630 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 9, NO. 2, MAY 2013

Transcription

1 630 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 9, NO. 2, MAY 2013 Development of High-Reliability EV and HEV IM Propulsion Drive With Ultra-Low Latency HIL Environment Evgenije M. Adžić, Member, IEEE, Milan S. Adžić, Vladimir A. Katić, Senior Member, IEEE, Darko P. Marčetić, Member, IEEE, and Nikola L. Čelanović, Member, IEEE Abstract This paper proposes an improvedandrobustmethod of minimizing the error in propulsion-drive line-currents that are reconstructed from a single dc-link current measurement. The proposed algorithm extends and then shortens the relevant phase pulse-widths in order to provide optimal sampling of the dc-link currents in two consecutive pulsewidth modulation (PWM) periods. The proposed PWM pattern control enables an improved sampling method which cancels offset jitter-like waveform errors present in all three reconstructed line-currents, which is due to a specific combination of nonsimultaneously sampled dc-link current and line-current PWM ripple. The improvement in induction motor drive accuracy using a single current-sensor and no shaft sensor (as proposed in this paper), over that of conventional methods, is shown. Thanks to an ultra-low latency hardware-in-the loop (HIL) emulator, the proposed algorithm, its implementation on a DSP processor, code optimization and laboratory testing were all merged into one development step. In order to perform final tests of the proposed current-reconstruction algorithm and to verify the usefulness of the developed HIL platform by means of comparison, experimental results obtained on a real hardware setup are provided. Index Terms Current measurement, electric vehicles (EV), hardware-in-the-loop (HIL), induction motor (IM) drives. I. INTRODUCTION WITH a large number of microprocessors controlling the majority of functions in a modern automobile, it is impossible to overemphasize the importance of hardware-in-the-loop (HIL) testing infrastructures. In fact, together with the microprocessor, software testing within a HIL framework is what has made all the increased safety and comfort features of modern automobiles possible. More recently, with the fast paced development of hybrid electric vehicles (HEVs) Manuscript received December 06, 2011; revised March 07, 2012, May 13, 2012, and July 17, 2012; accepted September 17, Date of publication October 04, 2012; date of current version January 09, This work was supported by the Ministry of Education and Science of Republic of Serbia under Project III Paper no. TII E. M. Adžić, V.A.Katić, D.P.Marčetić, andn.l.čelanović are with the Faculty of Technical Sciences, University of Novi Sad, Novi Sad 21000, Serbia ( evgenije@uns.ac.rs; katav@uns.ac.rs; damar@uns.ac.rs; nikolace@uns.ac.rs). M. S. Adžić is with the Subotica Tech College of Applied Sciences, University of Novi Sad, Novi Sad 21000, Serbia ( adzicm@vts.su.ac.rs). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /TII and electric vehicles (EVs), power electronics and digital control technology has entered another safety-critical function of the modern automobile; the propulsion. The propulsion system consists of one or more AC electric motors controlled by power electronic converters and supplied by energy stored in batteries or produced by fuel-cells or by an electric generator. Real-time simulation(rts)andhilsupport have been recognized as crucial tools for assisting engineers in efficient power electronics (PEs) and motors controls development [1] [3] and to this end, HIL emulators are increasingly exploited [4] [7]. This is known to be a very efficient method but some improvements are still needed. A recently proposed RTS/HIL platform based on a novel ultra-low latency (ULL) processor design, has greatly enhanced in-depth research into improving the control strategies of electric drives [8]. This paper addresses two key aspects of drives technology, which are not restricted to but are of particular interest for the automotive industry: 1) sensorless control technology that holds the key to reliable and maintenance-free operation under all operating conditions, and 2) drives-specificullhilthatdramatically accelerates the pace of product prototyping. In order to reach the full potential of power electronics in the field of motor control and satisfy the high expectations of the automotive market, reliability and control strategies must be significantly improved [1]. Large number of sensors, cable harnesses and signal-conditioning circuitry deteriorate system reliability, increase maintenance needs and raise system costs. The reliability of a speed sensor attached to a motor shaft has been recognized by academia and industry alike as a liability, resultinginthedevelopment of many shaft-sensorless techniques [9] [14]. In a first step, speed information from a shaft sensor is substituted by a speed-estimation using motor voltage and measured motor phase-currents [9] [12]. A second step in the effort to minimize the number of sensors and also the focus of this work is to substitute the two motor phase-current sensors by asingledc-link current sensor and a smart reconstruction algorithm [15]. Most of the problematic issues in this area, including unobservable, low modulation-index operation, partially observable sector transition instances and also ever-present current measurement phase-shifts have already been resolved [15] [24]. However, the application of a single-current sensor scheme in vector controlled drives with current regulation reveals more outstanding issues. The nature of the offset jitter-like waveform error in reconstructed line-currents, due to nonsynchronized /$ IEEE

2 ADŽIĆ et al.: DEVELOPMENT OF HIGH-RELIABILITY EV AND HEV IM PROPULSION DRIVE WITH ULL HIL ENVIRONMENT 631 Fig. 1. ULL HIL experimental setup with controller and interface board. Fig. 2. IM drive with single dc-link current sensor (line-currents measurement used for comparing and testing purpose). sampling and motor line-current ripple at pulsewidth modulation (PWM) frequency, is explained in detail [25]. As this kind of distortion is especially apparent during light loads and machine-flux reduction, some kind of correction technique is needed. Since on an urban driving cycle, a vehicle s traction machine operates most frequently at light loads and a wide speed range of 3 4 times the nominal speed is required [26], [27], this issue becomes important. After presenting the problem with a conventional reconstruction method, this paper proposes a simple correction scheme for eliminating the offset jitter-like waveform errors in the reconstructed line-currents. The reliability improvement of the complete single current-sensor/shaft-sensorless drive is demonstrated using a HIL drives emulator and finally proved using a real hardware (HW) test-bed. II. HIL DESCRIPTION In the early stages of the control algorithm development and further, in order to validate the effectiveness and reliability of the proposed current-reconstruction method, the new ULL HIL test platform was used. The ULL HIL emulator comprised a programmable, application-specific processing architecture based on a FPGA board with fast analog/digital input/output interfaces and with a supporting software tool-chain performing the function of power electronics circuit analyzer/compiler [8]. This approach provided real-time execution with a 1 emulation time-step including the input/output interface latency. ULL allowed the use of a relatively high, e.g., 2 khz, switching frequency for the PWM, throughout all the experiments of this paper. A PWM frequency of 2 khz does not represent the ULL HIL limit but is set by the used real HW limitations (bandwidth of the dc-link current sensor) and for the sake of comparing ULL HIL responses with those obtained by a real HW system. The HIL emulator s ULL properties proved to be particularly helpful during the control algorithm design because its time-interrupt handling nature requires precise PWM signal shifting and fine alignment of the sampling times inside the short PWM periods of 500.Fig.1showstheULL HIL setup for evaluation of the sensorless induction motor (IM) drive with the proposed line-current reconstruction method. The same controller interfaced with the ULL HIL system, based Fig. 3. Three-phase PWM pattern with two active voltage vectors and corresponding dc-link current waveform. Sampling moments included. on the Texas Instruments DSP TMS320F2812, also controls the real HW system. III. IM MOTOR DRIVE AND LINE CURRENT RECONSTRUCTION TECHNIQUE The most reliable three-phase motor line-current measurement is the use of only one current transducer placed in the inverter dc circuit, as shown in Fig. 2. Fig. 2 shows the ULL HIL software Schematic Editor/Circuit Compiler which allows users to build a variety of PE configurations such as the induction motor drive, which is the object of this paper. The correlation between the dc-link current and the motor line-currents depends on the states of the inverter switches. There are six switch-state combinations which yield the active voltage vectors and two which result in nonactive or zero vectors ( and ). During the application of nonactive vectors, the dc-link current is always zero, while during the application of the active vectors, it is always related to one of the line-currents [15]. A typical 3-phase PWM period contains two active voltage vectors that are on long enough for current reconstruction, as showninfig.3.thefigure shows three PWM signals for the voltage vector in the first voltage sector, in which phase A has the highest and phase C the lowest, voltage command. During each of the two active voltage vectors, and,onemotor

3 632 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 9, NO. 2, MAY 2013 Fig. 4. Areas in frame in which current reconstruction is not possible. line-current can be sampled, and, respectively. Since the sum of all three line-currents flowing to the motor must be zero, measuring the two line-currents allows a determination of the remaining third motor line-current, which stays unobservable during the entire PWM period ( in Fig. 3). For the three-phase current reconstruction principle to work, both dc-link current-sampling instants must be precisely synchronized with the PWM pulses, usually referred to the beginning or the middle of the PWM period (Trig signal in Fig. 3). According to Fig. 3, the optimal sampling instants should be calculated relative to the active vector transition moments. However, for a reliable dc-link current reading, the signal sampling must take place after an additional, precalculated delay, The sample delays include total switching-device turn-on delay-time, dc-link current measurement signal rise-time and signal settling-time. The parameter includes a dead-time that is automatically inserted by the DSP, IGBT driver response-time, PWM signal-processing time and worst-case switching-device (e.g. IGBT) on-time delay. The values of all the delays involved in the dc-link current sampling process are of the order of a few and this indicates that only a low-latency real-time emulator can provide adequate accuracy. A. Modified PWM Pattern and Sampling Instants for Reliable Line-Current Reconstruction Practical difficulties with dc-link current measurement can occur when a PWM period includes active vectors that are present only for a short time, which happens in two common cases (Fig. 4). The first case is a low modulation index, when both active vectors are present for a short time. The second case occurs regardless of the modulation index, when the reference voltage vector passes near or falls on one of the six active vectors and produces at least one short active vector. During these situations, the method may be unable to reliably measure and calculate the motor line-currents. Therefore, it is desirable to use one of the suggested methods which achieve effective dc current measurement by modifying (1) Fig. 5. Modified PWM pattern: the PWMB signal is shifted to the right to form the first sampling-window wide enough for reliable current reading. the three-phase PWM voltage patterns sufficiently, if longer active vectors are needed [15] [21]. Otherwise, one can estimate motor line-currents during nonmeasurable periods without modifying the PWM pattern using mathematical machine models and the reference voltage vector information [22] [24]. In this paper such a PWM scheme is used [18]. Fig. 5 shows the applied solution with a recorded example where the first active voltage vector on the lagging side of the PWM period is less than the predetermined shortest active vector time or minimal dc current sampling-window where is the minimum time during which the dc-link current signal has to be present at the DSP analog inputs after the sampling has been initiated. Generally, the PWM signal associated with the phase having the middle voltage command is shifted to the right in order to form a sufficiently long first sampling window on the lagging side of the PWM period. Then, the duration of the succeeding voltage vector is calculated and if needed, the PWM signal associated with the phase having the highest voltage command is also shifted to the right in order to form a second sampling window. After modifying the PWM, a dc-link current measurement can be reliably taken at the first possible instant. The sampling close to the beginning of the short active vector would result in its lowest possible extension and would have less impact on the current ripple and result in less audible noise. An implementation of precise PWM signal-shifting method and correct alignment of the sampling signals with beginning (or middle or end) of the active vectors can be accurately simulated and rapidly verified only by using an ULL HIL emulator. IV. RECONSTRUCTED MOTOR LINE-CURRENT WAVEFORM ERROR DUE TO CURRENT RIPPLE Regardless of the selected PWM shifting method for line-current reconstruction, the end results cannot be the line currents sampled in the middle of the PWM period. To make things even worse, two observable line-currents cannot be sampled simultaneously, each is sampled when available and thus with a certain time-displacement between the two samples. As a result, these two values are phase-shifted differently from the middle point of the PWM period, in which the average PWM value of the (2)

4 ADŽIĆ et al.: DEVELOPMENT OF HIGH-RELIABILITY EV AND HEV IM PROPULSION DRIVE WITH ULL HIL ENVIRONMENT 633 TABLE I RIPPLE ERROR IN RECONSTRUCTED LINE CURRENT Fig. 7. Experimental results reconstructed line-current samples in comparison to the line-current average values during PWM periods for no-load operation (25 Hz, 72 Vrms, 1.2 Arms). TABLE II FIRST ACTIVE VOLTAGE VECTOR COMPONENTS THROUGH EACH SECTOR Fig. 6. HIL experimental results reconstructed line-current samples in comparison to the line-current average values during PWM periods for no-load operation (25 Hz, 72 Vrms, 1.2 Arms). line-current is located (see Fig. 3). A combination of time-displacement of the two line-current samples and current PWM ripple produces a particularly shaped error of the reconstructed line-current, as uncovered in previous work [25] and now by extensive HIL emulations and experimental setup of which the main results are shown in this section. Table I and Figs. 6 and 7 give an overall view of the complex, ripple-caused error in the reconstructed line-current ( for example). Approximate errors are given for each sector, and especially for the reference voltage vector position at the beginning, middle and end of the sector. In Table I, qualitatively represents the error of the sampled (or calculated) current due to the line-current PWM ripple and sample time-displacement from PWM middle point. It is clear that two abrupt excursions in the current signal can be expected during one period of fundamental output voltage. The ripple-induced error exhibits steps of amplitude twice per period (transitions between sectors 1 2and5 6 for line-current ) and it practically produces an offset section. For the other two line-currents, transition offsets would appear at different positions (1 2and3 4for,and3 4and5 6for current). Fig. 6 shows HIL results that demonstrate the erroneous offset of the two sectors, with the measured and reconstructed motor line-currents for the open-loop control and no-load scenario (where the magnitude of the line-current is comparable to the current ripple level). The experimental results given in Fig. 7 and obtained on the matching real HW model show the same behavior concerning the reconstructed current-waveform error. A. Quantitative Analysis of Stator Current Rate of Change Qualitative analysis given in [25] revealed unusual reconstruction error in line-currents, which was confirmed in realtime emulated and real HW system. A possible way to quantitatively describe the stator current rate of change (during the zero voltage vector, and, and during the first active vector, and ), is to use the machine model in a stationary reference frame, which is native to the vector controlled drives. For induction motor drive, a current rate of change can be calculated using following equation: where are: equivalent leakage inductance of stator windings ; and applied voltage vector values during considered ripple; equivalent stator resistance ; electrical rotor angular frequency; and rotor fluxes; rotor time constant; and actual and previous PWM period. Equation (3) can be used to calculate the rate of change of the stator currents for a given voltage command. During the zero vector, components and are zero. During the first active vector (in lagging side of PWM period), voltage components and are dependent on the sector and on stator windings connection, shown in Table II. (3)

5 634 IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS, VOL. 9, NO. 2, MAY 2013 The obtained change of rates in original phase domain (4) together with the known durations of the voltage vectors (from PWM modulator), can be used to predicted the difference,, between the first and second sample from each line current mean value(seefig.3) Fig. 8. Proposed dc-link current sampling and modified PWM pattern. However, this paper proposes simple method (Section V) which does not need nor include prediction of stator current rate of change. V. IMPROVED SCHEME FOR THE CANCELATION OF OFFSET JITTER-LIKE ERROR IN THE RECONSTRUCTED LINE CURRENTS The motor currents are conventionally reconstructed using two dc-link current samples only, taken in a half (leading or lagging) of single PWM period at different time instants relative to the PWM line-current ripple middle point. The result is an erroneous two sectors offset in the line and currents and an increase of the inherent 6th harmonic jitter in the -currents. Predicted line current sample error [(5)] can be used to align sampled current values to the PWM middle point and to correct the reconstructed line currents. However, the basic idea suggested in this paper represents an improvement of the ingenious method proposed in [15], which used the line-currents measured in both halves of the naturally symmetrical PWM switching period. The method proposed in [15] samples the dc current in the center of the active voltage vectors four times during one PWM period and calculates the two available line-current values by averaging the samples from two matching vector pairs. This approach provides concurrent measurement of all three line-currents (referred to the center of a PWM period). It effectively cancels a current ripple-caused waveform error in the reconstructed line-currents and eliminates the current samples mutual phase-shift. Besides its simplicity, this method is completely insensitive to machine parameter variances as opposed to more complex observer-based methods. However, in [15] the critical cases of a reference voltage vector passing between the six possible active vectors or with a low modulation index are neglected and not considered. The authors in [16] clearly emphasize that during these cases and with the PWM modified scheme used (where PWM signals are not symmetrical as assumed), the simultaneously sampled line-currents cannot be acquired. It clearly indicates there is a need to provide an improved procedure for reliably and more accurately measuring the motor line-currents, while maintaining simplicity. (5) In view of the high PWM switching frequencies (up to 10 khz) and the motor power range (usually between 10 and 100 kw) used for electric vehicle drives, one can conclude that there is no need for the very highest current control-loop sampling rate at the PWM level. This fact allow us to record line-currentinformationonthelaggingsideofonepwmperiodand then on the leading side of the subsequent PWM period and calculate the available line-currents by simple averaging of the corresponding recorded values. In this way, all three estimated line-currents would be referred to the same instant reflecting the average current value in two consecutive PWM periods. The proposed method enables us to improve the PWM pattern control in order to account for critical cases. It represents an extension of the modified PWM pattern explained in Section II-A where under critical conditions, the lagging half-pulse width is shifted to the right and the leading half-pulse width in the subsequent PWM period is shifted to the left in order to create sufficient sampling windows for current measurement. Fig. 8 shows the dc-link current during two consecutive PWM periods and details related to the proposed method. The linecurrents at the time instant, representing average values in two consecutive PWM periods, can be obtained using simple calculation It remains to assign the resultant currents,,and to motor line-currents,,and depending on the actual sector number. The first improvements in the reconstructed line-current waveform can be observed in Figs. 6 and 7 by comparing and signals. VI. EXPERIMENTAL RESULTS An experimental setup which represents a typical electric vehicle drive-train with an induction machine, a two-level voltage (6)

6 ADŽIĆ et al.: DEVELOPMENT OF HIGH-RELIABILITY EV AND HEV IM PROPULSION DRIVE WITH ULL HIL ENVIRONMENT 635 Fig. 9. IM drive used in tests. Fig. 10. Implemented sensorless DFOC speed control scheme. TABLE III PARAMETERS OF EXPERIMENTAL SETUP (USED ALSO FOR HIL) TABLE IV PER-PHASE IM EQUIVALENT CIRCUIT PARAMETERS (ALSO IN HIL MODEL) Fig. 11. Real HW and HIL responses with change in speed reference from 0.05 to p.u., using conventionally reconstructed line-current feedback. (a) Dynamic speed response in real HW system. (b) Dynamic speed response in HIL model. source inverter and a battery, is shown in Fig. 9. Machine, inverter and control parameters are given in Table III. The perphase equivalent circuit parameters for this induction machine are given in Table IV. During all the reported experiments, the motor was lightly loaded (15% rated, 1.1 Nm) by another machine controlled by an industrial frequency-converter in torque mode. Light motor load resulted in a relatively low rms current value and in significant distortion of the reconstructed current waveforms, where usefulness of proposed method is more obvious. For experimental purposes, a direct rotor field-oriented control (DFOC) was implemented [14] and used (Fig. 10). For a speed controller gains set according to module optimum (Table III) and a ramp-change (0.4 s) in speed reference from 0.05 to p.u., Figs. 11 and 12 compare the dynamic speed responses of the sensorless IM drive with both a conventional and the proposed current-reconstruction method. Both, the HIL model (b) and the real HW experimental results (a) are presented, for the sake of comparison. Besides the measured and estimated speeds ( and ), Figs. 11 and 12 show rotating refer- ence frame ( ) current references ( and )and -components of the actual measured motor line-currents ( and ). Since current-reconstruction and current control loop dynamics are an order of magnitude faster than the speed control loop, the reconstruction method does not affect the speed dynamic performance of the drive. The results from Figs. 11 and 12 show an excellent match between hardware and HIL responses, except for slight differences in transition process and steady-state for the speed. Because the exact value of the considered system s moment of inertia is unknown, there are slight differences in transition. In the control system with a conventionally reconstructed line-current feedback, there are significant oscillations of -components of the actual line-currents compared to the case where the proposed current-reconstruction is used. Even more importantly, there is noticeable offset especially in -current component (Fig. 11) which is mostly canceled using the proposed algorithm (Fig. 12).