad]

-0.05

∆γ = γest - γm

-0.1

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

t [s]

200

nref

150

nest nm

n [rpm]

100

50

0

-50

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

t [s]

Figure 5.8: Position control of the 3 kW PMSM. Top: Position reference γref, estimated position γest and

measured position γm. Middle: Difference between estimated and measured position. Bottom: Speed

reference nref, estimated speed nest and measured speed nm.

Sensorless Speed Control of PMSM 99

The motor model of the PMSM as well as the implemented EKF contains four

electrical motor parameters: d/q-axis inductances Ld and Lq, stator resistance Rs and

permanent magnet flux linkage ΨMd. For the given PMSM, the ohmic voltage drop

is very small. Thus, the influence of a stator resistance variation is very low and

hardly measurably. The key mechanical drive parameter is the moment of inertia J.

A mismatch of the inertia affects the observer performance only during transients

and causes no steady-state error. Therefore, an adaptation of Rs and J is not

considered.

The most influential motor parameter, affecting the steady-state error and the

observer performance, is the permanent magnet flux linkage ΨMd. Applying flux

adaptation, the torque-current mapping via a look-up table according to figure 2.12

is no longer suitable. Assuming exact knowledge of the motor parameters and using

the d-axis current, the torque reference Tel*, determined by the speed controller, is

transformed to a q-axis current reference iq*:

Tel*

iq* = (5.27)

p (ΨMd − ( Lq − Ld ) id )

According to (2.50), the optimum torque control of the PMSM yields two solutions

for the d-axis current reference id*:

2

ΨMd ΨMd

i =

*

± + iq* 2 (5.28)

d

2 ( Lq − Ld )

2 ( Lq − Ld )

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.

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

adaptation must be disabled at steps of the speed reference to avoid erroneous flux

calculation during transients.

∆ω KΨ ∆ΨMd

∆ΨMd

sign( ω*) ∆Ψ Md = ∫ KΨ ∆ ω dt Flux

error

ω* Tel*

Kn

ω

|Tel| < Tmax

Tload Speed control and

α

J/p flux adaptation

i*q

ΨMd p

x1

Initial value

x2 x12 + x22 i*d

id

Lq-Ld 2

Initial value

Current mapping

with load torque rejection and modified current mapping (Ld ≤ Lq).

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.

0.31 0.26

0.3 0.25

estimated

[Vs]

[Vs]

0.29 0.24

flux linkage real flux

0.28 0.23

linkage

Md

Md

0.27 0.22

Ψ

Ψ

0.26 0.21

0.25 0.2

0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

t [s] t [s]

10 10

Parameter adaptation Parameter adaptation

5 switched ON 5 switched ON

∆ n [rpm]

∆ n [rpm]

0 0

-5 -5

∆n = nest - nm ∆n = nest - nref

-10 -10

0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3

t [s] 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.

Sensorless Speed Control of PMSM 101

Considering PMSM’s 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 ΨMd [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

10

5

i [A]

0

d

-5

-10

-15

0 5 10 15 20 25 30 35 40 45 50 55 60

t [s]

0.27

[Vs]

0.26

Md

0.25

Ψ

0.24

0 5 10 15 20 25 30 35 40 45 50 55 60

t [s]

4

∆n = nest - nm

2

∆ n [rpm]

-2

∆n = nest - nref

-4

0 5 10 15 20 25 30 35 40 45 50 55 60

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.

adaptation to compensate also for a slight mismatch of the q-axis inductance Lq as

well as for a load-dependent variation/saturation. Figure 5.12 demonstrates the flux

adaptation at variable load torque, id* = 0 A and a motor speed n = 1000 rpm. Again,

the coincidence of estimated and measured speed shows the validity of the approach.

Thus, all motor parameters, except for the flux linkage, are set constant. The

influence of parameter variations is compensated by flux adaptation. The approach

of setting the inductances of the given PMSM constant is also verified by numerical

calculations [Pah 98] and measurements [Van 98]. However, a mismatch of motor

parameters is not arbitrary. Approximate values, guaranteeing the stable operation of

the observer, are also required for exact tuning of the current controller.

102 Chapter 6

15

10

i [A]

q 5

0

0 5 10 15 20 25 30 35 40 45 50 55 60

t [s]

0.265

[Vs]

0.26

Md

0.255

Ψ

0.25

0 5 10 15 20 25 30 35 40 45 50 55 60

t [s]

4

∆n = nest - nm

2

∆ n [rpm]

-2

∆n = nest - nref

-4

0 5 10 15 20 25 30 35 40 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.

Sensorless Speed Control of PMSM 103

torque-current mapping. Therefore, a real-time flux adaptation scheme, tracking

motor parameter variations, has been implemented. All proposed control approaches

are verified by experimental results.

6. PV-Powered Water Pump Systems

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

[Mul 97], [Dus 92] for this kind of systems turned out to malfunction, when tested

under realistic conditions. Due to the lack of storage in the dc bus, the power of the

PV array must be used immediately to accelerate an ac motor. 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. Therefore, a novel MPPT

algorithm, realized by feeding back the dc voltage and current to a controller, has

been implemented. The entire system is controlled by a digital signal processor

(DSP) based developing platform realizing MPPT, dc bus voltage control,

speed/torque control of the drive and start-up and shut down automatism's.

Considering realistic conditions, advantages and drawbacks of the different control

units are discussed. Measured results are presented and evaluated to demonstrate the

performance and the stability of the system developed here.

The system, experimentally installed both at the K.U. Leuven and in industry, is a

PV powered water pump system, consisting of a PV array, a low cost inverter, a

permanent magnet synchronous motor (PMSM) and a water pump with water

storage. The PV array has a peak power of 4,32 kW. To avoid a power supply by an

electric grid, the system has been set up to work independently in island operation.

All control and measurement units are supplied by the dc bus between inverter and

PV array. A block diagram of the pilot installation is shown in figure 6.1. The

inverter operates as a variable frequency source (PWM) for the PMSM driving the

pump. Since a PMSM in open loop is unstable (see section 7.4.1), a field-oriented

control with feedback of speed and position is proposed. However, a mechanical

speed/position sensor has several drawbacks from the viewpoint of drive cost,

reliability and signal noise immunity. Especially in submerged-motor/pump systems,

an installation of the additionally sensor is problematic or even impossible. Here,

speed and field position are estimated by an extended Kalman filter described in

chapter 6.

Solar water

generator inverter PMSM pump storage

Phase current

for EKF

PWM MPP-

Tracking control

Current [A]

Voltage [V]

prototyping

DC bus voltage/current

measurement for MPPT

Ride-Through at Power Interruptions 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.

recording the experimental data. Provisionally, the entire control algorithm, safety-

related monitoring and the start-up and shut down automatism’s 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

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.

9 135

MPP

8 120

7 105

current

6 90

[W]

5 75

[A]

power

PV

PV

4 60

P

I

3 45

2 30

1 15

0

0 5 10 15 20 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 Number of

modules in modules in Imax [A] Umax [V]

series parallel

6 6 44,7 129

9 4 29,8 193,5

12 3 22,35 258

18 2 14,9 387

Surface applications for irrigation systems are mostly driven by dc machines while

for installations in the drilling holes submersible induction motor/pump systems are

used. Commutator motors have very desirable control characteristics, but they are

not applicable for submersible installations. Furthermore, their use is limited by a

number of factors [Bose 97]:

• Need for regular maintenance of the commutator;

• Relatively heavy rotor with a high inertia;

• Difficulty in producing a totally enclosed motor as required for some

hazardous (e.g. submersible motor/pump system) applications;

• Relatively high cost;

reliability and maintenance-free operation is important [Bhat 87]. However, small

Ride-Through at Power Interruptions 109

efficiency especially at partial load. Thus, motor selection and design theory

[Hen 96] were limited to a permanent magnet synchronous motor (PMSM) coupled

to centrifugal or submersible pumps.

and pump. Due to this hazardous application, placing additionally sensors, e.g.

encoder for speed and position measurement, is costly and problematic or even

impossible. Therefore, the first approach of the PV powered water pump system was

an open loop control of a PMSM with a damper cage (figure 6.3). Rotor bars have

been implemented in order start up the motor and to balance disturbances.

Copper bars

magnets

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 ω ω 2 (6.1)

where Kω 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.

This U/f ratio is pre-determined for every (steady state) motor speed, choosing a

voltage level corresponding to the lowest motor current. The voltage-current

characteristic at different load torque levels is presented for the studied PMSM in

figure 6.4. The calculations are made at a frequency of 10 Hz corresponding to a

motor speed of n = 300 rpm. If all motor parameters and the load characteristic are

known, the optimal U/f ratio can be calculated for every motor speed. The current is

settled automatically depending on the difference between induced (EMK) and

supply voltage. However, a direct control of the current is impossible proposing this

approach.

Considering figure 6.4, a slight variation of the voltage can easily lead to a very high

over-current of the motor with the risk of demagnetization. In fact, an erroneous

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

Torque

I [A]

[0->10Nm]

10

0

10 11 12 13 14 15 16 17

U [V]

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

ϑ = ϑ0 + ∆ϑ (6.2)

J d 2 ∆ϑ

+ Tmax ∆ϑ cos ϑ0 = 0 (6.3)

p dt 2

∆ϑ = sin( 2π f e t ) (6.4)

1 Tmax cosϑ0

⇒ fe = (6.5)

2π J p

The frequency fe of the PMSM used varies between 0 Hz < fe <10 Hz for maximum

load (ϑ0 = 90°) and no load (ϑ0 = 0°) respectively. This oscillation creates a

Ride-Through at Power Interruptions 111

electromagnetic instability of the drive. In order to soften the system, the PMSM

should be equipped with a damper cage [Hen 91]. The resulting damper time

constant Tdamp can be calculated using:

2 J ω12 '

Tdamp = R2 (1 + σ 1 ) 2 (6.6)

3 p 2 U 12

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.

[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.

Most PV powered water pump systems consist of two different control units. The

first is a dc unit with or without a battery as energy storage. In this unit, the MPPT is

controlled by varying the duty ratio of a dc-dc converter. Using this converter leads

to a less complicated control algorithm for the MPPT. Varying the dc bus voltage

can be done more quickly and without changing the power or frequency of the

motor. The influence of a changing irradiance level during a searching procedure is

reduced. On the other hand, this converter introduces many losses, amongst others:

• 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

Udc

IPV Tel*

Udc ua* SVM

Torque * Inverter

START-UP PI with Tel* u

Control b

& -1 anti windup

Udc*

*

Tel &

MPPT uc*

ω* |ω|<ωmax EKF

Voltage

anti windup anti windup

control

AC

pump

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.

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,

Ride-Through at Power Interruptions 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

Reference

2000

n [rpm]

1500

Measurement

1000

500

0

0 0.25 0.5 0.75 1 1.25 1.5

t [s]

30

20

iq

i [A]

10

id

0

-10

0 0.25 0.5 0.75 1 1.25 1.5

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 open-

circuit voltage guaranteeing a productive motor speed, being higher than the

minimum speed of the shut down automatism.

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 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

200

U dc [V]

Reference

150

Measurement

100

0 0.2 0.4 0.6 0.8 1

t [s]

30

15

i [A]

0

q

-15

-30

0 0.2 0.4 0.6 0.8 1

t [s]

900

n [rpm]

800

700

600

0 0.2 0.4 0.6 0.8 1

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.

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 ½ m (=½ bar) to 10 m (=10 bar) and back to ½ m. The

Ride-Through at Power Interruptions 115

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.

200 200

Reference

195 195

[V]

[V]

190 190

dc

dc

U

U

185 185

180 180

0 1 2 3 4 5 0 1 2 3 4 5

t [s] t [s]

1500 1500

1400 1400

n [rpm]

n [rpm]

1300 1300

1200 1200

1100 1100

0 1 2 3 4 5 0 1 2 3 4 5

t [s] t [s]

Left: ½ m ⇒ 10 m. Right: 10 m ⇒ ½ m.

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 dc-

link capacitor recharging cycle. The experimental results of such short time three-

phase power interruptions are shown in section 8.5.

maximum power point tracking (MPPT). Compared to a common voltage tracking,

the efficiency of the output power of a PV array can be increased about 2 % by

MPP-Tracking. The MPP is characterized by the voltage, where the PV array

generates maximum output power.

The main problems of matching the MPP with a PV array as power supply are

related to the non-linear, solar irradiance and cell temperature-dependent voltage and

current characteristics of the PV array. The characteristics are affected by the

contamination, the sunlight irradiance and the cell temperature. To reach the MPP at

rising irradiance level, the current in the dc bus must be increased while the dc bus

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.

initial values

Calculate Uopt

• Optimum voltage: Uopt

• Search range: ∆U

• Search speed: dU/dt

Calculate ∆U

Decrease Uref

NO

Uref < Uopt - ∆U

Uref

YES

Reference

Increase Uref

voltage Uref

Uref

NO Uref

Uref > Uopt + ∆U

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.

Ride-Through at Power Interruptions 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.

Figures 6.10-6.12 demonstrate the start-up procedure and the automatic operation

mode of the entire control system consisting of MPPT, voltage control and torque

control. Figure 6.10 shows the power generated by the PV array and the speed of the

motor driving the pump during 5 min of MPPT, while figure 6.11 exemplifies the dc

bus voltage for the same span of time. The characteristic of figure 6.12 indicated by

“Start 1” shows the mentioned power as a function of the dc voltage.

4

3

[kW]

2

PV

P

0

0 50 100 150 200 250 300

t [s]

2000

1500

n [rpm]

1000

500

0

0 50 100 150 200 250 300

t [s]

Top: PV power. Bottom: Motor/pump speed.

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

of reference voltage is very small and almost constant.

240

220

Udc [V]

Box 1

200

180

0 50 100 150 200 250 300

t [s]

195

Udc,ref

194

Udc [V]

193

192

191

Udc

190

265 270 275 280 285 290 295

t [s]

Top: dc bus voltage. Bottom: Details indicated by Box 1.

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.

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.

Ride-Through at Power Interruptions 119

3.5

MPP

3

2.5

2

[kW]

Artificial Start 2

PV

P

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).

voltage control are also suitable for an induction motor driving the pump. The

performance of induction motor and PMSM are similar, only the induction motor

efficiency is lower especially at partial load. Here, in contrast to all other earlier

presented results, a PV array with a peak power of 1,2 kW and a 1,5 kW induction

motor driving the centrifugal pump was used. The MPPT during a cloudy day, with

this second installation, is presented in figure 6.13. Four different starting conditions

(a-d) are shown. Characteristic ‘c’ and ‘d’ starts in an artificial operation point. The

characteristic ‘a’ is situated in a higher voltage range, because it shows the first

searching procedure at a lower cell temperature.

250

MPP

200

d

150

[W]

c

PV

b

P

100

50

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.

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

source inverter supplying both squirrel-cage induction motors and permanent

magnet synchronous motors. Basic control techniques, that allow dynamic torque

and flux control in a decoupled way, are direct torque control (DTC) and field-

oriented control (FOC). Summarizing, the DTC provides a better dynamic torque

response whereas the FOC provides a better steady-state behavior. With respect to

the planned applications where the motor speed is the main control variable, the

FOC has been chosen as final control scheme.

torque producing components of the supply currents. In particular, the concept of

field orientation and the resulting ability to directly control the electromagnetic

torque were discussed in chapter 2. Torque control, which constitutes the most basic

motor control function, maps very directly into current control because of the close

association between current and torque generation in any PMSM and induction

motor drive.

There are many excellent books on the topics of electrical machines and drives.

However, it is believed that the present thesis is novel in many respects. The basic

FOC-scheme is refined systematically adding additional features step by step. Flux

weakening is widely known in literature. Less appreciated is the ability to operate

the induction motor above the nominal flux at low speed to enhance the torque per

ampere relation and thus better utilize the available power supply current. The

approach has been further refined by flux optimization. Contrary to the assertions in

literature, this feature makes the induction motor superior compared to the PMSM in

a wide operation range when efficiency is considered especially in the range where

iron losses are dominant. The choice of a suitable flux control strategy depends on

the respective application. In the realized implementations, it can be switched over

easily and in real-time to different strategies. Considering flux weakening of the

PMSM, the algorithms presented in literature are based on pre-calculations

postulating a constant dc bus voltage. In this work, the dc bus voltage is variable

over a wide range requiring an alternative approach. Therefore, an automatic flux

adaptation scheme has been implemented.

122 Chapter 9

within the current controller is neglected since the maximum voltage is limited by

the power inverter itself. However, the presented current control with anti-windup is

essential considering the dc bus voltage variable over a wide range.

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. Chapter 3 presents the collaboration 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 measurement distortion/identification due to the inverter

non-linearity are discussed in detail.

encoder as measurement device. Among the speed, also rotor position and the

acceleration of the drive are estimated. The implemented algorithm is based on a

linear Kalman Filter. The discussion extends to the implementation of an advanced

speed control loop. It has been shown that this approach offers a significant

improvement of the entire drive performance. This chapter can be also regarded as a

smart introduction into observer theory. Advanced observer theory has been applied

to approaches eliminating the need of position/speed measurement.

The sensorless speed control of both permanent magnet synchronous motor and

squirrel-cage induction motor drives, which is nowadays the most attractive research

area of electrical motor drives, is presented in chapter 5 and 6. New models for

speed estimation are proposed. The structures of the implemented sensorless control

schemes are based on the extended Kalman filter theory. The approach requires no

additional measurements. The terminal voltages are not measured; they are

reconstructed by using the monitored dc bus voltage and the switching functions of

the inverter considering non-linearities due to the dead time of the power switches.

Among the speed, also rotor flux, flux position and the acceleration of the drive are

estimated. The speed estimation does not lag the actual motor speed, both in steady

state and during acceleration/braking. Compared to sensorless control schemes

described in literature, the experimental results have shown to offer a significant

improvement of the drive performance.

were developed and implemented. Special care has been taken for the viability of the

real-time implementation: A comprehensive and clear description of controller

design and affiliated parameter calculation is given for all treated applications.

Furthermore, all proposed control schemes were verified by experimental results.

shaft sensor is described in chapter 7. This system reflects one application

employing many of the drive features, designed and implemented in this work. New

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 high-

performance 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 ride-

through 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.

Using the TMS320C31 DSP providing 60 MFlops, the proposed algorithms have

been realized only by means of costly code optimizations. The limit of the possible

code and memory size has been reached. The calculation of the closed loop current

transfer function has shown the large influence of delays within the loop as e.g.:

measurement filter, sample time, PWM frequency and signal lag of data

transmission. Further increasing the program size will lead to execution times,

which are no longer suitable for high-performance motion control.

It is expected that the dynamic performance, especially of the torque control loop,

can be vastly increased. Presently, a new DSP development platform based on TI’s

most recent processor-generation, the TMS320C6711 DSP providing 1000 MFlops,

is under construction within the ELECTA group. This new development platform

provides a system, which will be capable of implementing even more extended and

computation time intensive algorithms.

Considering a more powerful control system, there are many applications possible,

e.g.: The noise covariance matrices within the mentioned sensorless speed control

system can be adapted in real-time dependent on the given operating point. This can

be done by e.g. another extended Kalman filter or artificial intelligence.

Various reluctance motors will have an increased role in the future. An expansion of

the proposed sensorless control schemes to these motor types forms surely an

interesting task.

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, time-

invariance 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.

- Power Inverter Hard Ware DesignUploaded byapi-3695801
- Tesla Model s (Electric Motor Specifications)Uploaded byseyyidMubeen
- Current Control Techniques for Three-phase Voltage-source PWM ConvertersUploaded byedvin_s667
- Speed control of induction motorUploaded bykunalpahari
- Modelling, simulation and analysis of low-cost direct torque control of PMSM using hall-effect seUploaded byarunsgr
- DC/AC Pure Sine Ware InverterUploaded byTrongTanLe
- Current Source InverterUploaded byjp-sharma
- Final Project ReportUploaded byShubhendra Singh
- Full Control of a PWM DC AC Converter for AC Voltage RegulationUploaded byapi-27465568
- Voltage Source Inverter_2Uploaded bydhamu6699
- Dstatcom Project ReportUploaded byMahendar Mahe
- Sinusoidal Pulse Width ModulationUploaded byabdurrahim_t
- CSI vs VSIUploaded byManikandan Durairaj
- Electrical Machines and DrivesUploaded byanon-427532
- 36-1 FIELD ORIENTED CONTROL OF A MULTILEVEL PWM INVERTER FED INDUCTION MOTORUploaded byapi-19508046
- IEC 61727.pdfUploaded bymiguelin226
- FULL REPORTUploaded byapi-3756989
- PSIM TutorialUploaded bybaohuy_pla
- Ansoft Maxwell 3D v11 UserguideUploaded byredoctober24
- dvr projectUploaded byRavi Morla
- Design and Construction of an InverterUploaded byAdeniji Olusegun
- Modelling of IM Using Dq TransformationUploaded bydivinelight
- Electric Vehicles - Modelling and SimulationsUploaded byJosé Ramírez
- 2 kVA Inverter ProjectUploaded bySalisu Webmaster
- Thesis PmsmUploaded bymolay_ee
- Motor Design Suite V12Uploaded byphieuxuatkho
- MODELING_AND_SIMULATION_OF_PERMANENT_MAGNET_SYNCHRONOUS_MOTOR_DRIVE_SYSTEMUploaded bySlimane Souag
- power quality improvement with DSTATCOMUploaded byprachi_shrivas
- A New Cascaded Multilevel Inverter WithUploaded byStephen Snyder
- Three-Phase Squirrel-Cage Induction Motor Drive Analysis Using LabVIEWUploaded byCarlos Roberto Amaya Rodriguez

- WatchPower User ManualUploaded by142777
- ups spec.pdfUploaded byGoutam Acharjee
- FUJI-FRENIC-5000G11S-P11S-User-Manual.pdfUploaded byKi Sey
- Selective_Harmonic_Elimination_in_Multil.pdfUploaded byAnonymous Xp6yLzD
- A Novel Overmodulation Technique for Space-Vector Pwm InvertersUploaded byacer.hemanth6952
- edriveqbanksem7 (1)Uploaded byalisha27
- PV Protection DCUploaded byPaul Moreno
- Design of FPGA Based Solar Power InverterUploaded byAbhilasha Mathur
- 123Uploaded byapi-3753995
- St2712w LaboratoriumUploaded byAnkit Desai
- Vantage LincolnUploaded byKevin Diestra Aliaga
- G120_CU240BE2_BA6_0414_eng.pdfUploaded byRebecca Patel
- Conversol Off Grid EV 1-5KVA Manual 20140327Uploaded byCaraiane Catalin
- A Hybrid Ac-dc Microgrid and Its ControlUploaded byTouseef Hussain
- 1100w Sharp 1 Cf MicrowaveUploaded bywissam.alnaser6332
- A Review of Variable Speed Wind Turbine System Employing DFIGUploaded byIJRT Online
- Analysis of Voltage Control for a Self-excited Induction Generator Using a Current-controlled VolUploaded byapi-27465568
- Advantage of Haier DC Inverter 180 Sine WaveUploaded byJuan Cásares
- Liebert-GXT3-5000VA-10000VA-SLI-23185_REV3-05-12Uploaded byHoàng Minh
- Codigos de Error Lg InverterUploaded bycipaarzu
- Mathworks.com GTO Based STATCOMUploaded byEngr Imtiaz Hussain Gilani
- Doctoral 2006 MultilevelUploaded byVenkateshChelladurai
- IRJET-Speed Control of Brushless DC Motor Using Fuzzy Based ControllersUploaded byIRJET Journal
- AC SneilheirdUploaded byFeby Choirul Umawardi
- Speed Control of Induction MotorUploaded byعمار سهيلي
- Portable Solar GeneratorUploaded byFarizal Jaafar
- Gutor_PxW_Q410_053_AUploaded bycedar1015
- Emergency PowerUploaded bysathish
- eecrs2009Uploaded byjaykay7069
- LM201W01-SLA1-LG.pdfUploaded byuim