Analysis of field-oriented controlled induction motor drives under sensor faults and an overview of...

Analysis of field-oriented controlled induction motor drives undersensor faults and an overview of sensorless schemes

Arun Dominic Dn, Thanga Raj Chelliah 1

Department of Water Resources Development and Management, Indian Institute of Technology, IIT-Roorkee, Roorkee, Uttarakand 247667, India

a r t i c l e i n f o

Article history:Received 28 November 2013Received in revised form14 April 2014Accepted 29 April 2014This paper was recommendedfor publication by Didier Theilliol

Keywords:Field-oriented controlInduction motor drivesReviewMine hoist loadSensorless control

a b s t r a c t

To obtain high dynamic performance on induction motor drives (IMD), variable voltage and variablefrequency operation has to be performed by measuring speed of rotation and stator currents throughsensors and fed back them to the controllers. When the sensors are undergone a fault, the stability ofcontrol system, may be designed for an industrial process, is disturbed. This paper studies the negativeeffects on a 12.5 hp induction motor drives when the field oriented control system is subjected to sensorfaults. To illustrate the importance of this study mine hoist load diagram is considered as shaft load ofthe tested machine. The methods to recover the system from sensor faults are discussed. In addition, thevarious speed sensorless schemes are reviewed comprehensively.

& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

1. Introduction

The induction motors consume major parts of electrical energyin any process industries. Control system employed in thesemotors plays a vital role in such industries where the speed hasto be varied as per the requirements. The rotor speed of theinduction motor can be varied by the following three ways [1–5]:

(a) By changing the number of poles,(b) by changing the stator supply voltage with fixed frequency

(VVFF), suitable to fan and pump loads, and(c) by changing the stator supply voltage with variable frequency

(VVVF), suitable to all types of industrial loads

The first one is achieved by designing the motor in such a waythat it works on two speed levels and this motor is called Pole-Amplitude-Modulated Induction motor [1,5]. It does not requirethe frequency converters and it can be easily implemented cost-effectively. However, this method is operated at two speeds ofcertain speed ratio and requires more care in the winding design.

The second method is based on varying the supply voltage toIMD. In 1960s, when thyristors was invented [6] and suggestedvarious topologies like a back to back thyristor configuration in

each phase, control of output voltage from a constant source wasachieved by delaying the conduction angle of thyristors. A pair ofthyristors connected back to back in each phases of a starconnected stator induction motor produce an acceptable steadystate performance (with energy conservation) for fan and pumptypes of loads. But the use of same topology in delta connectedinduction machine produced average torque per rms value ofstator current an inferior value [7]. Also the third order harmonicsgenerated by the motor gets circulated in the windings and causesheating effect. As a result, parameter variations occurred in themotor which leads to more heating effect. Furthermore, thevariable voltage applied to the machine is non-sinusoidal in naturewhich produces time-harmonic currents, increases the heatingeffect of the machine. Hence the variable voltage is normallyapplied to low and medium capacity pumps of power range 5–150 hp in which torque is proportional to the square of the motorspeed [7].

The third method is based on varying the supply frequency.For the variable voltage and variable frequency operation, a pulsewidth modulation (PWM) technique should be used for themotion control of electrical drives. The modulation techniquescan be of sinusoidal pulse width modulation (SPWM) or spacevector modulation for generating the PWM Pulses in digital form.Microcontrollers or digital signal processors (DSPs) or field pro-grammable gate array (FPGA) or dSPACE can be used to generatetriggering pulses based on the algorithms designed. These pulsesignals are compatible to power electronic switches present in thePWM inverter. The PWM inverter is connected to the stator

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/isatrans

ISA Transactions

http://dx.doi.org/10.1016/j.isatra.2014.04.0080019-0578/& 2014 ISA. Published by Elsevier Ltd. All rights reserved.

n Corresponding author. Tel.: þ91 7579079205, þ91 7500325877.E-mail addresses: [email protected],

[email protected] (D. Arun Dominic).1 Tel.: þ91 1332 285554.

Please cite this article as: Arun Dominic D, Chelliah TR. Analysis of field-oriented controlled induction motor drives under sensor faultsand an overview of sensorless schemes. ISA Transactions (2014), http://dx.doi.org/10.1016/j.isatra.2014.04.008i

ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

www.sciencedirect.com/science/journal/00190578

www.elsevier.com/locate/isatrans

http://dx.doi.org/10.1016/j.isatra.2014.04.008



mailto:[email protected]

mailto:[email protected]





terminals of the induction motor. By varying the modulation index(amplitude of reference wave to its corresponding carrier wave)the pulses are varied to get variable frequency.

The main contribution of this paper is to analyze the perfor-mance of vector controlled IMD under sensor faults with theconsideration of mine hoist load diagram. And then the paperreviews comprehensively the sensorless schemes including posi-tion estimation, stator current estimation, and fault detection andisolation techniques. In addition, stator current reconstructionfrom DC link current is also discussed whereas Ref. [27] (publishedrecently in this area) reviewed speed estimation techniques only.The paper is organized as follows: Section 2 discusses the differentmethods of speed control of Induction motor drives, Section 3analyzes the IMD under sensor faults, Section 4 reviews speedsensorless schemes, Section 5 discusses the current sensor faultdetection and isolation (FDI) techniques, Section 6 reviews phasecurrent reconstruction techniques from DC link current and finallythe paper concludes at Section 7.

2. Speed control techniques for induction motor drives

As discussed earlier, the speed control techniques of IMD canbe broadly classified into constant frequency drives and variablefrequency drives with further classifications as shown in Fig. 1.Variable frequency drives usually offer good dynamic performancewhich can be broadly classified into two: (i) scalar control and (ii)vector control.

2.1. Scalar control method

The air gap flux is kept constant by varying the correspondingvoltage in proportional to the frequency (keeping their ratio

constant for torque constant) as given by the linear relationship[1,2,9]

Eg ¼ 4:44KwNΦagf ð1Þ

where Kw is windage factor, N is the number of stator turns, Φag isthe air gap flux, Eg is the induced emf, f is the supply frequency, Kw

and N are constants. Eq. (1) can be rewritten as follows:

EgKgf

¼Φag ð2Þ

where Kg ¼ 4:44KwN is a constant.From Eq. (2), it is clear that the supply frequency can be varied

in linear proportion to the induced emf to maintain air gap flux asconstant [10]. When the load varies from no load to full load, theair gap flux has to be varied in linear proportion of voltage andfrequency. It is possible up to the rated voltage as windinginsulation deteriorated at over-voltages. There are two types ofconventional scalar control methods depending on control loops i.e. open loop V/f control and closed loop V/f control.

(i). Open loop V/f control:The open loop V/f control is shown in Fig. 2. A variablefrequency PWM inverter is connected to the stator circuit ofthe motor. The control variables are voltage and the frequency.In Fig. 2, reference voltage is generated from the frequencycommand value G so that the V/f ratio is maintained constant.These voltage and variable frequencies are generated byneglecting the stator voltage drop. However at low frequencyoperation the stator resistance drop cannot be neglectedsince, most of the stator voltage is absorbed by the statorwinding resistances. This voltage drop has to be compensatedby a boost voltage [3] and its value should be high at the timeof starting and can be reduced to a lower value once the motor

Methods with Constant Frequency

Speed Control on Induction Motor Drives

Methods with variable frequency

PoleChanging

VariableVoltage

Scalar orV/f

Control

Vector orField

OrientedControl(FOC)

SensorlessVectorcontrol

IndirectFOC

DirectFOC

Estimationof rotorSpeed

IndirectMethod

DirectMethod

Machine ModellingEquations

High Frequencysignal Injection

AdaptiveControl Method

ExtendedKalman Filter

Method

Model ReferenceAdaptive System

(MRAS)

Sliding ModeControl Method

(SMC)

Current SensorFDI for twosensor Fault

Observerbased FDI

ParityEquation

Based FDI

Phase CurrentReconstruction

Current SensorFDI for only one

sensor Fault

Estimationof StatorCurrent

Fig. 1. General classification of speed control of induction motor drives and its estimation techniques.

D Arun Dominic, T.R. Chelliah / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎2





is reach to its steady state value. In Ref [8], a solution for widerange of variation of flux in scalar control of induction motordrives with boost in voltage is proposed. The control loop has afrequency compensation that gives constant slip frequencyover the entire speed range. The voltage is varied dynamicallyand hence the dilemma in choosing the boost voltage isavoided. An automatic boost in voltage is reported in [11] inwhich the boost in voltage depends on the machine para-meters. For a low speed of 200 rpm there is a torque pulsationdue to the parameter variations. Hence the open loop V/fcontrol method is usually not applied to an application whereprecise speed control is necessary. Since the speed measure-ment is not taken as a feedback, steady state speed error ispresent more when load changes.In addition, open loop V/f control produces less torque andunstable oscillations at low speed operations [12]. When full loadis applied with rated stator voltage, the speed of the inductionmotor is reduced to a higher value or in other words the steadystate error is increased. Even when the frequency is varied thelinear characteristic is not obtained. This frequency drop iscompensated by a frequency compensation control [12] and itis designed by considering the linearized induction motor model.When the V/f ratio is kept constant irrespective of changes in theloads, the maximum torque is decreased and the voltage drop inthe stator terminals becomes dominant. To avoid this voltagedrop the voltage at no load condition was set to little more thanthe rated value.However this control depends on the machine parameters likerotor time constant andmechanical time constant like moment ofinertia and friction co-efficient. To overcome this problem amethod is proposed in [13] which compensated the stator voltagedrop vectorially, both in magnitude and phase of the voltage.It requires the value of stator resistance that can be measuredby offline. The frequency is compensated by estimation of airgap power which is based on the non linear speed –torquecharacteristic curve.

(ii). Closed loop V/f control:Closed loop operation has a speed control loop and the errorin speed is processed by a PI controller for generating slip

speed command [3,14]. The closed loop V/f speed control isshown in Fig. 3. This closed loop V/f control method is thesame as the open loop speed control by giving the voltage andspeed signal as the inputs to the PWM inverter. The speedcontrol loop is a closed loop and the slip speed is addedto the rotor speed to generate the frequency command ωe

n.The voltage generated by a function generator is added duringlow frequency operation for stator voltage drop compensa-tion. The output from the PWM Inverter, variable voltagevariable frequency, is given to the induction motor. The steadystate response of the drive with this control is goodwhereas the dynamic response of the drive is not good. Tohave enhanced dynamic response the speed as well theflux have to be controlled. For this control, both speed andflux has to be fed back and the field orientation concept wasinvented [15].

2.2. Vector control of induction motor drives

The vector control or the field oriented control was firstinvented by Blaschke [15] in the year 1972. In the field orientedcontrol, the rotor flux (ψr) is aligned with the rotating referenceframe that is rotating at the synchronous speed, as shown in Fig. 4.α and β are the stationary reference frame coordinates and d and qare the rotating reference frame coordinates. θ is the angle formedbetween the rotating reference flux vector and the stationaryreference frame axis. The block diagram of field oriented control isshown in Fig. 5 which has two control loops: (i) inner current orflux control and (ii) outer speed control.

The inner current control loop is used to regulate the fluxproducing component of the stator current (direct axis compo-nent) and the outer current control loop is used to regulate thetorque producing component (quadrature axis component) of thestator current. The inner current control loop limits the currentflow in the converter and motor within the safe limit. The outerspeed control loop regulates the speed. When the load is increasedthe shaft speed decreases and therefore a positive error Δωm

occurred, which indicates the speed needs to be increased. The

G

Integrator

PWMThree Phase

PWM Inverter

Diode BridgeRectifier

InductionMotor

V s*’

Vs*

Θ*e

Va*

Vc*

Vb*

3-PhaseSupply

ω*e

+

+

Vs/ωe

+

C

Vo

Vs

ωe

Vo

L

Fig. 2. Open loop V/f speed control [3].

Three PhasePWM

Inverter

DiodeBridge

Rectifier

InductionMotor

3-Phase Supply

+

C

PIController

ω r*

ω r

+ -

ω r

ωsl

++

Vs

ωe

L

Fig. 3. Closed loop V/f control of induction motor drives [3].

D Arun Dominic, T.R. Chelliah / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3





inner loop increases the flux producing component of the statorcurrent and the drive accelerates till the speed reaches to thedesired speed. When desired speed is achieved, that means themotor torque is equal to the load torque, there is no error in thespeed which indicate that there is no need of any moreacceleration.

Since the stator and rotor fluxes are coupled together, control-ling the phase angle of the rotation (θ) of induction motor drivesresults the motor acts like a fully compensated (θ¼901) separatelyexcited DC machine where the flux can be controlled indepen-dently. Based on the control of angle of rotation the vector controlcan be classified into two types namely Direct Field Oriented

Control (DFO) and Indirect Field Oriented Control (IFOC) dependson the unit vector (cos θ, sin θ) generation.

(i). Direct field oriented control (DFOC):The DFOC the angle of rotation is obtained as

θe ¼ tan �1 Ψ qr

Ψ dr

� �ð3Þ

where Ψ dr and Ψ qr are the d and q axis flux linkage which canbe obtained from the voltage model [3]

Vqs ¼ 23Va�1

3Vb�

13Vc ¼ 1

3ðVabþVacÞ ð4Þ

Vds ¼ � 1ffiffiffi3

p Vbþ1ffiffiffi3

p Vc ¼ � 1ffiffiffi3

p Vbc ð5Þ

Iqs ¼23Ia�

13Ib�

13Ic ¼ Ia ð6Þ

Ids ¼ � 1ffiffiffi3

p Ibþ1ffiffiffi3

p Ic ¼1ffiffiffi3

p ðIaþ2IbÞ ð7Þ

Ψ ds ¼Z

Vds�RsIdsð Þdt ð8Þ

Ψ qs ¼Z

Vqs�RsIqs� �

dt ð9Þ

Ψ dr ¼ LrIdrþLmIds ð10ÞΨ qr ¼ LrIqrþLmIqs ð11Þwhere Vds, Vqs, Ids, and Iqs are the corresponding d and q axisvoltages and currents of stator. Idr, Iqr are the corresponding dand q axis currents in the rotor circuit and Ψ ds; Ψ qs are thecorresponding flux linkages of the stator. In the DFOC methodFig. 4. Rotor flux orientation.

3-PhasePWM

Inverter

InductionMotor

SVMPWM

dqotoα,β

a, b, cto

α,β

α,βto

dqo

PI

PI

V

V

sdref

sqref

Vsαref

Vsβref

ψref

-

-+θ

iaibic

Isα

Isβ

Isd

Isq

ClarkTransform

InversePark

Transform

ParkTransform

ω ref PI+

-

+

3 Φ DiodeBridge

Rectifier

+C

L

3-Phase Supply

IDC LinkCurrent

CurrentSensors

SpeedEncoder

α,βto

a, b, c

InverseClark

Transform

-

ωr

Fig. 5. Vector control of induction motor drives.






the stator flux is derived from the basic equations of thevoltage model. However, the flux estimation is complicated atlow frequency operation. The next method called indirectvector control is performed to overcome from this problem.

(ii). Indirect field oriented control (IFOC):The angle of rotation is generated in a feed forward mannerwhich is obtained by adding slip frequency and the rotorfrequency [3,16] and the integration process is absent here.

θe ¼Z

ωedt ¼Z

ðωrþωslÞ ¼ θrþθsl ð12Þ

ωsl ¼LmRr

ψ rLrIqs ¼

Lmψ rTr

Iqs ð13Þ

Te ¼32

� �P2

� �LmLr

� �ψ r Iqs ð14Þ

where ωsl is the slip frequency and ωr is the rotor angularfrequency, Tr¼Lr/Rr is the rotor time constant.

3. Induction motor drives under sensor faults

The VVVF control applied in IMD successfully and received agood dynamic response [1–3]. As discussed in the previoussection, closed loop operation of IMD is possible by measuringcurrents and speed with the help of suitable sensors. There ispossibility to decrease the reliability of the drive when sensorsused [17–19]. The current sensor may be faulted when suddenchange in stator currents. The current sensor faults can be of opencircuit fault, omission faults, gain, bias, constant, noise and alsoover current faults. The speed sensors are more fragile since thesesensors are in contact with the shaft rotor. The speed sensor faultsare omission faults, gain constant fault and open circuit fault. Thissection analyses the simulation results obtained from a vectorcontrolled IMD serving to mine hoist load.

Fig. 6. shows a mine hoist load diagram which has threedifferent operating regions as marked. The machine is loaded with1.5 pu, 0.75 pu and 0.14 pu at region 1, region 2, and region 3,respectively. A 12.5 hp induction motor with the above loaddiagram is considered to study the negative effects of sensor faults.

Fig. 7 shows that the simulation results of the said vectorcontrolled IMD under healthy sensors and the obtained results(speed, torque and stator currents) are acceptable. Fig. 8 shows thesimulation results of IMD under speed sensor faults. Fig. 9 showsthe simulation results of IMD under current sensor faults.

It is noted on the speed profile (shown in Fig. 8a) that the rotorspeed is decreased from 0.5 pu to 0.09 pu when a speed sensorfault (open circuit) has occurred at region 1. During the fault speed

error is maximum as the feedback signal given to the controller iszero. Once the steady state speed error is very high the controllerincreases the reference current of Iqs which results more statorcurrent (2.0 pu) as shown in Fig. 8d. When a current sensor faulthas occurred, there is an oscillation in the motor torque and itoscillates between 1.2 pu to 2 pu as shown in Fig. 9b. In addition,stator currents profile is also disturbed. As a conclusion, IMD isfailed to produce the desired torque and speed profiles undersensor faults. The increment in stator current in mine hoist loadunder faults would create voltage drop in the lines with significantpower losses. This would affect the other machines connected inthe same bus. As mine hoist motors are rated even up to 2000 hpin practice the negative effects are more than obtained. Also theDC link current profile is analyzed under t3 region for both speedand current sensor faults. DC link current is increased from 0.2 puto 2 pu as shown in Fig. 10d under speed sensor fault whereas thisvalue is 0.2 pu to 4 pu in current sensor fault, as shown in Fig. 11d.This information is much helpful to decide the capacitor rating inDC link to withstand over currents caused by sensor faults.

4. Overview of speed sensorless schemes

4.1. Position estimation in sensorless vector control of inductionmotor drives

In the initial development stages of the research on motioncontrol of electrical drives, the speed of the ac motors aremeasured by a tachometer. The use of this speed measuringinstrument finds difficult for placing and holding in the rotor shaftof the induction motor. Due to this problem the speed is measuredby the help of sensors. The sensors can be an optical encoder typeor tachogenerators or rotational transducers. These sensors aremore fragile when mounted on the shaft of induction motor and itcauses less reliability in the overall system. So the concept ofsensorless control of ac motors was developed. When the induc-tion motor is loaded drooping in the speed takes place whichresults the rotor cuts more flux. As a result more emf is inducedwhich causes disturbance or harmonics in the induced emf. Theconcept of slot harmonics [20,21] was developed where harmonicsproduced by the rotor flux wave are studied. The air gap fluxdeveloped in the three phase induction motor is sinusoidal innature due to the rotating magnetic flux (RMF). These RMF whenpassing through the conductor present in the rotor slots causesharmonic distortions in the air gap flux due to the variation in thepermeance of the rotor slots. This slot harmonics are directlyrelated to the speed of the rotor. Since the slot harmonic wave ispresent along with air gap flux wave it is difficult to extract thespeed information. One method to extract the speed information isby adding a flux ripple component in the resulting air gap fluxwave so that the frequency of the slot harmonic wave is increasedwhich can be easily isolated by a Low Pass Filter (LPF) to extractthe speed information. But induction motors are designed withskewed rotor slots to reduce the harmonics produced by the rotorflux wave and adding flux ripple component increases the har-monics present in the resulting air gap flux wave particularly thirdorder harmonic value is increased in the resulting air gap fluxwave. So the method of inserting the flux ripple is not possible inthe isolation of slot harmonics. The other method of isolating theslot harmonics is by adding the individual voltages present in eachcoils in each slot and in each phase. To eliminate the third orderharmonics a switched Capacitor Filter is used. The cut offfrequency of the filter is adjusted by regulating the switchingfrequency of the capacitors present in the filter. This frequency iscaptured by Voltage Controlled Oscillator (VCO) and the frequencyof VCO is the frequency of the induced voltage. When theFig. 6. Mine hoist load diagram.






frequency of the slot harmonics is near to the frequency of theVCO the resulting frequency will be passed through the LPF andadjusts the output of the VCO. If the frequency of the slotharmonics and the frequency of the VCO are same the frequencyis locked. This method holds good only for applications wherehigh dynamic performance is not required. It consumes signifi-cant time as the frequency spectral analysis is needed for slotharmonics. Direct and indirect methods [22,23] can be utilized

for the sensorless speed estimation on IMD (as shown previouslyin Fig. 1).

(i). Model based estimation:The stator current of IMD in arbitrary reference frame isdetermined by the estimated stator currents [25] as:

iαs ¼ ias ð15Þ

0 5 10 15-0.5

0

0.5

1

1.5

Time (sec)

Spe

ed (p

u)

6 6.02 6.04 6.06 6.08 6.1-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Time(sec)

Sta

tor C

urre

nts

(pu)

1.001 1.0015 1.002 1.0025 1.003 1.0035 1.004 1.0045 1.005-1

-0.5

0

0.5

1

Time (sec)

DC

Lin

k C

urre

nt (p

u)

1 1.02 1.04 1.06 1.08 1.1-1.5

-1

-0.5

0

0.5

1

1.5

Time(sec)

Sta

tor C

urre

nts

(pu)

0 5 10 15-3

-2

-1

0

1

2

3

4

Time(sec)

Torq

ue (p

u)

11 11.02 11.04 11.06 11.08 11.1-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

Time(sec)

Sta

tor C

urre

nts

(pu)

0 5 10 15-3

-2

-1

0

1

2

3

4

5

Time(sec)Id

s &

Iqs

in (p

u)

IdsIqs

12 12.0002 12.0004 12.0006 12.0008 12.001-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

Time(sec)

DC

Lin

k C

urre

nt (p

u)

Fig. 7. Performance of induction motor drives under healthy conditions (no sensor fault). (a) Speed waveform. (b) Motor torque. (c) Stator currents in region t1. (d) Statorcurrents in t2. (e) Stator currents in t3. (f) Ids & Iqs references. (g) DC link current at t1 region. (h) DC link current at t3 region.






iβs ¼1ffiffiffi3

p ð2ibsþ iasÞ ð16Þ

Vαs ¼ Rsiαsþdψαsdt

ð17Þ

Vβs¼ Rsiβsþdψβs

dtð18Þ

ψαs ¼Z

ðVαs�RsiαsÞdt ð19Þ

ψβs ¼Z

ðVβs�RsiβsÞdt ð20Þ

jψ sj ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiψ2

αsþψ2βsÞ

qð21Þ

Cos θ¼ Ψαs

jΨ sj ð22Þ

Sin θ¼ Ψβs

jΨ sj ð23Þ

where Rs is the stator resistance of the machine and θ isthe stator flux angle with respect to α-axis of the stationaryα–β reference frame.

θ¼ tan �1 Ψβs

jΨαsjð24Þ

The synchronous frequency is obtained by

ωe¼ dθdt

¼ d½ tan �1ðΨβs=ΨαsÞ�dt

¼ ðpψβsÞψαs�ðpψαsÞψβsψ2αsþψ2βs

ð25Þwhere p¼d/dt called the differential operator.Substituting the value of Eqs. (17) and (18) in (25) the value ofestimated speed with the known parameters are obtained as

ωe¼ ðVβs�RsiβsÞΨαs�ðVαs�RsiαsÞψβsψ2αsþψ2βs

ð26Þ

The torque is estimated from the stator flux and stator current as

Te¼ 3P4ðψαsiβs�ΨβsiαsÞ ð27Þ

For the estimation of rotor speed, the slip speed compensation isnecessary which is given by

ωsl ¼ KsTe ð28Þ

0 5 10 15-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

Time(sec)

Spe

ed (p

u)

0 5 10 15-6

-4

-2

0

2

4

6

Time (sec)

Torq

ue (p

u)

0 5 10 150.5

1

1.5

2

2.5

3

3.5

4

4.5

Time(sec)

Ids

& Iq

s (R

ef) i

n pu

Iqs

Ids

3.2 3.21 3.221.5

1.55

1.6

1.65

Time(sec)

Torq

ue (p

u)

2.4 2.6 2.8-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

Time(sec)

Sta

tor V

olta

ge (p

u)

2.6 2.7 2.8 2.9 3-3

-2

-1

0

1

2

3

Time(sec)

Sta

tor C

urre

nts

(pu)

Fig. 8. Performance of induction motor drives under speed sensor faults created at region t1. (a) Rotor speed at t1. (b) Motor torque at t1. (c) Stator current at t1. (d) Referencecurrents Ids and Iqs at t1. (e) Line voltage at stator terminals.






where Ks is the ratio of rated slip frequency/rated torque thatcan be obtained from the name plate details of the machine.The rated rotor speed is given from the estimated parameters isgiven as

ωr ¼ωe�ωsl ð29Þwhere Vαs, Vβs, Iαs, Iβs, P,ωsl,ωr and θ are the stator voltages,currents in stationary reference frame, number of poles, slipspeed, rotor speed angle of rotation of rotor respectively.From the described equations the speed and torque areestimated using the (known) measured parameters like DClink current and flux linkage. PI controllers are used to achievethese estimations.

4.2. Estimation of rotor speed by high frequency signal injection

The other method of speed estimation is achieved from theinjection of high frequency signal from the inverter into the acmotor. A high frequency balanced three phase signal of equalamplitude [24,26] varies between the range of 500–2 kHz isgenerated and the signal is injected into stator windings. Oncethe high frequency signal of same magnitude is injected by theinverter, the modulation in the stator transient reactance dom-inates the stator impedance at high frequencies. The demodulationof the high frequency signal is used to extract the positioninformation. The demodulation signal is based on the heterodyning

process that is based on the addition of the cosine signal andsinusoidal signal that are the function of the saliency angle θΛe

� �and the combined signal from the heterodyning process is sub-jected to low pass filter (LPF) which allows the low frequencysignal in order to avoid the interaction between the three phasecurrents. This concept is called spatial modulation and it is appliedin indirect field oriented control of induction motor drives. Thevariations in the inductance of the rotor occur due to the magneticsaliency. The wide slot opening in the rotor causes more magneticreluctance and decreases the inductance value and the rotor slotwith narrow opening causes less magnetic reluctance andincreases the inductance value [21]. This difference in the induc-tance value is called a local inductance which is a function of rotorposition. The method is clearly discussed in [20–24].

4.3. Estimation of rotor speed by adaptive control method

In adaptive control, observers are used which are insensitive tomachine parameters. The method can be classified into threecategories namely the Extended Kalman Filter method, ModelReference Adaptive Control (MRAS) and Sliding Mode Control.

(i). Extended Kalman filter (EKF) method of speed estimation:The Kalman filter is a full-order stochastic observer used forstate estimation of a non linear dynamic systems [3] in real

0 5 10 15-15

-10

-5

0

5

10

Time(sec)

Torq

ue (p

u)

2.6 2.8 30

1

2

3

4

Time(sec)

0 5 10 15-0.2

0

0.2

0.4

0.6

0.8

1

1.2

Time(sec)

Spe

ed (p

u)

2.5 2.55 2.6 2.65 2.7-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Time (sec)

Sta

tor C

urre

nt (p

u)

2.5 2.6 2.7 2.8 2.9 30

0.2

0.4

0.6

0.8

1

1.2

1.4

Time(sec)

Ids

& Iq

s (R

ef) i

n pu

Ids

Iqs

a b

Torq

ue (p

u)

Fig. 9. Performance of induction motor drives under current sensor faults created at region t1. (a) Rotor speed at t1. (b) Motor torque at t1. (c) Stator current at t1.(d) Reference currents Ids and Iqs at t1.






time. State space model of induction motor is mandatory toimplement this method. State space model of IM [3,27] isfollows:

Vds ¼ RsIdsþddtψds�ωeψ qs ð30Þ

Vqs ¼ RsIqsþ ddtψqsþωeψds ð31Þ

Vdr ¼ RrIdrþddtψdr�ðωe�ωrÞψ qr ¼ 0 ð32Þ

Vqr ¼ RrIqrþddtψ qrþðωe�ωrÞψdr ¼ 0 ð33Þ

ψds ¼ LsIdsþLmIdr ð34Þ

ψ qs ¼ LsIqsþLmIqr ð35Þ

Te ¼ 32

� �P2

� �ðψdsIqs�ψ qsIdsÞ ð36Þ

In state space model the equations can be written as

X�¼ AXþBU ð37Þ

Y ¼ CX ð38Þwhere X¼[X1, X2, X3,……..Xn], X is the system state vector, X1,X2, X3, ……..Xn is the system state variables, A is the systemmatrix of order n X n, B is input column matrix, C is a statematrix, U is the system input, Y is the system output. Thestate variable is considered as X¼[Ids � Iqs �ψdr �ψqr], U¼[Vds �Vqs] and Y¼[Idr � Iqr �ψds �ψqs]. The various symbols usedin these equations are having usual meanings. s¼ 1�

ðL2m=LsLrÞ; τr ¼ Tr ¼ ðLr=RrÞ; s is the leakage co-efficient andτr or Tr is the rotor time constant.

A¼

1s

1τsþ L2m

LsLrτr

� �ih�ωe

LmsLsLrτr

LmsLsLrωm

�ðωe�ωmÞ � 1s

1τsþ L2m

LsLrτr

� �ihLm

sLsLrωmLm

sLsLrτr

�Lmτr 0 � 1

τr ðωe�ωmÞ0 �Lm

τr�ðωe�ωmÞ 1

τr

266666664

377777775

ð39Þ

B¼

� 1sLs 0

0 � 1sLs

0 00 0

266664

377775 ð40Þ

C ¼

�LmLr

0 1Lr

0

0 � LmLr

0 1Lr

sLs 0 LmLr

0

0 sLs 0 LmLr

26666664

37777775

ð41Þ

For speed estimation of induction motor the state spaceequations are modified as

X ðkþ1Þ ¼ AX ðkÞþBU ðkÞþV ðkÞ ð42Þ

Y ðkÞ ¼ CX ðkÞþW ðkÞ ð43Þ

V (k), W (k) are zero mean and white Gaussian noise vectorsof X (k) and Y (k) respectively. This noise and measurementsare given by three covariance matrices namely Q, R and P.The algorithm has two stages namely: (i) prediction stage and

0 5 10 15-1

0

1

2

3

4

5

Time (sec)

spee

d (p

u)

Act.Speed

0 5 10 15-3

-2

-1

0

1

2

3

4

Time(sec)

Torq

ue (p

u)

Act.Torque

13 13.002 13.004 13.006 13.008 13.01-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Time (sec)

stat

or c

urre

nts

(pu)

12.215 12.216 12.217 12.218 12.219 12.22-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Time(sec)

Dc

Link

Cur

rent

(pu)

Fig. 10. Performance of induction motor drives under speed sensor faults created at region t3. (a) Rotor speed at t3. (b) Motor torque at t3. (c) Stator currents at t3. (d) DC linkcurrent at t3.






(ii) filtering stage. In the prediction stage, the next state ispredicted by the machine model and its estimated states.Prediction state covariance matrix (Pn (kþ1)) is calculatedfrom the covariance matrix Q. In the filtering state, the nextestimated states X ðkþ1Þ are obtained from the predictionstate Xn (kþ1) by adding a correction term eK where e¼Y(kþ1)� Y ðkþ1Þand K is the Kalman filter gain. Iterations aredone till the error e value is equal to 0.The EKF has the following steps [3]:

Step1: Initialize the state vector and covariance matrix.Step2: Predict the state vector.Step3: Estimate co-variance matrix P (kþ1).Step4: Compute the Kalman filter gain.Step5: Estimate state vector.Step6: Update error covariance matrix.

The concept of Kalman filter was invented by R. E. Kalman inthe year 1960 [28] for aerospace applications. The methodwas applied to field oriented control of IMD [29] is in the year1993. The flux vector is estimated using the Kaman filteralgorithm. A method of speed estimation of induction motorby Complex Kalman Filter for field oriented control isexplained in [31]. It differs from the conventional speedestimation techniques as voltage equations are not required.Ref. [30] analyzed EKF with parameter uncertainties and theresults shows that inaccuracy present in the estimation whenstep variations in the load torque applied. With the help of

one voltage and one current sensor rotor resistance, speedand air gap flux are estimated. But this method introducesharmonics. A method of speed estimation of induction motorby speed adaptive observer is explained in [32]. But themethod failed to produce an acceptable dynamic perfor-mance at low speed (frequency) as the required time todetect the position is more. In addition, at low frequencyregion the sampling frequency to control the speed is lowthat may lead to poor dynamic performance. To improve thedynamic performance at low speed a method (consideration ofnoises), based on Kalman Filter is proposed in [33]. ExtendedKalman filter (EKF) [34] was introduced for non-linear sys-tems. In EKF, system noise and the output measurementnoises are considered and the experimental results [34] showthat speed control of induction motor produced a desirableperformance in the entire speed region.

(ii). Model reference adaptive systems (MRAS) method of speedestimation:Model reference adaptive system method (MRAS) was evolvedin late 1950s [35]. This method is more effective when poorplant dynamic characteristics and large unpredictable varia-tions in the system. The MRAS system has two models namelyreference model and the adjustable model as shown in Fig. 12.Reference model is the theoretical model with all the para-meters are known whereas parameters have to be estimatedin the adaptive model or adjustable model. u is the input givento both the models. Comparing these two models the errors

0 5 10 15-0.2

0

0.2

0.4

0.6

0.8

1

1.2

Time

spee

d (p

u)

0 5 10 15-10

-8

-6

-4

-2

0

2

4

6

8

Time(sec)To

rque

(pu)

12.14 12.15 12.16 12.17 12.18 12.19 12.2-5

-4

-3

-2

-1

0

1

2

3

4

5

Time(sec)

stat

or c

urre

nts

(pu)

12.188 12.19 12.192 12.194 12.196 12.198 12.2-1

0

1

2

3

4

5

Time(sec)

DC

Lin

k C

urre

nt (p

u)

Fig. 11. Performance of induction motor drives under current sensor faults created at region t3. (a) Rotor speed at t3. (b) Motor torque at t3. (c) Stator currents at t3. (d) DC linkcurrent at t3.






are generated. The errors of the parameters are given tothe adaptation mechanism block which has control lawsto adjust the system parameters in adjustable model. Theconventional MRAS speed adaptation is based on fourparameters like direct-axis, quadrature-axis voltages anddirect-axis, quadrature-axis currents from the motor. Themotor voltages and currents are measured in the station-ary reference frame. But the motor speed is time varyingin nature and expressing the motor equations in station-ary reference frame can lead to computational errors. Amethod for calculating the speed using MRAS is explainedin [36]. The input parameters considered in this methodare direct-axis, quadrature-axis currents and motor modelparameters. Ref. [37] presented a simulation study with apseudo order model of the induction motor as a referencemodel and concluded that higher the value of PI gainsfaster the convergence of speed estimation in IMD. Theaccuracy of the MRAS observer depends on the correctsettling of the machine parameter with estimator andcontroller. The method proposed in [36] has less stabilityunder low frequency conditions. An improved method forestimating the speed at low frequency and at light loads isexplained in [38] where a compensation of stator resis-tance variations has done. Adaptive flux observer pro-posed in [39] is more stable than the other two observerspresented in [36–38]. In order to improve the efficiency ofthe induction motor under light loaded condition byconsidering the core loss into account a method is pro-posed in [40]. The MRAS concept was applied effectivelyto doubly fed induction machine (DFIM) [41,42]. MRASapplied to DFIM is based on the small signal model thathas voltage model as reference and current model asadjustable in which rotor angular position is included. Incomparison with Colin Schauder model of MRAS [36] themethod produced better static and dynamic perfor-mances. But the concept of MRAS is not applied for lowspeed operation. A comparative study of MRAS observerwith Luenberger observer [43] is applied for speed esti-mation of sensorless field oriented control of inductionmotor drives. The simulation results depicts that thehybrid controller (combination of MRAS and Luenbergerobserver)is subjected to parameter variations at lowspeed and the speed torque response at low speed is notsatisfactory.

(iii). Sliding mode control (SMC) method of speed estimation:The sliding mode is an approach in dynamic systems gov-erned by ordinary differential equations with discontinuous

state functions in Eqs. (44) and (45) [44]. For a second orderrelay system the sliding mode equations are written as

€xþa2 _xþa1x¼ u ð44Þ

u¼ �MsignðsÞ ð45Þ

s¼ cxþ _x ð46Þwhen s¼0 then

cxþ _x¼ 0 ð47Þwhere a1, a2, M, and c are constants.Eq. (47) represents the sliding mode equation for a secondorder delay systems. The concept of the sliding modecontrol method applied on electrical drives was firstdiscussed in [45] and for sensorless control of inductionmotors in [46]. SMC is based on the variable structuresystem (VSS) technique which has two steps. The first stepis based on that when the input of the control element isdiscontinuous and approaches to zero the output voltageis kept constant by maintaining high gain. Due to highgains the uncertainties caused in the system is sup-pressed. The second step is based on the decoupling ofthe fluxes. Since the sliding mode trajectories dimensionis less than the motor equation dimension, decoupling ofthe fluxes are easier compared to the other conventionalcontrollers. It can be used for highly non linear systemswith high uncertainty conditions. The robustness of theSMC is less compared to the other controllers due to theoscillations occurred in the system. These oscillations inthe controllers are due to finite frequencies are calledChattering. Due to the chattering effect the pulsation inthe motor torque and heating losses occur. To overcomethese drawbacks a method is proposed in [47] which isbased on the boundary layer control of the controller. Theslope of the boundary layer control must be high toovercome the decoupling effect in the speed controlmethods of IMD. The other drawback of the discrete timeSMC is that the control depends on motor parameters. Butthe motor parameters vary when temperature variationand saturation occurred on the motor. The mismatch in theparameters causes the tracking error between the actual valueand the estimated value which leads to chattering. Thechattering can be suppressed by different ways suggested in[48] i.e. observer based chattering suppression, state depen-dent gain method, equivalent control dependent gain method,hysteresis loop control. In the observer based, chatteringsuppression is done by designing an observer which identifiesthe uncertainties and disturbances. The sliding mode control-ler requires variable switching frequency and the switchingfrequency can be limited within the hysteresis band so thatchattering can be reduced.To maintain the dynamic and robust control performanceunder parameter variations and load disturbances anadaptive uncertainty observer with feed forward controlmethod is proposed in [49]. In this adaptive uncertaintyobserver the robustness of the sensorless speed control ofinduction motor is obtained by adding a term called lumpof uncertainty in the control input. The term 'lump ofuncertainty' is assumed as constant during the course ofobservation of the system parameters. By adding the term'lump of uncertainty' term the tracking error is reduced tozero. The robustness of the controller is checked with 20%of increase and decrease of stator resistance for low speedand high speed operation. However in the speed estima-tion the system uses LPF instead of the integral term in the

AdjustableModel

ReferenceModel

AdaptationMechanism

єU

Fig. 12. Basic structure of MRAS [35].






controller causes instability problem. The instabilityoccurrence is due to DC drift and initial value problems.In order to overcome this problem two sliding mode basedMRAS [50] is proposed. One controller is based on thecontinuous time SMC and the other is based on thediscrete time SMC. The switching frequency is kept asinfinite and the sampling frequency is kept as finite sothat the chattering in the flux waveforms is reduced. Filterneed not to be used when the chattering of the fluxwaveforms are reduced to minimum. If the machineparameters are known then the discrete SMC with highbandwidth of frequency is chosen so that the chattering inthe flux wave form is reduced. If the machine parametersare unknown the continuous time SMC can be used. Inboth the methods by adjusting the gain of the SMCcontroller the chattering is reduced. The results presentedin [50] show that there is no under damped poles or zeroson the right half of the s-plane that improves stability ofthe system. The concept of a robust SMC was introducedin variable speed wind energy conversion systems byBeltran et. al. [51] and the stability was analyzed at lowspeed and high speed regions. SMC based Direct PowerControl of DFIM is discussed in [52] but it needs thecoordinate transformation in synchronous referenceframe [51–54]. Hu et. al. [55] discussed a method toestimate the parameters in stationary reference frame.As a conclusion, SMC offers chattering effect which can bereduced but cannot be eliminated.

(iv). Estimation of rotor speed at zero frequency condition:The speed estimation of rotor at zero frequency or at lowfrequency is obtained by injecting voltage and current signalson the flux axis is proposed [56,57] where the field orientedcontrol of the induction motor is considered. Since thesesignals are not rotating in nature there is no torque ripple. In[57], phase angle of the injected signal is aligned with theestimated d-axis of the rotating rotor. Once the signals areinjected the position has to be estimated by the properdemodulation technique [58] which approximates the flux.Ref. [59] improved the concept of demodulation that afrequency of very low 0.5 (50–100) Hz signals is injected inthe sinusoidal component of the stator current reference andgenerates a sinusoidal magnetizing force which interactswith the main rotating magnetizing field. This modulatesthe saturation level of the machine core and results a newcomponent of zero sequence voltage is generated that has theposition information. An estimation algorithm that estimatesthe speed and stator resistance is developed in [60] at lowfrequency operation.

5. Current sensors fault detection and isolation

A fault in the current sensor can be detected if the sum of thethree phase currents is not equal to zero in a balanced system. Afault in the current sensor can also be determined by the error inthe measured and estimated values. The faults may be of omission,gain, bias and constant [18,19] whereas the fault in the speedsensors are same as that of current sensors but no bias type faultsoccurred. fault detection and isolation (FDI) can be implementedwith the help of measured and estimated currents, residualgeneration, change detection, fault diagnosis and the reconstruc-tion of the current signals. The FDI techniques can be categorizedinto types namely

(A) Observer based FDI.(B) Parity equation based FDI.

5.1. Observer based FDI

The immeasurable signals that occur during the faults can beestimated and reconstructed by an observer [61]. A Luenbergerobserver [61] is used for residual generation (difference in mea-sured and estimated quantities) when a fault is occurred. A controlsystem adapted in the system is switched the feedback signalsfrom the measurement to the observed signal. A bilinear observer[62,63] that separates the systemmatrix into two terms where oneis independent and other is dependent on mechanical speed of therotor. By separating the rotor speed term the non linear term isisolated. Also the Eigen values of the errors are independent ofrotor speed. But it needs two observers namely stator currentobserver and the rotor current observer. Ref. [64] explains the FDIfor heat exchanger process that requires reliable model of obser-vers when fault occurs in the system. When there is a fault in anyone of the observers, the other observer takes the value of theprevious observer (fault observer) and produce incorrect residuals.This method does not consider the machine parameter variationwhich is a major drawback. Also the machine will run in open loopduring the fault. Ref. [66], effective switching pattern proposed toovercome the difficulties faced in [65] but it required threeobservers. A single Lyapunov-based observer is proposed in [67]which is able to estimate two phase currents, speed, and rotorresistance. Another single observer is reported in Ref. [68] con-sidering the input saturation, external disturbances for FDI. Theabove method of observer design [61–67] does not consider thesaturation effects of the sensors. A method is proposed in [69] thatconsidered the saturation effect in the modeling of the observerfor fault sensors. The system states are estimated from thecontroller outputs and are given as feedback to the system. Byconsidering the actuator fault the controller gain and the observergain cannot be determined by pole placement technique since thefeedback controller is not controllable. Based on the linear matrixinequalities (LMI) the controller and observer gains are founded.This method of finding the gains by LMI has the advantage offinding the gains that are fully immeasurable.

5.2. Parity equation based FDI

In this method, the residuals during a fault are generated fromthe parity equations [70]. The decoupled equations are derivedconsists of linear and non linear terms based on the state spaceapproach with unknown linearity. The linear terms indicateknown faults and the non linear terms indicate unknown faults.The linear terms are formed from the known mathematical modelof the system and the non linear term consists of noise and timevarying parameters. In this approach the faults occurred aredescribed in the linear model and the recursive least squaremethod is used to identify the faults from the residual of the FDIalgorithm and hence complex calculations are avoided. The effectsof parameter variations on the observer are overcome by enhancedparity equation [71]. In the enhanced parity equations method, thefaults detection does not depends upon the system model but itdepends on the previous measurements taken from the system.The algorithm memorizes the early measurements and generatesthe residuals. The method is verified experimentally in DC motorcontrol systems [72,73].

The other method of fault detection and isolation (FDI) is basedon SMC [74–76], since the SMC is more robust to parametervariations and can handle more uncertainty and external distur-bance in the system. Also another method is reported in [77] whichis signal based that depends only on the previous measured phasecurrents and compensates the initial current sensor offsets faults. Inthis method the faults are identified by the diagnostic variable of






the measurable parameter and when the values exceed the thresh-old value the sensors are said to be faulted and it has to be replaced.

6. Phase currents reconstruction from DC link current

The concept of three phase currents estimation from the DClink current is first explained by Evans and Cottingham [78]. TheDC link current per phase is obtained by multiplying the phasecurrents of each leg with its switching frequency. By summing theDC link current of each phase, total DC link current value isobtained which is shown in Eq. (48).

idc¼ ∑q

k ¼ 1ikFk ð48Þ

where idc is the DC link current, ik is the current in each of the legsof the sinusoidal PWM inverter and Fk is the switching frequency, qis the number of legs present in the inverter. Substituting the valueof Fk theoretically in Eq. (48), the resultant DC link currentexpression has only harmonic currents. For a three phase inverteronly triplens are present and the sum of other harmonic current iszero. For single phase PWM H-Bridge inverter only even currentharmonics are present. Due to the presence of even orderharmonics in the single phase inverter the output frequency ofthe inverter is twice the frequency across the load. In [79], threephase motor currents are derived from the DC link current. Toderive the motor currents from the DC link current the time periodfor switching pulse must be maintained constant even for higherswitching frequencies like 20 kHz so that the variation in the linecurrent will be very less. To measure the DC link current practi-cally a flux nulling hall-effect sensor or a sense resistor with high-slew-rate amplifier is used. The voltage across the DC linkcapacitor is fed to a low pass filter (LPF) with the cut-off frequencyless than half of the switching frequency that removes the ripplespresent in the switching frequency. Another method to measurethe DC link current is used in [80] where a filter and hold circuitare used.

To avoid the diode recovery current and snubber currents dueto high di/dt and dv/dt in the reconstructed signal a blankingperiod is introduced between the inverter switches. Due toblanking time the dynamic performance of the drive at low speedis poor. The improved dynamic performance at low speed isdiscussed in [81] with the space vector modulation (SVM) techni-que. The blanking time can be compensated by implementing thedouble sided modulation. The double sided modulation is theprinciple of using two voltage vectors switching at switchingperiod. The SVM used at over modulation region [82] to obtainmaximum output voltage from the PWM Inverter. But in overmodulation region the phase current gets distorted. Each andevery sector operation two phase currents can be measured usingtwo current sensors and the third value of current can becalculated from the other value in such a way the sum of thecurrents are zero. Because of dead time effect there may be a shortcircuit in the switches. Few methodologies proposed for undermodulation region (low modulation index) when the currents aredifficult to reconstruct [83]. One disadvantage of the low modula-tion index is the presence of active voltage vectors in SVM forshort time and therefore sampled current cannot be extractedfrom the DC link current. A novel method for the reconstruction ofphase current at low modulation index is explained in [84].Themethod is called the measurement vector insertion method(MVIM) where the pulses are switched on at equal intervals oftime. By doing this the quality of estimation is improved since allthe three phase currents are measured every time whenever theMVIM algorithm is activated. The method offers reduced totalharmonic distortion (THD) but inverter losses and the controller's

disturbance rejection capability (also called dynamic stiffness) aremore due to the added switching events developed by MVIM. Toimprove the dynamic stiffness (DS) a virtual resistance [85] isadded to the controller.

The other method for measurement of DC link current is basedon the current observer [86]. An advantage of using the observer isphase current can be estimated even the modulation index is verylow. The observer is the adaptive phase current observer sincesynchronization of the sampling instant as well as the observerinput values takes place. Three independent adaptive phasecurrent observers are used to identify the information about zerosequence current also. A disadvantage in the observer is thedeviation in the current is more when compared with a singlethree phase adaptive observer and it requires motor parameters.To overcome the disadvantage Curve fitting Luenberger observer isproposed in [87]. In this method the sinusoidal three phasecurrents are estimated.

In practice there is a difference between estimated and themeasured values due to the disturbances and noise in the currentwaveform which results a ripple in estimated current waveform.To avoid the disturbances and current ripple a method called stepby step prediction method [88] is proposed. In this method thecurrent is measured on the DC link for a PWM period. Thephilosophy behind the step by step prediction is the carrierfrequency is injected at the sectors of low modulation index sothat the current is measured for a PWM period. For every halfperiod the actual voltage is the average of voltage injected and theoutput voltage. The switching frequency is higher than thefundamental frequency and hence the current ripple is reduced.However the carrier frequency injection will lead to harmonicdistortion and phase current reconstruction is not possible whenthe dead zones are located near the origin due to the currentoffset present in the reconstructed currents. In this case phasecurrent reconstruction is possible when current sensor onlineoffset compensation [89] is used. It is a method of adding a notchfilter whose cut off frequency is equal to the fundamentalfrequency ωr in the β-axis of the stationary reference frame. Thisnotch filter is cascaded with the PI controller and gives theproportional value of the reconstructed current. But this methoddoes not hold good at zero frequency operation as the operatingfrequency approaches to zero and setting the cut off frequency inthe notch filter is not possible. Also all the three phase currentsare not reconstructed at the same time. At very low modulationindex the reconstruction of three phase current is not possible viaDC link current sensor since the active switching time should begreater than the acquisition time (sum of dead time of inverter,rising time of the inverter and Analog to Digital conversion timeof signals and DC link current settling time). So to overcome thisproblem of three phase reconstruction at very low modulationindex, a hybrid modulation technique (combination of spacevector modulation and PWM modulation technique) is proposedin ref. [90].The authors has proposed a concept of inserting twonon null voltage vectors of equal in magnitude and opposite indirection instead of null vectors in the low modulation range. Byinserting the voltage vectors, active switching time is increased

9%

19%

24%6%

19%

23% EKF

MRAS

SMC

Low Frequency operation

Current Sensor Fault

Phase Current Reconstruction

Fig. 13. Distribution of publications considered for review (sensorless schemes).






than acquisition time, the sectors in the SVM region falls in theactive measurement area and also the SVM provides less harmo-nic distortion. In case when the modulation index falls in theimmeasurable region, the voltage PWM method is used withoutinserting any null voltage vectors for the extension of all threephase current reconstruction. Fig. 13 shows the distribution ofpublication considered for review on sensorless schemes.

7. Conclusions

The paper discussed the dynamic performances of a vectorcontrolled IMD under speed and current sensor faults (open circuitfault). The following conclusion we have obtained from the studyof IMD with mine hoist load diagram:

(a) when speed sensor fault has occurred:(i) The speed profile of the motor is badly disturbed and is

oscillated at around 0.08 pu when 0.5 pu is demanded asper the region 1 of mine hoist load.

(ii) The torque ripple is more due to sudden change in Iqs.(iii) The stator current is raised to nearly double the value of

its original (before fault). This is because of the incrementin Iqs due to high speed error.

(iv) Voltage sag in stator terminals is occurred from 1 pu to0.6 pu.

(b) when current sensor fault has occurred:(i) Slight oscillation in the speed profile.(ii) Torque and stator current profiles affected badly.(iii) Iqs reference oscillates between 0.2 pu to 1.2 pu.(iv) The stator current increases from 1 to 2 pu with more

distortions.(v) DC link current is raised up to 4.0 pu during the fault.

In addition, the paper has reviewed comprehensively the vari-ous sensorless schemes used in IMD including position estimation,stator current estimation and fault detection and isolation techni-ques and stator current reconstruction from DC link current. Theauthors have tried to include as much descriptions of the contentsas possible in order to point out the important and unique aspectsof each scheme. In summary there is a promising approach insensorless schemes for further progress i.e., research on efficientsensorless scheme suitable to low speed operation of IMD willattract academia and industry as very few publications reportedso far.

Appendix

Table A1.

References

[1] Krishnan R. Electric motor drives modeling, analysis and control. India: PHILearning; 2006.

[2] Leonhard W. Control of electrical drives. Berlin, Heidelberg, New York:Springer-Verlag; 1985.

[3] Bimal K. Bose, modern power electronics and AC drives. India: PHI Learning;2010.

[4] Abu-Rub Haitham, Iqbal Atif, Guzinski Jaroslaw. High performance control ofAC drives. West Sussex, United Kingdom: John Wiley & Sons; 2012.

[5] Fong W. Induction-motor speed-changing by pole-amplitude modulation. IEEProc 1958;2597:411–9.

[6] Paice DA. Induction motor speed control by stator voltage control. IEEE TransPower Appar Syst 1968;PAS87:585–90.

[7] Lipo TA. The analaysis of induction motors with voltage controlled bysymmetrical Thyrisors. IEEE Trans Power Appar Syst 1971;PAS90:515–25.

[8] Boys JT, Walton SJ. Scalar control: an alternative AC drive Philosophy. IEE Proc1988;135(3, Part B):151–8.

[9] Tsivitse PJ, Klingshirn EA. Optimum voltage and frequency for polyphaseinduction motors operating with variable frequency power supplies. IEEETrans Ind Gen Appl 1971;IGA7(4):480–7.

[10] Ioannides MG, Papazis SA, Ioannidou FG. Implementation of scalar controlscheme for variable frequency induction motor actuator system. Sens Actua-tors A Phys 2003;106(1–3):306–9.

[11] Gastli A, Matsui N. Stator flux controlled V/f PWM inverter with identificationof IM parameters. IEEE Trans Ind Electron 1992;39(4):334–40.

[12] Koga K, Ueda R, Sonoda T, Sonoda Toshikatsu. Constitution of V/f control forreducing the steady-state speed error to zero in induction motor drive system.IEEE Trans Ind Appl 1992;28(2):463–71.

[13] Garcıa AM, Lipo TA, Novotny DW. A new induction motor V/f control methodcapable of high performance regulation at low speeds. IEEE Trans Ind Appl1998;34(4):813–21.

[14] Upasan A, Kumsuwan Y. Closed loop speed control of induction generator withscalar-control inverters. Energy Procedia 2013;34:371–81.

[15] Blaschke F. The principle of field orientation as applied to the new transvektorclosed-loop control system for rotating-field machines. Siemens Rev 1972;34:217–20.

[16] Guey Z, Pan C. Accurate rotor flux orientation technique for indirect rotor fluxoriented control of induction motor drives. J Chin Inst Eng 2004;27(7):1033–45.

[17] Dominguez-garcia AD, Kassakian JGSchindall JE, Zinchuk, JJ. On the use ofbehavioral model for the integrated performance and reliability evaluation offault-tolerant avionics systems. In: Proceedings of the IEEE conference on 25thdigital avionics systems; 2006. p. 1–14.

[18] MBazzi A, Dominguez-Garcia AD, Krein.PT. A method for impact assessment offaults on the performance of field-orieinted control drives: a first step toreliability modeling. In: Proceedings of the applied power electronics con-ference and exposition; 2010. p. 256–63,.

[19] Bazzi AM, Dominguez-Garcia AD, Krein PT. Markov reliability modelling forinduction motor drives under field – oriented control. IEEE Trans PowerElectron 2012;27(2):534–46.

[20] Zinger DS, Lipo TA, Novotny DW. Using induction motor stator windings toextract speed information. In: Proceedings of the conference record of the IEE,IAS annual meeting; 1989. p. 213–18.

[21] Holtz J. Sensorless position control of induction motors—an emerging tech-nology. IEEE Trans Ind Electron 1998;45(6):840–52.

[22] Holtz J. State of the art of controlled AC drives without speed sensors. Int JElectron 1996;80(2):249–63.

[23] Finch JW, Giaouris D. Controlled AC electrical drives. IEEE Trans Ind Electron2008;55(2):481–91.

[24] Jansen PL, Lorenz RD. Tranducerless position and velocity estimation ininduction and salient AC machines. IEEE Trans Ind Appl 1991;31(2):240–7.

[25] Xue Yanhong, Xu Xingyi, Habetler Thomas G, Divan Deepakraj M. A statorflux-oriented voltage source variable-speed drive based on DC link measure-ment. IEEE Trans Ind Appl 1991;27(5):962–9.

[26] Liu C, Chang H, Wang C. Very low speed sensorless control of induction motordrives using high-frequency signal injection. J Chin Inst Eng 2005;28(6):957–66.

[27] Alsofyani IM, Idris NRN. A review on sensorless techniques for sustainablereliablity and efficient variable frequency drives of induction motors. RenewSustain Energy Rev 2013;24:111–21.

[28] Grewal Mohinder S, Andrews Angus P. Applications of Kalman filtering inaerospace 1960 to the present. IEEE Control Syst Mag 2010:69–78.

[29] Ammar FB, Pietrzak-David M, de Fornel B; Godfroid H. Experimental results ona high power induction motor drive using a field-oriented control based on aKalman filter implemented on a floating point RISC processor. In: Proceedingsof the European conference on power electronics and applications, vol. 6;1993. p. 117–22.

[30] Barut M, Bogosyan S, Gokasan M. An EKF-based estimator for the speedsensorless vector control of induction motors. Electron Power Compon Syst2005;33(7):727–44.

[31] Menaa M, Touhami O, Ibtiouen R, Fadel M. Sensorless direct vector control ofan induction motor. Control Eng Pract 2008;16(1):67–77.

[32] Kubota H, Matsuse K, Nakano T. DSP-based speed adaptive flux observer ofinduction motor. IEEE Trans Ind Appl 1991;31(2):240–7.

Table A1Induction motor specifications and parameters.

Motor Rating 12.5 hp Number of poles 4Motor speed 1450 rpm Voltage 415 V

Rs 0.3 puRr 0.25 puLs 0.0415 puLr 0.0412 puLm 0.0403 puJ 0.1 Kg m2

B 0.02 Kg m2/s



http://refhub.elsevier.com/S0019-0578(14)00087-1/sbref1






































































[33] Kim HW, Sul SK. A new motor speed estimator using Kalman Filter in low-speed range. IEEE Trans Ind Electron 1996;43(4):498–504.

[34] Kim YR, Sul SK, Park MH. Speed sensorless vector control of induction motorusing extended Kalman filter. IEEE Trans Ind Appl 1994;30(5):1225–33.

[35] Landau YD. Adaptive control—the model referencing approach. New York:Larcel Dekker; 1979.

[36] Schauder C. Adaptive speed identification for vector control of inductionmotor without rotational transducers. IEEE Trans Ind Appl 1992;28(5):1054–61.

[37] Kojabadi HM. Simulation and experimental studies of model referenceadaptive system for sensorless induction motor drive. Simul Model PractTheory 2005;13(6):451–64.

[38] Mitronikas ED, Safacas AN. An improved sensorless vector – control methodfor an induction motor drive. IEEE Trans Ind Electron 2005;52(6):1660–8.

[39] Ohyama K, Asher GM, Sumner M. Comparative analysis of experimentalperformance and stability of sensorless induction motor drives. IEEE TransInd Electron 2006;53(1):178–86.

[40] Matsuse K, Taniguchi S, Yoshizumi T, Namiki K. A speed-sensorless vectorcontrol of induction motor operating at high efficiency taking core loss intoaccount. IEEE Trans Ind Appl 2001;37(2):548–58.

[41] Cardenas R. MRAS observer for doubly fed induction machines. IEEE TransEnergy Convers 2004;19(2):467–8.

[42] Cardenas R. MRAS observer for sensorless control of standalone doubly fedinduction generators. IEEE Trans Energy Convers 2005;20(4):710–8.

[43] Paladugu A, Chowdhury BH. Sensorless control of inverter-fed inductionmotor drives. Electron Power Syst Res 2007;77(5–6):619–29.

[44] Sabanovic Asif, Fridman Leonid, Spurgeon Sarah. Variable structure systems:from principles to implementations. IEE Control Engineering Series 2004;66,United Kingdom.

[45] Utkin VI. Sliding mode control design principles and applications to electricdrives. IEEE Trans Ind Electron 1993;40(1):23–36.

[46] Ferrara A. Discussion on: an adaptive variable structure control law forsensorless induction motors. Eur J Control 2007;4:393–7.

[47] Proca A, Keyhani A, Utkin V, Miller J. Discrete time sliding mode, continuoustime sliding mode and vector control of induction motors. Int J Control2002;75(12):901–9.

[48] Lee H, Utkin VI. Chattering suppression methods in sliding mode controlsystems,. Annu Rev Control 2007;31(2):179–88.

[49] Jeng Lin F, Jong Wai R, Chuan Lin P. Robust speed sensorless induction motordrive. IEEE Trans Aero Electron Syst 1999;35(2):566–78.

[50] Comanescu M, Xu L. Sliding mode MRAS speed estimators for sensorlessvector control of induction machine. IEEE Trans Ind Electron 2006;53(1):146–53.

[51] Beltran B, Ahmed-Ali T, Benbouzid MEH. Sliding mode power control ofvariable-speed wind energy conversion systems. IEEE Trans Energy Convers2008;23(2):551–8.

[52] Beltran B, Ahmed-Ali T, Benbouzid MEH. High-order sliding mode control ofvariable-speed wind turbines,. IEEE Trans Ind Electron 2009;56(9):3314–21.

[53] Ben Elghali SE, Benbouzid MEH, Ahmed-Ali T, Charpentier JF, Mekri F. High-order sliding mode control of DFIG-based marine current turbine. In:Proceedings of IEEE IECON. Orlando, FL; November 2008. p. 1228–33.

[54] Beltran B, Ahmed-Ali T, Benbouzid MEH.High-order sliding mode control of aDFIG-based wind turbine for power maximization and grid fault tolerance. In:Proceedings of the IEEE IEMDC. Miami, FL; May 2009. p. 183–89.

[55] Hu J, Nian H, Hu B, He Y, Zhu ZQ. Direct active and reactive power regulationof DFIG using sliding mode control approach. IEEE Trans Energy Convers2010;25(4):1028–39.

[56] Ha JI, Sul SK. Sensorless field orientation control of an induction machine byhigh frequency signal injection In: Proceedings of the conference record IEEEIAS annual meeting; 1997. p. 426–32.

[57] Linke M, Kennel R, Holtz J.Sensorless speed and positon control of synchro-nous machines with alternate carrier injection. In: Proceedings of the IEEEinternational electric machines and drives conference, IEMDC'03, vol. 2; June2003. p. 1211–17.

[58] Holtz J, Quan J. Sensorless vector control of induction motors at very lowspeed using a nonlinear inverter model and parameter identification. IEEETrans Ind Appl 2002;38(4):1087–95.

[59] Consoli A, Scarcella G, Bottiglieri G, Scelba G, Testa A, Triolo DA. Low-frequency signal-demodulation-based sensorless technique for inductionmotor drives at low speed. IEEE Trans Ind Electron 2006;53(1):207–15.

[60] Zaky MS, Khater M, Yasin H, Shokralla SS. Very low speed and zero speedestimations of sensorless induction motor drives. Electron Power Syst Res2010;80(2):143–51.

[61] Luenberger David G. An introduction to observers. IEEE Trans Autom Control1971;AC-16(6):596–602.

[62] Rothenhagen K, Fuchs FW.Current sensor fault detection and reconfigurationfor a doubly fed induction generator. In: Proceedings of the IEEE IECON. Taipei,Taiwan; 2007. p. 1115–20.

[63] Rothenhagen K, Fuchs FW. Current sensor fault detection by bilinear observerfor a doubly fed induction generator. In: Proceedings of the IEEE conference;2006. p. 1369–74.

[64] Escobar RF, Astorga-Zaragoza CM, Téllez-Anguiano aC, Juárez-Romero D,Hernández Ja, Guerrero-Ramírez GV. Sensor fault detection and isolation viahigh-gain observers: application to a double-pipe heat exchanger. ISA Trans2011;50(3):480–6.

[65] Rothenhagen K, Fuchs FW. Current sensor fault detection and reconfigurationfor doubly fed induction generators. IEEE Trans Ind Electron 2009;56(10):4239–45.

[66] Romero ME, Seron M. Sensor fault tolerant control of doubly fed inductionmachines using multiple switching mechanisms. In: Proceedings of theAustralian Control Conference (AUCC). Australia; November 2011. p. 191–97.

67] Najafabadi TA, Salmasi FR, Jabehdar-maralani P. Detection and isolation ofspeed-, DC-link voltage-, and current-sensor faults based on an adaptiveobserver in induction-motor drives. IEEE Trans Ind Electron 2011;58(5):1662–72.

[68] Fan J, Zhang Y, Zheng Z. Observer-based reliable stabilization of uncertainlinear systems subject to actuator faults, saturation, and bounded systemdisturbances. ISA Trans 2013;52(6):730–7.

[69] Fan J, Zhang Y, Zheng Z. Observer-based reliable stabilization of uncertainlinear systems subject to actuator faults, saturation, and bounded systemdisturbances. ISA Trans 2013;52(6):730–7.

[70] Chan CW, Hua S, Yue ZZH. Application of fully decoupled parity equation infault detection and identification of DC motors. IEEE Trans Ind Electron2006;53(4):1277–84.

[71] Berriri H, Slama-belkhodja I. Enhanced parity equations method for sensorfault detection. In: Proceedings of the electrical drives conference on controland fault tolerant systems. France; 2010.

[72] Berriri H, Wissem NaouarM, Slama-Belkhodja Ilhem.Parity space approach forcurrent sensor fault detection and isolation in electrical systems. In: Proceed-ings of the 8th international multi-conference on systems, signals & devices.Tunisia; 2011.

[73] Berriri H, Naouar MW, Slama-belkhodja I. Easy and fast sensor fault detectionand isolation algorithm for electrical drives. IEEE Trans Power Electron2012;27(2):490–9.

[74] Wang T, Xie W, Zhang Y. Sliding mode fault tolerant control dealing withmodeling uncertainties and actuator faults. ISA Trans 2012;51(3):386–92.

[75] Chen M, Wu Q-X, Cui R-X. Terminal sliding mode tracking control for a class ofSISO uncertain nonlinear systems. ISA Trans 2013;52(2):198–206.

[76] Hao L-Y, Yang G-H. Robust fault tolerant control based on sliding modemethod for uncertain linear systems with quantization. ISA Trans 2013;52(5):600–10.

[77] Freire NMA, Member S, Estima JO, Cardoso AJM, Member S. A new approachfor current sensor fault diagnosis in PMSG drives for wind energy conversionsystems. IEEE Trans Ind Appl 2014;50(2):1206–14.

[78] Evans PD, Hill-Cottingham RJ. DC link current in PWM inverters. IEE Proc1986;133(4, Part B):217–24.

[79] Green TC, Williams BW. Derivation of motor line-current waveforms from theDC-link current of an Inverter. IEE Proc 1989;136(4, Part B):196–204.

[80] Boys JT. Novel current sensor for PWM AC drives. IEE Proc 1988;135(1, Part B):27–32.

[81] Blaabjerg.F, Pedersen.JK. An ideal PWM-VSI inverter using one current sensorin the DC-link. In: Proceedings of the conference record power electronics andvariable – speed drives; 1994. p. 458–64.

[82] Sun K, Wei Q, Huang L, Matsuse K. An over modulation method for PWM-inverter-fed IPMSM drive with single current sensor. IEEE Trans Ind Electron2010;57(10):3395–404.

[83] Lee WC, Lee TK, Hyun DS. Comparison of single-sensor current control in DClink for three-phase voltage-source PWM converters. IEEE Trans Ind Electron2001;48(3):491–505.

[84] Kim H, Jahns TM. Phase current reconstruction for AC motor drives using a DClink single current sensor and measurement voltage vectors. IEEE Trans IndElectron 2006;21(5):1413–9.

[85] Kim H, Jahns TM. Current control for AC motor drives using a single DC-linkcurrent sensor and measurement voltage vectors. IEEE Trans Power Electron2006;42(6):1539–47.

[86] Wolbank TM, Macheiner PE. Current-controller with single DC link currentmeasurement for inverter-fed AC machines based on an improved observer-structure. IEEE Trans Power Electron 2001;19(6):1562–7.

[87] Saritha B, Janakiraman PA. Sinusoidal three-phase current reconstruction andcontrol using a DC-link current sensor and a curve-fitting observer. IEEE TransInd Electron 2007;54(5):2657–64.

[88] Ha J. Current prediction in vector-controlled PWM inverters using single DC-link current sensor. IEEE Trans Ind Electron 2010;57(2):716–26.

[89] Cho Y, Labella T, Lai JS. A three-phase current reconstruction strategy withonline current offset compensation. IEEE Trans Ind Electron 2012;59(7):2924–33.

[90] Lai Y, Lin Y, Chen C. New hybrid pulsewidth modulation technique to reducecurrent distortion and extend current reconstruction range for a three-phaseinverter using only DC-link sensor. IEEE Trans Power Electron 2013;28(3):1331–7.





































































































































Analysis of field-oriented controlled induction motor drives under sensor faults and an overview of...

Documents

Transcript of Analysis of field-oriented controlled induction motor drives under sensor faults and an overview of...