Development and Series Application of a Vehicle Drivetrain Observer Used in Hybrid and Electric Vehicles

Transcription

1 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0364 EVS27 Barcelona, Spain, November 17-20, 2013 Development an Series Application of a Vehicle Drivetrain Observer Use in Hybri an Electric Vehicles Dr. Gunther Götting 1, Markus Kretschmer 2 1 Robert Bosch GmbH, Postfach , Stuttgart, gunther.goetting@e.bosch.com Abstract This paper introuces the evelopment an calibration process of a vehicle rivetrain observer use in hybri an electric vehicles for active amping control (ADC) an for the improvement of the electric machine s rotor angle signal uality. This approach starts with creating an overall vehicle moel that inclues the electric machine, the transmission, sie shafts, the tires an the vehicle boy. For control engineering purposes, that multi-orer-moel is then reuce into a two-mass-oscillator which can be easily escribe in state-space form. Using this reuce rivetrain moel an applying a Luenberger observer approach, not only the signal uality of both the instrumente rotor angle an the spee of the electric machine can be improve consierably but also the oscillation ynamics of this vehicle rivetrain can be estimate. If not compensate uring vehicle operation, rivetrain oscillations might lea to increase rivetrain wear, NVH issues an limite rie comfort; therefore, the oscillation spee is very important in computing an active amping torue that is to compensate rivetrain oscillations. Calibration of the vehicle rivetrain observer is one using specific vehicle test ata that are fe into a stanalone calibration tool ientifying the parameters of the vehicle rivetrain as well as the Luenberger feeback vector. Base on these ata, a proper active amping control application is set-up an verifie in various vehicle tests an to lea to the calibration finally to the application in several hybri an electric vehicle series projects (e.g. Peugeot 8 HYbri4). Keywors: active amping control, hybri vehicles, electric vehicles, observer, rivetrain ynamics 1 Introuction Hybri (HEV) an electric vehicles (EV) o show a uite ifferent vehicle rivetrain setup compare to conventional, combustion-engineriven-only vehicles. Electric vehicles an several hybri vehicles usually lack a launch element like a friction clutch or a torue converter which prevents the combustion engine from stalling uring vehicle launch. When launching a HEV or an EV from stanstill using the electric machine, there is no nee for a friction clutch since an electric machine is able to provie torue from zero spee alreay. In this case however, torue excitations on a vehicle rivetrain that oes not inclue some sort of launch element result in e-machine spee oscillations that not only limit rie comfort but also ten to significantly increase rivetrain wear. To provie a proper electric machine torue, the machine s rotor angle has to be known very EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 1

2 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0365 accurately in computing the phase currents to setup the machine torue. This paper introuces a control engineering approach that uses a simplifie vehicle rivetrain moel within a Luenberger observer to both estimate rivetrain oscillations an improve the signal uality of the measure rotor angle of the electric machine. I.e. by using this engineering approach (1) the positioning angle uality is improve leaing to an improve torue an current control of the electric machine an (2) the rivetrain oscillations can be estimate which are necessary to compute an active amping torue compensating such rivetrain oscillations. Calibration of the rivetrain observer moel is one using vehicle test ata (e.g., electric machine torue an spee) an feeing it into a stanalone parameter ientification tool that computes rivetrain an observer feeback parameters. Applying these parameters into the electric machine s torue an current control loop an properly calibrating the active amping control, rivetrain oscillations can be reuce consierably while improving rie comfort. 2 Vehicle Drivetrain Dynamics A vehicle rivetrain consists of several mechanical elements (e.g. engine/electric machine, transmission, ifferential gearbox, sie shafts, tires, ) which all introuce its respective characteristics (e.g. inertia, stiffness) into the overall rivetrain. Base on these rivetrain parts an elements, it might be possible to come up with a very sophisticate simulation moel of a vehicle rivetrain. Such a multi-orer rivetrain moel, however, is too complex to be use in real-time software applications controlling an electric machine. Owing to this fact, the vehicle rivetrain nees to be moele ifferently without neglecting ominant rivetrain ynamics. Using an electric axle rive as a reference vehicle powertrain, Figure 1 shows how to reuce moel complexity of a vehicle rivetrain moel leaing to a so calle two-massoscillator. The ynamics of a rotational twomass-oscillator can be represente by applying the torue euilibrium for inertias J 1 an J 2 E. (1) where T Em inicates electric machine torue acting on inertia J 1 while T Loa represents the rag or loa torue exerte on inertia J 2 (being the euivalent of vehicle mass plus wheel inertia). T Em, Des T Loa J 1 k J 2 k k c k c c c Figure 1: Vehicle Drivetrain Moeling (1) Parameters c an k are the corresponing stiffness an amping coefficients of the reuce rivetrain moel. Inices an inicate angle, rotational spee an rotational acceleration of inertias J 1 an J 2, respectively. It might be note here, that most vehicle rivetrains can be escribe using such a reuce two-massoscillator moel. 3 Torue an Current Control In normal rive operation, the electric traction rive of an EV/HEV is operate in torue control moe, i.e. the vehicle manager commans a esire torue, erive from the accelerator peal. In recent years, permanent magnet synchronous machines (PMSM) are wiely use for this application. This machine type reuires an accurate rotor angle signal to provie a proper torue output by fiel oriente control (FOC) [1]. The electromagnetic torue T Em,Des can be erive from the / components of the stator current in the fiel oriente reference system as shown in E. (2). Remark: Electrical values are given in rms values, here as well as in the following. T 3 p I Ψ I L L (2) Em, Des PM The corresponing voltage euations are given in E. (3). I U Rs I L L I t (3) I U Rs I L L I Ψ PM t EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 2

3 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0366 An error in the rotor angle irectly influences the irection of the fiel oriente reference system an therefore etunes the irection of the controlle stator current phasor compare to its esire value. Figure 2 emonstrates how a given angle error of the stator current phasor I Stator is propagate to a torue error. 3 Current Torue UStator@OP1 OP i in A I / A, U / V OP2 IStator@OP2 OP1 angle error IStator@OP1 Figure 2: Angle error of stator current phasor an resulting torue error of a typical traction rive PMSM for ifferent operating points (OP1 an OP2) It is obvious, that the effect of an angle error on the resulting torue error is epenent on the operating point (OP). If the rive is operate in the base spee region following the maximum torue per ampere curve (MTPA), see e.g. [2], an angle error of 5 el. as shown in Figure 2 results in a torue error < 1% (OP1). Compare to this, in the fiel weakening area, the same angle error leas to an increase torue error of about % (OP2). Aitionally, a etune current phasor results in a etune voltage phasor. This is emonstrate in 3 I / A, U i in / A V Figure 2, where the resulting voltage phasors are shown for the three ifferent current phasors of OP1, calculate from E. (3). From a control point of view, this effect can be seen as a isturbance voltage, epenent on the angle error. In actual traction rive applications, the rotor angle is usually measure by a high resolution sensor system like e.g. a resolver (resolution: 1 el.). In aition, epening on the vehicle setup, also a igital sensor system featuring a uite low angle resolution may be use (resolution: el.). In the latter case, the sensor output cannot be use irectly but its resolution must be refine by aeuate extrapolation algorithms. However, inepenent of the sensor system in use, the measure angle signal is isturbe by noise an affecte by mounting an prouction tolerances, which, in turn, might result in corrupte phase currents as mentione above. This introuces an aitional torue ripple to the rivetrain an might even increase the rivetrain oscillations. Earlier investigations [3] have shown, that the overall performance of the torue control can be significantly improve, when the fiel angle use for the FOC is the output of an aeuate state observer instea of irectly using the sensor signal. Figure 3 shows a block iagram of the resulting torue control loop with the fiel oriente current controller incluing an observer an the physical system consisting of the inverter, the electric machine an the mechanical rivetrain. Since the observer structure is base on a mechanical moel of the rivetrain, see chapter 4, it elivers besies the rotor angle aitional internal state variables as well as the wheel an the rotor spee which can be use to implement an active amping control. It might be note here that the hat symbol ^ enotes an estimate number. Figure 3: Block iagram of torue control loop using an observer for provision of proper rotor angle an internal rivetrain variables for active amping control EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 3

4 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0367 angle / el. t / s rotor angle sensor angle observer angle Figure 4: Comparison of physical rotor angle, raw signal of the low resolution sensor an output of an angle observer If a low resolution angle sensor is use, the observer can help to reuce the angle error uring extrapolation time, i.e. the time interval between two sensor pulses. In Figure 4, the output of an angle observer is compare to the raw signal of a low resolution igital sensor system, here without extrapolation, an the physical rotor angle. Obviously, the observer signal can follow the physical angle uite well, though using a low resolution sensor as feeback input. For best results it is of particular importance to carefully aapt the structure an the internal parameters of the observer to the vehicle rivetrain, as will be shown in chapter 5. In case of a high resolution angle sensor, the maximum angle error is usually uite low. Nevertheless, the raw signal of the sensor is affecte by perioic an stochastic isturbances cause by mounting tolerances, signal processing an measurement noise. This, in turn, leas to isturbances on the measure fiel oriente current as well as on the output voltage of the current controller, as mentione above. Eviently, the current ripple can be clearly reuce by using the observer output instea of the raw signal. Since the observer is able to improve the uality of the angle signal, the isturbance voltage which results from sensor noise an acts on the output voltage of the current controller can be eliminate. 4 Drivetrain Observer an Active Damping Control The two-mass-oscillator moel introuce in chapter 2 sets the basis for computing the oscillation spee Osc (4) of a vehicle rivetrain. Using E. (1) an applying an electric machine torue of T Em,Des = Nm, T Loa = 0 Nm, J 1 = 0.02 kgm 2, J 2 = 2.0 kgm 2, c = Nm/ra an k = 0.1 Nms/ra, the rotational spees an as well as the oscillation spee Osc can be compute (Figure 6). without observer with observer I,Des Figure 6: Drivetrain oscillations I in A I,Des t in s Figure 5: Performance of current control with HFisturbances on rotor angle signal, without an with observer Figure 5 shows a measurement result of a fiel oriente current control with a high resolution angle sensor. While for t < 2.2 s the fiel angle is irectly erive from the raw signal of the sensor, for t > 2.2 s an aeuate observer is use for this task. Looking up the torue control loop escribe earlier (Figure 3), it can be notice that the positioning angle of the electric machine an the esire torue are the only input signals fe into the rivetrain observer the oscillation spee Osc, however, cannot be measure without proper wheel spee information. To compute oscillation spee, therefore, some sort of estimation algorithm is reuire to provie the oscillation spee estimation using the positioning angle, the esire torue an the two-mass-oscillator moel. For this purpose, a so calle Luenberger observer is use to estimate the necessary system state Osc EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 4

5 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0368 information reuire to apply an active amping torue. 4.1 The Luenberger Observer A Luenberger observer [3, 4] combines a moel of a real-worl process (e.g. a vehicle rivetrain) with one or several internal measurement signals erive from the real-worl process ( y ) leaing to moel state estimates which cannot be measure (Figure 7). From control engineering textbooks, one can erive the state-space representation of a linear, time-invariant system shown in E. (5): x A x B u y C x D u The matrices an vectors are efine as follows: x state vector x time erivative of state vector x A system matrix B input matrix C output matrix D feeforwar matrix u input signal(s)/vector y output signal(s)/vector (5) ˆ Osc. The former estimate can be use to improve torue an current control, the latter is use to compute a proper active amping torue. 4.2 The Active Damping Control To compensate rivetrain oscillations (Figure 6), the electric machine torue acting on a vehicle rivetrain nees to be moulate. I.e. epening on the esire torue T Des erive from the river s input on the accelerator (1) the oscillation spee ˆ Osc has to be compute an, base on this number, (2) a so calle amping torue T Dmp is to be create an subtracte from T Des leaing to the (3) electric machine torue T Em,Des (Figure 8). The amping factor k Dmp converts the oscillations spee ˆ Osc into the amping torue T Dmp, see also [6, 7]. Note, that k Dmp might vary epening on input torue T Des an electric machine spee ˆ. Figure 8: Creating Damping Torue T Dmp Applying E.(5) on the Luenberger observer, the state-space euations become: xˆ Â xˆ Bˆ u L ŷ Ĉ xˆ y ŷ (6) Note, that L is the so calle Luenberger feeback vector (or matrix) which is necessary to connect the observer with the real-worl process via the estimation error y ŷ. Figure 9: Active Damping Control (ADC) Operation Figure 7: Luenberger Observer This observer structure then provies estimates of the rotor angle ˆ an the oscillation spee Applying the same parameters that have le to the results epicte in Figure 6 an setting k Dmp = 0.1 Nms/ra, rivetrain oscillations can be compensate significantly (Figure 9). Since a vehicle rivetrain features spring elements (e.g. rive shafts) any change in torue acting on the rivetrain leas to a change of the relative motion between the electric machine an the vehicle s wheels. Therefore, the spee peak at t 0.02 s cannot be fully compensate it is rather the effect of applying tension on the vehicle rivetrain. EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 5

6 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page Ientifying Vehicle Drivetrain Parameters Knowing a physical representation of a vehicle rivetrain which contains four parameters only (J 1, J 2, k, c) to moel the two-mass-oscillator, one nees to fin correct parameter settings for vehicle applications. To provie a uick an easy way to ientify such rivetrain parameters, a stanalone software tool was evelope allowing calibration engineers to efficiently o both testing an classification of ientifie rivetrain parameters without having to use MATLAB or other available state-of-the-art tools. 5.1 Parameter Ientification Tests Control engineering textbooks introuce several test signals resulting in a system response which can be use to ientify unknown system parameters. To investigate rivetrain parameters, the test vehicle whose rivetrain ynamics has to be ientifie is place on a flat, even roa. Then an electric machine torue step is exerte on the rivetrain measuring the electric machine spee (Figure 10). Figure 11: IPE-ADC Stanalone Tool Comparing the results of the initial vehicle test (Figure 10) with the ientification results shown in (Figure 12), it can be note that the IPE-ADC software provies sufficient two-mass-oscillator parameter estimation for moeling a vehicle rivetrain. By looking up the ratio between torue input (40 Nm) an the first spee peak ( rpm), one can also get a first hint on how to set-up the amping factor k Dmp (Figure 8). After setting up both the moel parameters an the Luenberger observer, the active amping control an the observer s angle estimation algorithms proviing ˆ have to be teste in everyay riving an in several specific vehicle tests. Figure 10: Drivetrain Ientification Test These torue an spee ata are then fe into the stanalone software tool IPE-ADC (Figure 11). For each of the four rivetrain parameters a specific search space is efine setting the basis for ientifying J 1, J 2, c an k. Besies the ientification routine, IPE-ADC also provies the numbers for the Luenberger feeback vector L. Once the proper parameters have been estimate an the observer feeback vector is etermine, a software calibration file is create which can be easily use to enable proper active amping control operation of the inverter software. Figure 12: Drivetrain Ientification Results EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 6

7 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page Vehicle Testing To valiate active amping control operation an the signal uality of the estimate electric machine rotor angle ˆ, various vehicle tests nee to be performe. This chapter presents test results that feature some specific vehicle tests: Vehicle Launch Torue Loa Changes Siewalk Climb 6.1 Vehicle Launch In chapter 5.1, the ientification process to investigate rivetrain parameters was introuce. One coul notice significant spee oscillations cause by an e-machine torue step an the ynamics of the vehicle rivetrain (Figure 10). In everyay riving however, such torue steps o not occur the vehicle control unit rather converts the river s input (via the accelerator) into a pre-filtere set-torue T Des that the electric machine is to eliver. But even if the esire settorue is create by some sort of filtering algorithm, the vehicle rivetrain still represents an unerampe system with only little amping. Figure 13: Vehicle Launch without ADC Figure 14: Vehicle Launch with ADC Therefore, there is still the necessity to apply a proper active amping control to establish a smooth riving behavior. Figure 13 an Figure 14 compare a vehicle launch test without an with ADC operation active, respectively. The test result without ADC shown in Figure 13 emonstrates how the vehicle rivetrain is excite to oscillations by torue ramps, which can be seen from the rotor spee signal. Due to the low physical amping, these oscillations ecrease very slowly, which leas to a long-lasting isturbance of riving comfort. However, when the vehicle launch is performe with activate ADC (s. Figure 14), the rotor spee signal shows a very smooth behavior, inicating that rivetrain oscillations are well ampe by the interaction of the amping control. It is remarkable, that in this situation only a very low amount of amping torue is sufficient to prevent the rivetrain from oscillating. To achieve this excellent performance, it is inispensable to carefully tune the observer parameters to the specific vehicle rivetrain, as shown in chapter Torue Loa Changes In vehicle tests featuring torue loa changes, the river applies an releases the accelerator repeately; leaing to significant torue inputs (Figure 15, Figure 16). When releasing the accelerator, it might happen that the electric machine is commane into regeneration moe, i.e. the electric machine torue becomes negative (while riving forwar). Such a sign change in electric machine torue not only leas to a significant change in vehicle acceleration, it also results in a tooth flank contact change because of transmission play. Since the two-mass-oscillator moel oes not feature such transmission play, it is vital to test whether or not the observer can cope with that kin of moel uncertainty. As seen from Figure 15, the riving situation specifie above leas to significant rivetrain oscillations, again to be etecte by the oscillations in the rotor spee signal. Figure 16 shows the test results for the same riving situation, this time with ADC turne on. Right after a sign change in torue (t 145 s, t s) a significant peak in active amping torue T Dmp can be notice. This torue peak is cause by releasing the rivetrain spring (which basically consists of the vehicle s sie shafts) an running the rivetrain through the transmission play. Obviously, the propose active amping control approach can eal very efficiently with these effects. EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 7

8 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0371 Figure 15: Torue Loa Changes without ADC Figure 17: Siewalk Climb without ADC Figure 16: Torue Loa Changes with ADC Figure 18: Siewalk Climb with ADC 6.3 Siewalk Climb The siewalk climb is probably one of the most challenging tasks for both torue/current control an active amping control application: The vehicle s riving wheels are place right at a siewalk; then, the accelerator is applie an the electric machine is to provie enough torue (here: > Nm) while being operate at very low rotational spees such that the vehicle s riving wheels crawl up the siewalk, basically lifting up half the vehicle weight (Figure 17, Figure 18). This exerts not only significant mechanical wear on the rivetrain but also a consierable thermal stress on the inverter itself. When performing this test without ADC, the siewalk climb process goes along with strong rivetrain oscillations, s. Figure 17. An activate amping control can clearly reuce these oscillations, resulting in a smoother an more comfortable behavior, as shown in Figure Conclusions This paper provies some insight on the necessity to improve the signal uality of rotor angle measurements use in the torue an current control of hybri an electric vehicles as well as the nee to efficiently compensate vehicle rivetrain oscillations. Since such oscillations o not only result in consierable wear but in rie comfort issues as well, there is an essential interest in creating a proper electric machine control that leas to both smooth an sportive riving of hybri an electric vehicles. The propose observer base solution turns out to be very successful in improving the uality of the rotor angle signal, use for the FOC, as well as in estimating an aeuate oscillation signal as input for an active amping control. For best results, the tuning of the observer parameters can be precisely one with the help of the presente calibration tool. The control engineering approach presente here, has foun its way in several current an future hybri an electric vehicle series projects (e.g. Peugeot 8 HYbri4). EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 8

9 Worl Electric Vehicle Journal Vol. 6 - ISSN WEVA Page Page 0372 References [1] S. Macminn an T. Jahns, Control Techniues for Improve High-Spee Performance of Interior PM Synchronous Motor Drives, IEEE Trans. on In. Appl., Vol. 27, No. 5, Sept/Oct [2] B. Cheng, T. Tesch, Torue Feeforwar Control Techniues for Permanent-Magnet Synchronous Motors, IEEE Trans. on In. Appl., Vol. 57, No. 3, March 2010 [3] G. Götting, Ermittlung er lage zur Regelung einer PSM in Hybir- un Elektrofahrzeugen, AUTOREG 2011 [4] Norman S. Nise, CONTROL SYSTEMS ENGINEERING, ISBN , John Wiley & Sons, 0 [5] Otto Föllinger, Regelungstechnik, ISBN , Heielberg, Hüthig, 1994 [6] M. Menne, R. W. De Doncker, Active Damping of Electric Vehicle Drivetrain Oscillations, Proceeings of Power Electronics an Motion Control Conference (EPE-PEMC) 9, Kosice, 0 [7] G. Götting, R. W. De Doncker, Comparison of Algorithms for Suppression-Control of Drivetrain-Oscillations in Electric Vehicles, Proceeing of Electric Vehicle Symposium (EVS) 19, Busan, 2 Authors Gunther Götting hols both a iploma an a Ph.D. in Electrical Engineering from RWTH Aachen, Germany. He is a senior expert in electric machine control an also works in the fiels of observer evelopment an active amping control. Gunther Götting might be contacte via gunther.goetting@e.bosch.com Markus Kretschmer hols a iploma in Mechanical Engineering from the University of Stuttgart, Germany. His fiels of work inclue active amping control, observer an control engineering applications as well as vehicle tests. Markus Kretschmer might be contacte via markus.kretschmer@e.bosch.com EVS27 International Battery, Hybri an Fuel Cell Electric Vehicle Symposium 9