application of an e-machine emulator for power converter ... · application of an e-machine...

EEVCBrussels, Belgium, November 19-22, 2012

Application of an e-machine emulator for power convertertests in the development of electric drives

Dipl.-Ing. Stefan Uebener1, Prof. Dr.-Ing. Joachim Bocker21Daimler AG, e-Drive and Future Mobility, [email protected]

2University of Paderborn, Power Electronics and Electrical Drives, [email protected]

Abstract

In the development and validation of electric drives developers are facing great challenges. The existingfacilities for testing power electronics do not always fit the desired requirements. In this paper a test benchincluding an electric machine emulator will be presented, which physically reproduce the power flow ofa permanent magnet synchronous motor. The core of the e-machine emulator is a real-time flux basede-machine model that contains the characteristics of the desired machine. Furthermore, the performanceof an e-machine emulator in comparison to a dyno test bench and the benefits for the development ofalternative propulsion has been investigated. In the investigation could be shown, that the flux deviationsat stationary workpoints are under 1% in the area of constant torque.

Keywords: hardware-in-the-loop (HIL), simulation, permanent magnet motor, electric drive, AC motor

1 IntroductionIn the development of hybrid and electric vehi-cles the permanent magnet synchronous motor(PMSM) is preferred by many automobile manu-factures due to its high efficiency and power den-sity. Voltage-source inverters with field-orientedcontrol are generally used to control these ma-chines. There are two classic test bench config-urations for the application and validation of theinverter hard- and software:

• Hardware-In-the-Loop (HIL) systems withe-machine simulation on signal level

• Dyno test benches with target machine andload machine

In HIL systems the gate signals to the powerswitches are connected to a PWM-acquisitionand a real-time e-machine model, which calcu-lates the proper phase currents. The current sen-sor signals of the power inverter are stimulatedartificially through a d/a converter. A classicHIL system is only valid within its system lim-itation. Physical effects on the IGBTs, currentsensors and on the power parts are only repro-duced by simulation. Dyno test benches are

closer to the target system, because the entiredrive including power electronics and target e-machine is present. The mechanical load is gen-erated by a load machine. This configuration isvery close to reality; however it is very expen-sive, maintenance-intensive and has only limitedspeed dynamics because of the high inertia ofthe load machine. Furthermore, a cost intensivetarget machine especially designed for integratedtransmission is necessary. An e-machine emu-lator is an alternative test concept which emu-lates the power flow of the real e-machine, butcompletely without rotating mechanical parts. Inaddition, it is possible to simulate many differ-ent e-machines with little effort through varia-tion of the model parameters. Compared to theHIL system the e-machine emulator has the addi-tional benefit, that the power stage of the invertercan be operated without hardware changes. Theinterfaces between emulator and inverter are incontrast to HIL not signal lines, but phase lineswith real power flow. Four possible test configu-rations for inverter test are shown in fig. 1. Firstapproaches and prototypical implementations ofe-machine emulators were published in the endof the 1990s [1][2][3]. The described implemen-tations (here: Virtual Machine) used simple in-dustrial B6-Inverters as actuators to control the

EEVC European Electric Vehicle Congress 1

Figure 1: Test configurations for inverter tests

target phase currents. However, it turned out thatthe e-machine emulator requires much higher dy-namics and lower current ripples as the DeviceUnder Test (DUT). In publications [4] and [6]approaches were presented that reaches higherswitching frequencies, lower current ripples andan improvement of current dynamics by using amulti-level inverter. For some years commercialvendors distribute such systems, which make itinteresting for the automotive industry. In thispaper, such an e-machine emulators test bench ispresented. It focuses on the accuracy improve-ment through the application of a flux based e-machine model. Finally, the static and dynamicemulation performance of the e-machine emula-tor in comparison to a dyno test bench will beinvestigated.

2 System DescriptionThe inverter test bench shown in fig. 2 wasbuild up together with a commercial vendor fore-machine emulators (S.E.T. GmbH, Wangen imAllgau). Two power supplies are required to sup-ply the DUT and the e-machine emulator withDC voltage. The three phase wires of the DUT

Figure 2: Overview of the inverter test bench

are connected to the e-machine emulator via cou-pling inductance. Fig. 3 illustrates the struc-ture of the hardware setup. The phase currentsand voltages are measured in parallel through thee-machine emulator. The measured phase volt-ages are input variables for a real time e-machinemodel. Inside the model the target phase cur-rents will be calculated by the numerical solutionof a differential equation. The target and the ac-

tual phase currents are used as input variable tothe current controller, which controls a high dy-namic multi-level inverter. The ripple current of

Figure 3: Em-emulator hardware setup

the multi-level inverter is much smaller in com-parison to a two-level-inverter due to the highersum clock rate of the PWM. Studies on this topicand control procedures were presented in [4] and[7]. The multi-level inverter consists of parallelinverters (B6 bridges) with phase-shifted controlof the gate signals and coupling inductance be-tween each of the inverter legs. The structure ofthe the multi-level inverter is shown in fig. 4.The appropriate phase currents, which flow in

Figure 4: Structure of the multi-level inverter withfour inverter legs per phase [7]

the real e-machine, are continuously adjusted bycombining e-machine model, current controllerand multi-level inverter. The position of the e-machine is necessary for the field oriented con-trol, which is measured by a resolver, an encoderor Hall sensors. This signal is simulated elec-trically by the e-machine emulator. The voltagedetection, model calculation, current regulationand PWM generation takes place in a FPGA. Theemulation accuracy depends mainly of the per-formance of the e-machine model. This will beconsidered in the next section.

3 E-Machine Model

Many publications [4][5] dealing with e-machineemulation uses a linear PMSM-model with con-


stant inductances shown in the eq. 1.

ud = Rsid + Ldd

dtid − ωLqiq

uq = Rsiq + Lqd

dtiq + ωLdid + ωψp (1)

The disadvantage of the model is that the satu-ration of the electric motor ferrite core and thespace harmonics in the air-gaps are neglected. Inthis paper an alternative e-machine model will bepresented that overcome these restrictions. Themodel equations are described in a rotor fixeddq-coordinate system. In this case, the first stepis to transform the input phase voltages into thedq-system by the Clarke and Park transforma-tion. The equation below describes the flux basedmodel.

ud = Rsid +dψd

dt− ωψq

uq = Rsiq +dψq

dt+ ωψd (2)

The eq. 3 calculates the air gap torque.

Te =3

2p (ψdiq − ψqid) (3)

In fig. 5 the implementation of the flux basedPMSM model is shown. The ψdq2idq-Block con-

ψdq2idq

ud

uq

23p

Mload

ωM - J1

-

Rs

Rs

-

- -

ω

ψd

ψq

id

iq

ϕel

Figure 5: Implementation for the flux based PMSMmodel

tains the conversion of the flux linkages ψd andψq to the stator currents id and iq. Flux harmon-ics can optionally be added to the flux linkagesψd and ψq as function of the electrical machineangle ϕel.

ψd = ψd + ψd,harm = ψd + fhd (ϕel)

ψq = ψq + ψq,harm = ψq + fhq (ϕel) (4)

An approach to calculate the harmonic functionsfhdq(ϕel) was already presented in [8]. The eq. 5shows how the look-up tables for the harmonics

can be calculated, where ψh,n are the flux ampli-tudes of the nth harmonic of the back emf.

fhd (ϕel) =∞∑k=1

(ψh,6k−1 + ψh,6k+1) · cos (6ϕel)

fhq (ϕel) =

∞∑k=1

(−ψh,6k−1 + ψh,6k+1) · sin (6ϕel)

(5)

For considering the saturation effects, the fluxlinkages ψd and ψq has to convert to stator cur-rents. The id and iq currents components arestored in two 2D-lookup tables, where ψd and ψq

are the input values. The data acquisition of theflux tables is discussed in the next section.

id = f−11

(ψd, ψq

)iq = f−12

(ψd, ψq

)(6)

Alternatively, the d-and q-currents can be calcu-lated by constant inductances, which correspondsto the equation 1 (linear PMSM model withoutsaturation).

id =ψd − ψp

Ld

iq =ψq

Lq(7)

The preferred model can be chosen through aswitch. The fig. 6 shows the implementation ofthe Ψdq2idq Block, that contains the flux and har-monics tables of the PMSM-model. The advan-

ψd

ψq

id

iq

fhd

fhq

ϕel

+

+

f1-1

f2-1

ψp -

Ld

1

Lq

1

Mo

del

Sele

ct

ψd ^

ψq ^

Figure 6: Implementation of the Flux-to-Current-Block (ψdq2idq)

tage of the flux based model in comparison to thelinear model is the high accuracy in the simula-tion of saturation and harmonic effects.

4 Model Data AcquisitionFirst, its necessary to obtain the e-machine char-acteristics to fill the 2D lookup tables in eq. 6


with accurate data. One possibility obtaining theflux fields ψd and ψq with respect to id and iqis to perform a Finite Element Method (FEM)analysis. This solution has the benefit, that the e-machine don’t have to exist as a physical unit e.g.in early development stages. If a DUT of the tar-get machine is present, an e-machine characteri-zation on a dyno test-bench is a better choice dueto the higher accuracy. The structural representa-tion of the used dyno test bench and the measure-ment setup is shown in fig. 7. The main compo-

Inverter with

controllerEM

Te

Loadmachine

OscilloscopeSlave

OscilloscopeMaster Sync

ia ib uab ubc

Torque sensor

33uResolver

Figure 7: Measurement Setup on e-machine testbench

nents of the test-bench setup are the inverter withcontroller, e-machine and load machine. Two os-cilloscopes with together eight channels are usesin the measurement setup. Through the star-pointsymmetry just two currents ia, ib and two volt-ages uab, ubc are necessary to determine the sys-tem fully. The inverter uses an integrated resolverfor the rotor position examination. The electricalangle ϕel can be obtained by the measurementand demodulation of the three resolver signals(Sin, Cos, Ext). With the measured phase cur-rents iabc, phase voltages uabc and the calculatedelectrical angle ϕel the stator fixed values aretransformed into the rotor fixed coordinate sys-tem by the Clarke and Park transformation. Withthe transformed idq and udq components the sta-tionary magnetic flux ψd and ψq can be obtainedwith the eq. 8.

ψd = f1 =uq −Rsiq

ω

ψq = f2 =Rsid − ud

ω(8)

For a complete flux field several workpoints withvariation of id and iq have be recorded. The re-sults for ψd = f1(id, iq) are shown in fig. 8. Thelast step is a field inversion of ψ

dqw.r.t. idq.(

ψd = f1 (id, iq)ψq = f2 (id, iq)

)inverse→

(id = f−11 (ψd, ψq)

iq = f−12 (ψd, ψq)

)(9)

With the inverted flux tables the e-machinemodel in the previous section (see eq. (6)) canbe filled with data of a real e-machine . At lastthe lookup tables for the flux harmonics fhd andfhq can be filled by Fourier analysis of the targetmachine back emf with eq. (5).

Figure 8: Flux field ψd after e-machine characteriza-tion

5 Emulation ResultsIn this section the results of the emulation accu-racy are presented. In the first step the dynamicperformance of the em-emulator is investigatedthrough a step response. As an example a torquestep of 100 Nm at 400 min−1 is representedin fig. 9. In the first sub-figure the phase cur-rents with the electrical motor angle are shown.The angle ϕel has been calculated through de-modulation of the resolver signals. The secondsub-figure shows the phase voltages, where theIGBTs begin switching at timepoint 0.035 s. The3th and 4th sub-figures shows the transformedcurrents idq and voltages udq in dq-coordinatesystem. The time duration the DUT needs to con-trol the target currents to a stationary endpoint isabout 10 ms.In fig. 10 the torque step comparation be-tween e-machine emulator (EME) and real e-machine (EM) is shown. The flux harmonics one-machine emulator were deactivated to neglectthe effect of harmonic disturbance on the cur-rent controllers. The flux harmonics in the realmachine have significant impacts on the currentand voltage ripples. The current iq reached thestationary endpoint at the EM 3 ms faster as onthe EME. On EME the control rise times of bothcurrents id and iq are nearly equal whereas thecontrol rise time of the id current on EM is muchslower with high oscillations. The stator voltagesud and uq differs between EM and EME mostlyin the dynamic area, when the currents are ris-ing, between 5 and 10 V. Based on the measure-ments it can be recognized, that the em-emulatoris an idealize environment to design current con-trollers.In the next step the static performance of themodel shall be considered. The e-machinemodel, including the e-machine characteristicsmeasured on the dyno test bench, was uploadedon the em-emulator. The e-machine character-ization, described in section 4, has been done atthe e-machine emulator test bench, where severalwork-points with id, iq variation were recorded.The results are displayed as an e-motor map in


Figure 9: Torque step from 0 to 100 Nm at 400 min−1 on e-machine emulator

Figure 10: Torque step comparation between e-machine emulator (EME) and real e-machine (EM) at 1000 min−1


Figure 11: E-motor map comparation between e-machine emulator (EME) and real e-machine (EM)

Figure 12: Flux linkage field comparation of the e-machine emulator (EME) with real e-machine (EM)


Figure 13: Speed and torque workpoint variation on e-machine emulator

Figure 14: Harmonics comparation of back emf between e-machine emulator (EME) and real e-machine (EM)


fig. 11. To identify the e-machine emulation ac-curacy, the flux ellipses ψ =

√ψ2d + ψ2

q and thetorque hyperbola Tel (eq. 3) of the real e-machineand of the e-machine emulator measurements arerepresent in one figure w.r.t. id and iq. In thefigure it can be seen that the emulation accuracyis very satisfying over a wide range in torque aswell as in flux linkage. There are only minor dif-ferences of the flux linkage at work-points withhigh absolute id and low absolute iq.Fig. 12 displays the same results, whereas theflux linkage components ψd and ψq are sepa-rated. In the lower-left and lower-right sub-figurethe relative variations of the flux linkage com-ponents are shown. The Mean Absolute Error(MAE) of the ψd component is with 1.1% almosttwice as large as the ψq component with 0.59%.In sum, this is a very satisfactory result, becausethe parameters scattering through the productionof electric motors is higher than the error of theemulation.Up to this point, all the measurements has beendone at a fixed rotor speed in the area of constanttorque. In the next step the e-machine emulatoraccuracy has been investigated in the flux weak-ening area. Therefore, a set of work-points withrotor speed n and torque demand T ∗ variationhas been recorded at the dyno- and e-machineemulator test-bench. The result are several mapswith id, iq, ud and ud w.r.t n and T ∗, where udand uq are the output voltages of the current con-troller. The currents id and iq are the processvariables and controlled by current controllers.Fig. 13 displays the results obtained from thedifferences of the dyno- and e-machine emula-tor measurements. It should be noted that theflux weakening area begins at 1200 min−1. Thed- and q-currents in the area of constant powermatches exactly. Inside the flux weakening areathere are some differences of id between 4-14%.However iq has minor divergences with a MAEof just 0.2%. In the sub-figures bottom left andbottom right the controller output voltages whereshown. The MAE of ud is just 1.3% and theMAE of uq is 2.6%. As can be seen the error inthe area of constant torque is less than inside theflux weakening area. One reason for the differ-ences may be the higher line contact resistancesat e-machine emulator, which triggers to a volt-age drop on the phase lines. Through the volt-age drop the inverter controller increase the neg-ative d-current to reduce the d-flux. So the higherdeviations of id in the flux weakening area canbe explained. In the next time the test benchphase connections will be optimized to reducethe phase line voltage drop off. Last but not leastthe capabilities of the em-emulator to replicateflux harmonics has been examined. Therefore,the back emf of the real e-machine has been an-alyzed with a Fourier analysis. The lookup ta-bles of the harmonics has been calculated witheq. 5 and uploaded to the em-emulator. The backemf comparison between real e-machine and em-emulator are illustrated in fig. 14. The back emfhas a distinct 5th harmonic, which has been repli-

cated from EM-emulator with good precision.

6 ConclusionsIt has been shown that an e-machine emulator isan equal and inexpensive alternative to conven-tional dyno test benches for the test of electricvehicle inverters. The advantages are: no movingparts and no maintenance, flexibility, by chang-ing the model another motor type can be emu-lated. The challenges are: Data from the reale-machine is needed for the flux look-up tables.These must be acquired from measures of a dynotest bench. Alternatively, data can be acquiredfrom a FEM analysis. Generally the flux linkagedeviations are under 1%, which is less than theproduction tolerances for most electric motors.The largest deviations between model and mea-surements has been measured in the flux weak-ening area with 4-14%. Generally the system isperforming satisfactory, and will definitely be apart of the future development of electric motordrives and controllers.

References[1] H. Slater Real Time Emulation Environment for

Digital Control Development, PhD thesis, New-caste upon Tyne, Newcastle University, 1996.

[2] A. Jack, D. Atkinson and H. Slater, Real-timeemulation for power equipment development.Part 1: Real-time simulation, Electric PowerApplications, IEE Proceedings Vol 145, No. 2,1998

[3] H. Slater, D. Atkinson and A. Jack, Real-timeemulation for power equipment development.Part 2: The virtual machine, Electric PowerApplications, IEE Proceedings Vol 145, No. 3,1998

[4] S. Trabelsi Umrichterprufung mit Hardware-in-the-Loop und Einsatz einer neuar-tigen schnellen oberschwingungsar-men Leistungsendstufe, PhD thesis,ISSN 9783183364213, VDI-Verl., 2004

[5] B.Cebulski and M. Thom, Power-HiL-Simulation von Elektroantrieben fr Hybrid-fahrzeuge, Internationaler ETG-Kongress, 2007

[6] S. Grubic et.al, A High-Performance ElectronicHardware-in-the-Loop Drive - Load SimulationUsing a Linear Inverter, IEEE Transactions onIndustrial Electronics, 57(4), 12081216., 2010

[7] C. Nemec, J. Roth-Stielow Investigations onthe ripple current of a multiphase interleavedswitched inverter, Power Electronics and Appli-cations Conference, 2011


[8] M. Meyer Wirkungsgradoptimierte Regelunghoch ausgenutzter Permanentmagnet-Synchronmaschinen im Antriebsstrangvon Automobilen, PhD thesis, University ofPaderborn, 2010

AuthorsStefan Uebener studied Computerengineering at Technical Univer-sity of Ilmenau, Germany, from2004 - 2010. He is working asphd student at Daimler AG in Sin-delfingen since 2010 in the field ofe-machine analyse and emulation.

Prof. Dr.-Ing. Joachim Bockerstudied electrical engineering atthe Technical University of Berlinand received his phd there in 1988.He worked at AEG Research In-stitute, later Daimler Research andDevelopment, in Berlin from 1988-2001. Since 2003 he is profes-sor for Power Electronics and Elec-trical Drives at the University ofPaderborn.


application of an e-machine emulator for power converter ... · application of an e-machine...

Documents