New improved modulation method for acycloconvertor driving an induction motor

G.P. Hunter, MEngV.S. Ramsden, MEngSc, PhD

Indexing terms: Converters, Induction motors

Abstract: This paper describes new modulationtechniques that remove all the major limitations ofthe noncirculating current cycloconvertor, includ-ing low maximum frequency, subharmonics, poorbank crossover timing, and distortion due to dis-continuous current, when used to drive an induc-tion motor. The new modulation techniques workby keeping the flux in the induction motor (sensedby voltage feedback) as close to sinusoidal as pos-sible and give superior performance to the presentcurrent feedback techniques. An experimental 18-thyristor 3-pulse cycloconvertor showed that thenew modulation techniques allow smooth speedcontrol of an induction motor from 0 to 25 Hz(limited by software) with a peak output voltage of95% of the input voltage. It is expected that thesame techniques applied to a 36-thyristor 6-pulsecycloconvertor will allow smooth speed control upto 50 Hz. To obtain fast speed response, vectorcontrol can be easily added to the new modula-tion techniques. The new modulation techniquescan also be used to advantage in other cyclo-convertor and convertor applications. Patentshave been applied for.

List of symbols

fi = input frequencyf0 = output frequencytt = starting time of a trigger periodt2 = ending time of a trigger periodt0 = time of occurrence of an arbitrary fixed phase angle

of Vt for a trigger periodtf = time of triggering of thyristor in a trigger periodtc = time that current drops to zero (if this occurs) in a

trigger periodVo = output voltage on one phase of the cycloconvertorVr = output reference voltage on one phaseVb = boost voltage in one phase applied to overcome IZ

voltage dropVn = common mode voltage added to outputsVt = input voltage connected to the thyristor to be trig-

gered in a trigger periodVp — input voltage connected to the thyristor that is on

at the start of a trigger periodij/ = reference value of flux linkage in one phase of the

induction motor

Paper 6216B (P6/P1), first received 1st April 1987 and in revised form23 May 1988The authors are with the University of Technology, Sydney, PO Box123, Broadway, NSW 2007, Australia

K = constant determining stabilityLx = per phase stator leakage inductanceL2 = per phase, stator referred, rotor leakage inductance

1 Introduction

The noncirculating current cycloconvertor has manyadvantages over other forms of AC variable speed drive.Its maximum power output is virtually unlimited, itspower circuit is very simple (consisting of only phase-controlled thyristors and their associated snubbers), it isvery efficient, and it is naturally regenerative. With thepresent modulation methods in use, though, it suffersfrom some severe disadvantages. It has a low maximumoutput frequency (of about 25 Hz for a 6-pulse system)due to subharmonics appearing on the output, and itsuffers from voltage distortion and the associated torquepulsations due to inaccuracy in the choice of the bankcrossover times and the inability of the modulationmethods to compensate for discontinuous currents in thethyristors.

The performance can be improved by adding an extracurrent feedback loop around the cycloconvertor and itsmodulator [1, 2], but the improvement is limited by sta-bility considerations, and the cycloconvertor thenbecomes a current controlled device, rather than themore ideal voltage controlled device. Current control isparticularly a problem with multimotor drives.

It is much better to solve the performance problems byimproving the basic modulation method, rather than byattempting to linearise the present methods with currentfeedback. Attempts to improve the modulation methodhave been made in the past [3, 4], but these addressedonly the subharmonic problem and did not work verywell with the noncirculating current cycloconvertor. Thispaper describes a new modulation method that addresses,and should solve, all the problems of past modulationmethods. With this method, the. theory shows that sub-harmonics should be reduced to insignificant levels, theoccurrence of discontinuous current actually reduces theoutput voltage distortion, rather than increasing it, andcrossover between thyristor banks, should occur at theoptimum time. The maximum output frequency using thenew method is at least 25 Hz for a 3-pulse cycloconvertor(this was confirmed on a prototype that has beenconstructed) and is expected to be 50 Hz for a 6-pulsecycloconvertor. Further experimental work will investi-gate the extent to which the theoretical improvementsexpected are achieved.

Past attempts to find improved modulation methods[3, 4] have concentrated on trying to improve the outputharmonic spectrum to make it more suitable for aninduction motor load. This approach has the disadvan-tage that some problems, in particular distortion due to

324 IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988

discontinuous current and poor selection of the bankcrossover times, are ignored. Because of this, the newmodulation method was derived using a time domainapproach. The new method was chosen to minimise thedistortion in the motor flux waveform rather than tominimise unwanted harmonics in the output voltage fre-quency spectrum.

The new modulation method is particularly attractivewhen used with a 3-pulse cycloconvertor in an inductionmotor drive. The power circuit consists of only 18 thyris-tors and has the same efficiency and size as the equivalentconvertor for a DC motor drive. The performance is atleast as good as the equivalent 12 thyristor, 4-quadrantDC motor drive, with the advantage of using the morerugged induction motor.

The price paid for the improved performance of thenew modulation method is a more complex controlcircuit. The new modulation method is considerablymore complex than the present methods based on cosine-wave crossing control. Microprocessor control is essen-tial. The prototype 3-pulse system that has beendeveloped uses the extremely fast (200 ns instructiontime) 16 bit TMS32010 microprocessor from TexasInstruments. A slower microprocessor could be used, butat the sacrifice of response time (at present 7 ms for the3-pulse system) and with an increase in current ripple.

The new modulation method does not improve theinput power factor, but various techniques are availablefor achieving this [5]. Combining these techniques withthe present modulation scheme will be the subject offuture research.

2 Motor requirements

In developing the new modulation method, the motorrequirements when driven by a cycloconvertor werelooked at very carefully, particularly for the inductionmotor, as this is the preferred motor for most applica-tions.

One way to get good performance out of an inductionmotor drive using a cycloconvertor is to make it closelysimulate a DC motor drive using a thyristor convertor,and this is the approach taken here. In a DC drive, themotor flux is kept constant by a constant field currentand speed is controlled by a highly distorted DC voltageapplied to the armature. The distortion in the armaturevoltage produces a high ripple current and a correspond-ing increase in motor heating, but has little effect on themotor performance. This is because the torque is pro-portional to the product of flux and current and so theripple current produces only a corresponding high fre-quency ripple torque without affecting the averagetorque.

To simulate the conditions in a DC machine for aninduction motor, the components of flux linkage in thethree phases must be kept as close as possible to threesine waves of equal amplitude and displaced by 120°,which means that the integrals of the three input voltagesmust be kept likewise. This is the criterion on which thenew modulation method was developed. If the fluxlinkage components in each phase can be kept sinusoidalby the cycloconvertor, only the component of current ofthe same frequency can contribute to the DC componentof torque. There would be a pulsating torque owing tothe ripple components of the currents, but this would beno worse than that of the equivalent DC convertordriving a DC machine and containing two thirds of thenumber of thyristors as the cycloconvertor.

3 Basic modulation method

Described here is a greatly simplified version of the newmodulation method introduced as an aid to understand-ing the final version. This basic modulation method,which actually could be used to advantage in AC to DCthyristor convertor control, will be called pre-integrationcontrol. The method is described for the positive currentthyristor bank only. Operation with the negative bank isidentical, except that the waveforms are inverted.

The method is illustrated in Fig. 1. Shown are theinput voltage waveforms, the wanted fundamental output

tf t2

Fig. 1 One pulse of the output of a cycloconvertor using pre-integration controlArea C = Area D

voltage, Vr, and the trigger instant, tf, of the thyristor. Inthis control scheme, the thyristor must be triggeredwithin the time interval t^ to t2, where tx and t2 aredefined as the instances when the reference waveformintersects the input waveforms fed to the incoming andoutgoing thyristors. The interval tt to t2 is called thetrigger period. The time tf, when the thyristor is trig-gered, occurs when area c is equal to area d. Here, area dcannot be measured directly because it occurs after thethyristor is triggered. It must be precalculated. Expressedmathematically, tf is chosen so that:

0 = (Vo - Vr) dt (1)

This scheme keeps the integral of the output voltage fromtl-t2 equal to the integral of the reference voltage overthe same time, and thus keeps the average of the twowaveforms over this interval the same. An advantage ofthis method for driving an induction motor is that if themotor flux is at the correct value at time tlt then it willalso be at the correct value at time t2 and at the end ofevery subsequent trigger period. Also, because the inte-grals of the reference and output waveforms are equal toeach other at the end of each trigger period, there can beno long term build-up in errors (which plague cosinusoi-dal control) causing subharmonics.

Pre-integration control is similar to integral control[6] in that direct control is kept over the integral of thedifference between the output and reference voltages, butdiffers in that it calculates part of the integral in advance.This precalculation greatly improves stability. Pre-integration control is inherently stable, but integralcontrol is not, and must be supplemented with othertechniques to make it stable.

3.1 Compensating for discontinuous currentThe negative voltage excursions below Vr that occurbefore the trigger instant, tf, when the output phasecurrent is positive, can bring this current to zero for a

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988 325

short time (i.e. make the current discontinuous). A similarsituation can occur for a negative phase current due topositive voltage excursions above Vr. During this time ofzero current, the phase voltage, Vo, on the motor terminaldepends on the voltages of the other two phases and islargely indeterminate.

An improved calculation method which results in thesame triggering time as the above method, but allows thevoltage distortion caused by discontinuous current to becompensated for is illustrated in Fig. 2. Area a in this

Fig. 2 A practical method of implementing pre-integration controlshaded region is Area bArea b = Area a

figure is first precalculated. This is given by the formula:

Next, the voltage difference V, — Vo is integrated in realtime from time tx (area b). When this integral reaches thevalue of area a (which was precalculated), the thyristor istriggered. The algorithm is really the expansion of eqn. 1to the following:

o = j V , - vr) dt - [ \ v t - v0) dtJtl Jtl

(3)

The first term is precalculated at the start of the triggerperiod. The other term is calculated repeatedly from thestart of the period with tf replaced by the current time, t.When the second term reaches the value of the first term,the thyristor is triggered.

The effect of this method with discontinuous current isshown in Fig. 3. At time tc, the current drops to zero and

ti tc tf

Fig. 3 Effect of discontinuous current on the output of a cyclo-convertor using pre-integration controlArea b still equals Area a

the thyristor that was previously on turns off. With noconnection between the input and output of the cyclo-convertor, Vo is now indeterminate and may even exceedthe peak value of Vt owing to induced voltage from theother motor phases. The current remains zero until time

tf when the next thyristor triggers, re-establishing a posi-tive current flow. As can be seen from Fig. 3, if the thyris-tor is triggered by the previously described method, i.e.when area b is equal to area a, the integral from tx to t2of Vo — Vr is again zero, and thus the effects of the discon-tinuous current are compensated for.

3.2 Error cancellationFrom Fig. 3, it can be seen that correct operation of thepre-integration control method relies on:

(a) that Vt remains undistorted during the triggerperiod

(b) that the commutation time of the thyristors is veryshort.

Neither of these conditions may necessarily hold in apractical cycloconvertor. This may result in the unwantedbuild-up of the integral of Vo — Vr over several triggerperiods if the error is of the same sign over these triggerperiods.

This unwanted build-up of the integral of Vo — Vrwhen sequential trigger periods have calculation errors ofthe same sign can be corrected by the following additionto the control method: The voltage Vo — Vr is fed to anintegrator. The output of the integrator at time t2 rep-resents the error in the area between Vo and Vr at thistime. If this error is added to the next precalculated areaa in Fig. 2, it will automatically be corrected for in thenext trigger period. The integral of Vo — Vr at the end ofeach period will now be in error by the calculation errorof this period, and not by the sum of the calculationerrors of all previous periods.

3.3 Bank switchingWith pre-integration control, the time when bank switch-ing should occur is fairly easy to determine. If, in Fig. 3,the next thyristor is not switched on by the end of thetrigger period at t2, then it is not possible to maintain theaverage output at the reference voltage and this is thetime when bank switching to the negative bank shouldoccur. To calculate the next triggering time after the bankcrossover, the time of bank crossover can be made thestarting time of the next trigger period. Note that this is avery different approach to that used in present modula-tion schemes which switch banks at an estimate of thezero crossing of the fundamental component of outputcurrent. Instead, the instant of bank switching is selectedto minimise the output voltage distortion, which is abetter method for an induction motor, although thisshould also minimise the current distortion.

This technique is not used in the final design becausethe double pre-integration control method outlinedbelow allows a much more accurate method to be used(see Section 4.5).

4 Further improvement: double pre-integrationcontrol

4.1 Problem with pre-integration control methodFig. 4a shows the output waveform Vo that would beobtained from the reference waveform Vr using pre-integration control. It is assumed for simplicity that theoutput current is in phase with the output voltage anddoes not become discontinuous. Fig. 4b shows the inte-gral of Vr and the integral of Vo. It can be seen that theaverage of the integral of the output waveform is badlydistorted with the pre-integration control method. This isbecause the integral of Vo is always more negative than

326 IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988

the integral of Vr when the positive bank is operating, butis always more positive when the negative bank is oper-ating. In an induction motor, this would cause a corre-

•/Vodt

Vvrdt

Fig. 4 Typical waveforms with pre-integration control. The outputcurrent is assumed to be sinusoidal and in phase with the output voltagea Output and reference voltages, Vo and Vr, shown with input voltagesb Integrals of Vo and Vr, showing the distortion produced with pre-integrationcontrol

sponding distortion in the flux waveform which woulddegrade the performance of the motor. This distortion isremoved with an improvement on pre-integration controlcalled here double pre-integration control. This newmodulation method retains the advantages of pre-integration control, namely elimination of subharmonics,compensation for discontinuous current and improvedbank switching determination.

4.2 Description of double pre-integration controlmethod

As can be seen from Fig. 4, with pre-integration controlthe average of Vo is maintained at the average of Vr overone trigger period, but this is not the case for the inte-grals of these two waveforms. To maintain correct flux inan induction motor, though, keeping the average of thetwo integrals equal is the main requirement. This cri-terion can be met by applying pre-integration control tothe integrals of the output and reference waveformsrather than to the waveforms themselves. This is the basisof double pre-integration control. Expressed mathemat-ically, tf in a trigger period is chosen so that

0 = f2

Jti

- vr) dt2 (4)

where time t = 0 is any fixed time, but can convenientlybe made the time when the cycloconvertor is started.

Unfortunately, as can be seen in the example in Fig. 5,this method of modulation is unstable, as there is morethan one solution for the voltage waveform. The insta-bility shows itself as an oscillation in the value of theintegral of Vo — VT at the end of each trigger period. Tostabilise this modulation method, one technique is to adda term to eqn. 4 which is zero under steady state condi-tions without instability and thus does not interfere withthe basic modulation method, but which is non-zero in adirection which suppresses the instability when it occurs.Looking at Fig. 5, a suitable term is one proportional tothe difference between the integral of Vo — VT at times txand t2. Eqn. 4 with this incorporated becomes

o-ff,Jn Jo

(Vo - Vr) dt2 + K(t:- h) f Vo " KJti

Note that the constant K has been multiplied by the term(*2 ~~ ti) t 0 make it dimensionless. To determine theoptimum value of the constant, a rough computer simu-

(=0)

onetrigger period

Fig. 5 Illustration of instability with unstabilised double pre-integration controla Output and reference waveforms, Vo and Vr, with input waveformsb Integral of Vo and Vr

lation of the modulation method, which assumes theinput waveforms are trapezoidal rather than sinusoidal(this is closer to the actual waveforms in the prototype —see Section 5) was carried out [see Appendix]. It wasfound that the optimum value of K, which correspondsto critical damping, is 0.5. The simulation also showedthat with this value of K, recovery from a disturbance isvery fast. If a disturbance occurs at the start of a triggerperiod, then the integral of Vo — VT reaches 96% of itssteady state value by the end of the period. It is hoped toderive the optimum value of K mathematically at a latertime in order to find out whether it is waveform depen-dent.

In order to determine a practical algorithm to imple-ment this modulation method, and to have compensationfor discontinuous current while minimising the calcu-lations at each sampling time, eqn. 5 is best expandedinto a similar form to eqn. 3, with no integrations start-ing from tf. There are many expansions that fulfil thesecriteria. One that is particularly suitable to the micro-processor used in the prototype is:

o = f'2 [vt dt2 - [2 fVP dt2

Jto Jto Jti JO

-h) [Jt

[Vtdt-K(t2-t,) [to Jti

- K(t2 - tx)

n Jo

> (6a)

(6b)

(6c)

- I \Vt dt2 + (t2 - tf) V o AJto Jto Jo

+ K(t2 - tx) PV0 dtJo

- (h - tf) \"vt dt - K(t2 - tx) ( \ dtJto Jto

Vr)dt (5)

(6d)

This is because the TMS320 microprocessor is optimisedto calculate equations consisting of groups of two values

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988 327

multiplied together and separated by additions and sub-tractions.

The time t0 is a time chosen before the start of thetrigger period which corresponds to an arbitrary fixedphase angle of Vt. This is introduced to enable the termsinvolving Vt to be calculated using fast look-up tables. Inthis algorithm, the output voltage is not required, butonly its continuous integral. This eases transducerrequirement, as the integral can be obtained in digitalform directly via an integrating type V/f convertor, anisolating pulse transformer and a counter. It is notrecommended that the integration of the output voltagebe done in software, as the accumulation of round-offerrors could result in subharmonics occurring on theoutput. In operation, the terms involving tf are calcu-lated repeatedly from the start of the trigger period withtf replaced by the current time, and then added to theother pre-calculated terms. When the total goes throughzero, the thyristor is triggered.

Eqn. 6a of the equation can be calculated any time upto, and perhaps just after the start of the trigger period.In the prototype it is calculated just after the start inorder to minimise the system response time. Eqn. 6b isfound by reading the integral at the start of the triggerperiod, then carrying out multiplications, and so cannotbe done until after the start of the trigger period. Eqns. 6cand d contain the variable tf, and so must be calculatedrepeatedly as described above. The double integral ineqn. 6c can be calculated numerically by successivelyreading the value of the integral of Vo and adding it to anaccumulator.

4.3 Control of flux and voltage boostIn a real motor, the motor flux is not the integral of theapplied voltage as assumed so far, but is the integral ofthe applied voltage less the voltage drop across themotor leakage reactance and the stator resistance. If wesplit the reference voltage Vr into a boost voltage com-ponent Vb, to compensate for this voltage drop, and acomponent due to a new reference, if/(t), representing themotor flux at time t. The integral of Vr can then beexpanded to

Vrdt = ij/(t) + \Vbdt (7)Jo Jo

Using this expansion, eqn. 6 now becomes:

f'2 ff f12 f'2 f'0 = \Vt dt2 - \ t(t) dt- \Vb dt2

Jto Jto Jti Jti Jo

f'2+ K(t2 — tj \ Vtdt- K(t2 - t1)(«A(r2) — «A(*i))

JtoCti fn

- K(t2 - tj Vbdt- K(t2 - tt) Vo dtJti Jo

+ \ \v0 dt2 - \ \vtdt2 + (t2 - tf) v0 dtJti Jo Jto Jto Jo

f"+ K(t2 - tj Vo dt

Jo- ( h - tf) \ V t dt - K ( t 2 - t j \ V t dt ( 8 )

Jto Jto

For normal motor control without field weakening, theflux waveform should be kept constant, so ip(t) in eqn. 8can be found from a look-up table. The amplitude and

328

phase of the boost voltage, Vb, can be fixed for simplecontrol schemes, or for fast response can be controlled bya field-oriented control scheme (also known as vectorcontrol). If field-oriented control is used, it needs to be ofthe voltage control type [7, 8] because the double pre-integration modulation method controls the motorvoltage rather than the motor current. This is no dis-advantage since recent work [8] indicates that with field-oriented control, voltage control is to be preferred tocurrent control.

4.4 Effect on ripple currentA per-phase motor equivalent circuit that is adequate fordetermining the current ripple waveform is shown in Fig.6. The stator resistance does not appear in the circuit

Fig. 6 Simple per phase equivalent circuit for ripple current determi-nation

because it is assumed that the leakage reactance is muchlarger than the stator resistance at the ripple frequency.Voltage Vr is the reference voltage of the correspondingphase of the cycloconvertor and is equal to the funda-mental component of the supply voltage. Using thisequivalent circuit, the ripple current is given by

Ripple current = ] \ (Vo - Vr) dt*-l + ^2 J

(9)

A typical ripple current waveform for a positive currentfrom the cycloconvertor is shown in Fig. 7.

/YWVWY-Fig. 7 Typical waveform for j (Vo — Vr) dt and ripple current (= 1/Lj+ MW>- K)dt)

Comparing eqn. 9 to eqns. 4 and 5, it can be seen thatthe double pre-integration control method keeps the inte-gral of the current ripple waveform as close as possible tozero during each triggering period. This should also keepthe amplitude of the current ripple near its minimumvalue. This indicates that double pre-integration controlis also a very good modulation method for cyclo-convertors used in other applications where the ripplecurrent is limited primarily by inductance, such as highfrequency power system interties and synchronous motordrives.

In very small motors, the stator resistance may be sig-nificant at the ripple frequency. The possible effects ofthis have not been investigated for this paper, but is anarea of future research.

4.5 Improved bank crossover determinationThe optimum bank crossover time is the first time theactual current is zero (and thus all thyristors in thatphase are off) after the fundamental component ofcurrent passes through zero. Since the instantaneousvalue of the current ripple is proportional to the integralof Vo — Vr, which is a value which is easily calculatedwhen the double pre-integration control method is used,it is quite easy to accurately determine this optimum

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988

bank crossover time. As shown in Fig. 8, for a positiveoutput current, the crossover to the negative bank shouldoccur at the first instant when the output current is zero

ripple component of current

optimum bank crossover time

Fig. 8 Illustration of optimum bank cross-over time (actualcurrent = 0 and ripple current >0)

fundamental component of current

and the integral of Vo — VT is positive (the integral shouldbe negative for a negative output current). This is the firstpoint when the current is zero and the fundamental com-ponent of the current is negative, since, according to eqn.9, the ripple current is proportional to the integral ofVo-K.

This bank crossover determination method is animprovement over the one used for pre-integrationcontrol described in Section 3.3. Bank crossover is initi-ated at a current zero without waiting for the end of thetrigger period.

The mechanism of bank crossover used is to first turnoff all the gate signals to the thyristors in that phase.Then a check is made about 300 /is later that the currentis still zero to ensure the thyristors are turned completelyoff. The thyristor gate signals are then left off until thenext calculated triggering time. With double pre-integration control, leaving the thyristors off during thefirst part of a trigger period does not affect the accuracyof the modulation method.

To calculate the next triggering time after the bankcrossover, the technique described in Section 3.3 can beused where the time of bank crossover is made the start-ing time of the next trigger period.

When the cycloconvertor has a 3-phase load with noneutral connection, as in this case with an inductionmotor load, an extra modification must be made to takeaccount of the fact that any voltage distortion that iscommon to all three outputs does not produce any corre-sponding current ripple. Any voltage distortion waveformwhich is added equally and in phase to all three outputscannot affect the winding voltages or currents on amotor, or the voltages across any type of load, which isonly connected to the three outputs. The instantaneousvalue of the current ripple in a particular output is in thiscase proportional to the integral of Vo — Vr less thecommon mode component, or instantaneous average, ofthe output voltages (Vo) minus their references (Vr). This isexpressed as:

Ripple current oc | (Vo - Vr) dt - \ £ \(V0- Vr) dtJ -* u,v,w J

(10)

To compensate for this, the crossover from the positive tothe negative bank should now occur at the first instantwhen the output current is zero and the expression onthe right hand side of eqn. 10 is positive (and negative fora negative to positive bank crossover). This is the schemeused in the prototype.

5 Improvements to output voltage range anddistortion

Using the above modulation method allows the 3 pulsecycloconverter to efficiently and accurately control thespeed of an induction motor. On the prototype cyclo-convertor, though, some further modifications wereadded to maximise its performance and these will bedescribed below. With these modifications, the maximumoutput voltage before clipping is increased to 95% of theinput voltage and the distortion when operating at ornear maximum output voltage is improved. The modifi-cations can be used with most modulation methods, notjust the modulation methods described here.

5.1 Improvement of the output voltage range bychanging the output neutral voltage

The basic circuit of a cycloconvertor with a 3-phaseinduction motor load (assumed here to be star connected)is shown in Fig. 9. Normally the neutral voltage, Vn, is

Fig. 9 Basic three pulse cycloconvertor with motor load showing theneutral voltage, Vn

kept as close to zero as possible, but in actual fact, Vn canbe any value without affecting the motor, provided thevoltages between U, V and W are 3-phase sine waves. Bychoosing a suitable waveform for Vn and adding this toeach of the three output reference waveforms, it is pos-sible to increase the line to line voltages on the output ofthe cycloconvertor for the same peak voltages withrespect to neutral. The normal waveform chosen for Vn isa sine wave of frequency three times the output frequencyand an amplitude that will minimise the peak value of theoutput reference voltages. This procedure is well knownand has been documented many times, an example beingNakajima et al. [5].

The same method to improve the output voltage rangeis used here, but instead of a sine wave, the waveform ischosen to maximise the effect. Fig. 10a shows the wave-form used for Vn. It is the same as waveform VA (made upfrom portions of the U, V and W output referencevoltage waveforms) in the Figure, but with half the ampli-tude. It is generated in software from a look-up table.Fig. 10b shows the resulting output reference waveforms.With this modification, the peak line to line fundamentaloutput voltage before clipping is 95% of the inputvoltage. With the normal method of choosing a sine wavefor Vn, the peak line to line fundamental output voltage isalso improved to 95% of the input voltage, but theoutput reference voltage is at its peak level for a longerproportion of each cycle resulting in any clipping produc-ing more severe output voltage distortion.

5.2 Improving distortion by changing input referenceSo far it has been assumed that the measuring referencepoint used by the cycloconvertor control circuits is theinput neutral point. This is the natural reference pointand is the one that has always been used in previous

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988 329

cycloconvertor designs. As there is no actual neutral sup-plied to the cycloconvertor, this reference point wouldhave to be obtained using a star network of resistors. In

Fig. 10 Derivation of new output reference voltage waveforms byadjustment of Vn

a Output reference voltage waveforms and neutral voltage waveform, Vn, withoutput neutral as reference. Vn is chosen to be one half VA

b Output reference voltage waveforms with VK added

the prototype cycloconvertor, a new alternative referencepoint, obtained with the circuit of Fig. 11, was used. Thevoltage waveform at this new reference point with respect

i i i

f f finputreference

Fig. 11 Circuit used to create the new input reference point

to the true input neutral point is similar to Vn in Fig. 10a,but with a frequency of three times the mains frequency.With respect to this new reference point, the input wave-forms are no longer sine waves, but are the same as thewaveforms of Fig. 10b.

The advantage of using the alternative reference is thatwhen operating at maximum output voltage, the voltagedistortion is reduced and the ripple frequency is doubled.This effect is shown in Fig. 12. An important side benefit

Fig. 12 Output voltage ripple waveforms at maximum positive voltagelevel. Both diagrams are at the same scalea With neutral referenceb With new reference

to using the alternative reference is that the input wave-forms can be approximated by trapezoidal waveforms.This considerably eases the calculations required in themicroprocessor to determine the start and end times ttand t2 of the current trigger period. These times aredetermined from the intersection of the input waveformsand the output reference waveform, as shown in Fig. 1.With the input and reference waveforms approximated

by trapezoidal waveforms, this involves finding the inter-section of straight lines. If any of the waveforms had beensinusoidal, an iterative method would have had to beused.

Note that the evaluation in the software of the expres-sions in eqns. 6 and 8 involving Vt is carried out usinglook-up tables generated from the exact values of the newinput waveform.

6 Implementation

6.1 HardwareFig. 13 is a block diagram of the prototype cyclo-convertor-induction motor drive using a three pulsecycloconvertor with double pre-integration control. Thezero current detectors on each output phase work bysensing the voltage across each thyristor as described byHamblin and Barton [9]. To measure the integral of theoutput voltages, three voltage to frequency convertors ofthe integrating type interfaced to the microprocessor viacounters are used. The V/F convertors operate at acentre frequency of 100 kHz, ensuring the output of thecounters show the integrals of the voltages applied to theconvertors to a high enough accuracy. An offset voltage(not shown in Fig. 13) is applied to the input of thevoltage to frequency convertors to enable them tooperate in the bipolar mode. This is compensated for bythe microprocessor software. The input analogue speedcommand is also measured by a voltage to frequencyconvertor coupled to a counter. This has the advantagesof being cheaper than an analogue to digital convertor,allows the average speed over each speed samplingperiod to be measured rather than the speed at each sam-pling instant, and gives infinite speed resolution. Thespeed command determines the output frequencydirectly. The output voltage is determined by the fre-quency because the flux waveforms, which is held in alook-up table, has fixed amplitude.

The microprocessor is timed by two interrupt signalssupplied by a phase-locked-loop locked to the mains.One is at the same frequency as the mains and is used tosynchronise the microprocessor to the mains. The otheris at 60 times the mains frequency and determines thesampling instances.

For accurate control of voltage boost to enable accu-rate and fast motor response, perhaps by using field-oriented control (see Section 4.3), the tachometer can beadded as shown. The prototype does not include atachometer at present.

6.2 Microprocessor requirementsThe main limitation on the choice of microprocessor isprocessing speed. As can be seen from eqn. 6, a largenumber of calculations are required during each sam-pling interval, as well as other jobs such as checking forzero current. To reduce the load on the microprocessor,the sampling interval should be as long as possible, but alonger sampling interval produces extra voltage distor-tion because the thyristor firing time can be delayed byup to one sampling interval from the ideal time. It wouldbe ideal if the sampling interval can be made shortenough so that the maximum error in the integral of Vothat is introduced by delaying the firing time by one sam-pling interval is much less than the normal output distor-tion, which can be quantified as the normal peak value ofthe integral of Vo — Vr. The sampling interval chosen for

330 IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988

the prototype is 333 JIS which results in a maximum pos-sible error of the integral of Vo — VT (when Vo - Vr is amaximum and the firing error is 1 sample interval) ofabout one quarter the peak excursion of this integral.This is greater than the ideal, but was limited by thespeed of the microprocessor used.

output frequency of 50 Hz can be obtained from a6-pulse cycloconvertor using double pre-integrationcontrol, but at the expense of twice the number of thyris-tors in the power circuit. A standard mains-voltage delta-connected induction motor can be used for the 3-pulsecycloconvertor by reconnecting it to star configuration.

peripheral interfaceadapter

interruptsspeedV/Fconvertor

Fig. 13 Block diagram of the prototype cycloconvertor with the optional tachometer circuit

analoguespeed command

The microprocessor chosen, the Texas InstrumentsTMS32010, is one of the very few on the market withenough processing speed without going to bit-slicedevices. It is designed for digital signal processing, buthas an instruction set powerful enough for generalcontrol use.

6.3 Performance of prototypeCycloconvertors are known for their smoothness at lowspeeds. With the new method of modulation, this isimproved even further. The double pre-integrationcontrol technique eliminates any possibility of sub-harmonics and prevents distortion being introduced bydiscontinuous currents.

The software used in the prototype was designed for 0to 25 Hz operation, so its performance could only beevaluated up to this frequency. With a two-pole induc-tion motor, this allows a speed range from 0 to1500 RPM which is adequate for most applications. Per-formance evaluation at frequencies above 25 Hz will bepossible after software modifications. This will be investi-gated in later research work. Note that a maximum

The line-to-line voltage required for 25 Hz operationwould then be 86-6% of the mains voltage, which is areasonable match to the cycloconvertor. This is what wasused for testing the prototype.

The motor used for testing was a 4 pole, 7.5 kW (at50 Hz) cage induction motor which was connected to aDC generator as a load. No tachometer feedback wasused and the voltage boost from the cycloconvertor wasfixed at a level that would give a maximum torque at lowspeed of one half full load torque. From 0 to 25 Hz, thehighest frequency tested, the drive performance wasexcellent with no hint of instability or torque pulsations.The motor rotated smoothly even at 0.1 Hz. Below0.5 Hz, multiple switchings between the positive andnegative banks occurred near each true current zeropoint, but this did not affect the performance. In factattempts to prevent this using either hysteresis or by pre-venting further crossovers for a period after the firstcrossover introduced unwanted torque pulsations.

Figs. 14, 15 and 16 show oscillograms of some voltageand current waveforms obtained from the prototype.Note that in Figs. 14 and 15 the output phase voltage

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988 331

can go higher than the peak input voltage when theoutput current is zero. This is owing to induced voltagefrom the other two phases. Note that the current wave-

tortion than this integral waveform due to the filteringeffect of the leakage inductance. Note that at this fre-quency, a subharmonic at 150 — 6 x 24.2 = 4.8 Hz is

M ?, i Mi

Fig. 14 Waveforms recorded with an output frequency of 7 Hz and noload on motorTop: output voltage, phase to reference point, 200 V/div.Bottom: output phase current, 10 A/div.Horiz.: 20ms/div.

Fig. 16 Waveforms recorded with an output frequency of 25 Hz and aload of 2.5 kW on the motorTop: output voltage, phase to reference point, 200 V/div.Bottom: output phase current, 10 A/div.Horiz.: lOms/div.

Fig. 15 Waveforms recorded with an output frequency of 25 Hz andno load on motorTop: output voltage, phase to reference point, 200 V/div.Bottom: output phase current, 10 A/div.Horiz.: lOms/div.

form is fairly distorted, but because the flux waveform issinusoidal, this does not degrade performance, except forincreased I2R losses. In Figs. 14 and 15, bank crossoveroccurs near the peaks of the voltage waveform. This iswhere the fundamental component of the output currentwould be zero for an unloaded motor, indicating thebank crossover algorithm works correctly even withsevere discontinuous current.

Fig. 17 shows the integral of the UV line to linevoltage together with the U phase current for anunloaded motor at an output frequency of 24.2 Hz. Theoscillogram is taken over many cycles in order to observesubharmonic effects. As can be seen, this integral wave-form is nearly sinusoidal even near 25 Hz and with longperiods of discontinuous current (the proportion of dis-continuous current in these conditions can be seen moreclearly in Fig. 15). The actual motor flux waveformappearing at the rotor conductors will have even less dis-

332

Fig. 17 Waveforms recorded with an output frequency of 24.2 Hz andno load on the motor . •••• •••Top: integral of U — V line to line voltage using- an electronic integrator.Uncalibrated.Bottom: output current in U phase, 10 A/div.Horiz.: 50 ms/div. ' '

normally expected. Inspection of Fig. 17 shows thatalthough a beating effect at around 5 Hz can beobserved, no significant subharmonic current at 5 Hz ispresent.

The response time of the cycloconvertor with doublepre-integration control depends on how the algorithmsare implemented in the microprocessor. In the prototype,as explained in Section 4.2, of eqn. 6a for a given triggerperiod is calculated just after the start of that period. Todo this calculation, the reference voltage waveform to theend of the period must be known. This problem is over-come in the prototype by measuring the input variables,speed reference and tachometer output, or just the speedreference when there is no tachometer feedback, every120° advance of input phase, but delaying the use of thesereadings until after the next 120° advance. This gives an

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988

effective response time delay of about 7 ms, which is asgood as the best D.C. drives.

7 Conclusions

This paper has described new phase control methods(pre-integration control and double pre-integrationcontrol) for the noncirculating current cycloconvertortogether with new techniques to improve the selection ofthe bank crossover time, improve the output voltagerange, and reduce the output voltage distortion. Doublepre-integration control maintains sinusoidal flux in aninduction motor drive, minimises current ripple andallows better selection of the bank crossover time. It hasbeen shown experimentally that an induction motorrated at 7.5 kW at 50 Hz can be successfully driven at afrequency varying from 0 to 25 Hz with an 18-thyristorcycloconvertor. The method requires a fast micro-processor for its implementation.

Since with double pre-integration control the currentripple in an inductive circuit is minimised, this controlmethod should also be useful for other applications suchas synchronous motor drives and high frequency powersystem interties.

Future research will include a more rigorous mathe-matical analysis of the modulation method and its effecton the induction motor in order to find the theoreticallimits to performance and to find possible improvementsto the modulation method. The incorporation of powerfactor improvement techniques [5] will also be investi-gated.

Future development work will involve the incorpo-ration of vector control which will give the cyclo-convertor and induction motor drive the characteristicsof a wide-speed-range regenerative DC drive. As it standsit can compete with a regenerative PWM inverter driveor a current source inverter drive.

Patents have been applied for world wide.

8 References

1 AKAGI, H., FUJITA, M., FUKAO, T., and MIYAIRI, S.: 'Applica-tion of microcomputer to current-controlled cycloconvertor systems',Electr. Eng. Jpn., 1980,100, (4), pp. 86-94

2 KUROSAWA, R., SHIMAMURA, T., UCHINO, H., and SUGI, K.:'Microcomputer-based high power cycloconvertor-fed inductionmotor drive'. IEEE-IAS Annual Meeting, 1982, pp. 462-467

3 BIRD, B.M., and FORD, J.S.: 'Improvements in phase-controlledcirculating-current cycloconvertor using communication principles',Proc. IEE, 1974,121, (10), pp. 1146-1149

4 SMITH, G.P., and RAMSDEN, V.S.: 'Study of new techniques forthe control of cycloconvertors, with reference to AC machine control',I.E. Aust. Electr. Eng. Trans., 1978, EE14, (2), pp. 92-96

5 NAKAJIMA, K., SHIMIZU, I., and KUNIYOSHI, M.: 'Reactivepower reduced cycloconvertor with bias voltage at the neutral point'.IEEE-IAS Annual Meeting, 1980. Pt. II, pp. 785-790

6 PELLY, B.R.: Thyristor phase controlled convenors and cyclo-convertors'(Wiley, 1971)

7 YAMAMURA, S., NAKAGAWA, S., and KAWAMURA, A.:'Voltage type control of induction motor by means of field acceler-ation method', Electr. Eng. Jpn., 1984,104, (4), pp. 89-94

8 HARASHIMA, F., KONDO, S., OHNISHI, K., KAJITA, M., andSUSONO, M.: 'Multimicroprocessor-based control system for quickresponse induction motor drive', IEEE Trans., 1985, IA-21, (4), pp.602-609

9 HAMBLIN, T.M., and BARTON, T.H.: 'Cycloconvertor controlcircuits', ibid., 1972, IA-8, (4), pp. 443-452

9 Appendix

To get a rough idea of the optimum value of the stabilityconstant K an approximate simulation of one output of

the cycloconvertor was undertaken. The input andoutput waveforms were approximated to the waveformsshown in Fig. 18. Time and voltage are assumed to be

Vr( = 0.5)

0

Fig. 18 Approximation of input and output waveforms used forexamining stability

normalised to the values shown in this Figure. It isassumed that positive current is flowing from the output.

For the trigger period tx to t2 as shown in Fig. 18, andwith the reference voltage Vr fixed at 0.5, eqn. 5 becomes:

0 = t\ - 2K) + 0.5K (11)

where

fJo

A = f Vo - Vr) dtJo

Solving this for ta:

(12)

The value of A for the start of the next period is given by:

A(next) = I Vo - K) dtJo

= -ta + li + A (13)

The variation in the value of A from period to period is agood indication of stability. To determine stability, A wasset to zero at the start of the first trigger period and thencalculated for each subsequent period for different valuesof K. The results are listed in Table 1. These results indi-

Table 1 : Simulation results

Periods elapsed

0123456

Value of A

K = 0A

0106.5151

99.2547100.082599.9908

100.001099.9999

(% of final

K = 0.5

096.598099.9958

100.0000100.0000100.0000100.0000

value)

K = 0.6

088.323598.901499.899899.990999.999299.9999

Final value of ta = 0.25, final value of A = ^ (for Vr = 0.5)

cate that the best value of K is about 0.5. This gives thefastest settling time to equilibrium conditions.

IEE PROCEEDINGS, Vol. 135, Pt. B, No. 6, NOVEMBER 1988 333