IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

Sensorless Control of Permanent Magnet Synchronous Motor – A Survey
Li Yongdong* and Zhu Hao**
*Department **Department

of Electrical Engnieering, Tsinghua University. Beijing, China. Email: of Electrical Engnieering, Tsinghua University. Beijing, China. Email: extended Kalman filter (EKF), sliding model observer (SMO), model reference adaptive system (MRAS), etc. Fundamental model presents some limits at low speed, where the rotor-induced voltage is relatively small, and can not be used to precisely identify motor position information. Further researches have found that motor anisotropic properties provide additional information on the field angle or the position of the rotor. This inherent property makes it possible to use transient excitations by injected signals having other frequencies than the fundamental, or transients caused by inverter switching, to reliably identify and track rotor position even when the rotor is at standstill. Relative researches have been mentioned in many articles. This paper gives an overview of the extensive research on sensorless operation of PMSM. Those sensorless techniques are categorized as: open loop methods, close loop observers, and methods based on motor non-ideal property, each will be described in detail in Sections II. II. PMSM SENSORLESS CONTROL SCHEMES A. Open loop methods 1. Direct calculating [1][2] In the steady state, motor voltage equations in the rotor synchronous frame are time invariant, and they contain the speed information. By substitute Ud and Uq with stationary frame axes current and voltage, motor speed can be expressed with measurable variables, and its value can be calculated directly. As is described in [2], the equation for calculation rotor position is θ = arctg ( A / B) Where A = ua − Ria − Ld pia + ωiβ ( Lq − Ld )

Abstract—Vector control has been widely used in control of permanent magnet synchronous motors (PMSM) where the rotor position information is required. The cost of mechanical sensors and the difficulty to incorporate them make it necessary to avoid their uses and to study the mechanical Sensorless control. The idea behind such methods is to extract the motor position and speed information in functions of the terminal quantities from the motor equations. There are many methods which vary in principle and observer structure. In this paper some most promising methods are classified for different application conditions. Keywords—Sensorless control, PMSM, back EMF, EKF, MRAS, adaptive control, sliding mode observer, HF injection.

I. INTRODUCTION Permanent magnet synchronous motors have found wide applications due to their high power density, high efficiency, easy of control, high torque-to-inertia ratio and high reliability. In the past decade, vector control of permanent magnet synchronous motor has emerged as a mature technology. The best performance is achieved with a position sensor (resolvers, encoders or Hall Effect sensors) attached to the rotor shaft. Because of economical, robust and compact drives are needed, the elimination of the position/speed sensors is of high interest. The advantages of sensorless control are reduced hardware complexity and lower cost, reduced size of the drives, elimination of the sensor cable, better noise immunity, increased reliability, and less maintenance requirements. In the past different sensorless control schemes have been proposed, of them one big category is based on motor fundamental equations. A sinusoidal flux density distribution in the air gap is assumed which neglects motor space harmonics and other secondary effects. These methods are either open loop structure, or close loop observers. Open loop methods have a straightforward structure which is easy for calculating. But the estimation is greatly influenced by motor parameters variation and current measurement noise. In order to increase the robustness and to improve the dynamics of the estimation, it is necessary to use the error signals between measured and estimated quantities as a feedback, thus composing a close loop observer like

B = −uβ + Riβ + Ld piβ + ωiα ( Lq − Ld ) The calculation is direct and easy with a very quick dynamic response, and no complicated observer is needed. However, the stator current deviation used in above equations will introduce calculation error due to measurement noise. And any uncertainty of motor parameters will cause trouble to the motor position estimation, which is the biggest problem of this method.

C 2008 IEEE 978-1-4244-1849-7/08/$25.00○

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

2. Method based on stator inductance calculation For Interior Permanent Magnet Synchronous Motors (IPM), stator phase inductance is a function of rotor position θr, and it varies twice the frequency of motor speed [3]. At any time if this inductance is known, motor position can be exactly decided. Based on this property, paper [4] proposed a method to calculate motor rotor position using stator phase current and voltage. Take stator phase A as an example, stator inductance is calculated as:

Many control methods suitable for SMPM cannot be used directly to IPM. In the mathematical model of the IPM, position information is included not only in the flux or EMF term but also in the changing inductance because of its saliency. The model of SMPM is a special symmetrical case of IPM, which is relatively easy for mathematical procession. In order to apply the method suitable for SMPM to a wide class of motors, i.e. the IPM, in [7] a novel IPM model is suggested with an extended EMF.

Lsa = [ va − ea − Rs ia ] / ( dia / dt )
Where ea = K ωe = K dθ = K ⎡θ r (t2 ) − θ r (t1 ) ⎤ ⎢ ⎥ dt Δt ⎣ ⎦ By comparing motor stator inductance of the calculated value with the measured one in advance, rotor position is acquired.Yet in dynamic period, it is not so easy to calculate exact back EMF value. The calculated inductance will inevitably have some errors, which lead to an inaccurate estimation. Also, this method is an open loop method, the correctness of the estimation can not be guaranteed. If stator current is so high that stator flux saturation occurs, the rotor position estimation based on look-up table will lead to a wrong result. 3. Back EMF integration [5][6] In the steady state, stator and rotor flux vectors rotate synchronously. The angle difference between these two space vectors is load torque angle. If stator flux vector is known, then rotor flux angle can be calculated which indicates rotor position. For SMPM motor, its mathematic equations are symmetric in α – β coordinate. By simple integration the stator flux can be easily obtained
r r r ψ ds = ∫ (vds − rs ids )dt r r r ψ qs = ∫ (vqs − rs iqs )dt

⎡vd ⎤ ⎡ R + pLd − ω re Lq ⎤ ⎡id ⎤ ⎢ ⎥ ⎢v ⎥ = ⎢ ω L R + pl q ⎥ ⎦ ⎣iq ⎦ ⎣ q ⎦ ⎣ re q 0 ⎡ ⎤ +⎢ ⎥ − − + ( L L )( ω i pi ) ω K q re d q re E ⎦ ⎣ d By rewriting motor voltage equations into a symmetric form, the author defined an extended EMF E ex = ( Ld − Lq )(ω re i d − pi q ) + ω re K E In this way, the voltage equation of IPM is fully identical to that of SMPM without any approximation. Paper [10] exemplified an application based on this extended EMF method, where the adaptive observer is constructed on α − β coordinate for IPM sensorless drives. The problem is that, this so-called extended EMF is influenced by stator current id and iq, which vary during motor transient state. This will cause troubles to the speed estimation. In the low speed range, the signal-tonoise ratio of the extended EMF is relatively small and the speed estimation result is still not so good.
B. Close loop methods 1. Extended Kalman filter The Kalman Filter is a stochastic state observation based on least–square variance estimation. Extended Kalman Filter is an extension application of Kalman filter in nonlinear system. EKF based observers are ideally suited to estimate rotor speed and position. A great advantage is that it is less influenced by measurement noise, and parameters inaccuracy is not as critical as in conventional back EMF estimation based methods. A standard EKF observer contains three steps (1) Prediction step xek k −1 = xek −1 k −1 + [ F ( xek −1 k −1 ) xek −1 k −1 + B(u k −1 )]Tc

Where the angle of stator flux vector is
r r θψ = arctg (ψ qs /ψ ds )

Then the rotor flux vector can be calculated as ψ r = ψ s − R s is This method is sufficiently robust and accurate at higher stator frequency. The decisive parameter in the integration is stator resistance Rs which increases with temperature. A parameter error in Rs affects the value of Rs·Is, this signal dominates the integrator input when the magnitude of Us reduces at low speed. To summarize, two basic deficiencies let this model degrade as the speed reduces: the pure integration problem and the sensitivity of the model to stator resistance mismatch. Though not proper for high accuracy application, it gives a way out on how to estimate motor speed by using rotor flux space vector, which is closely related to the back-EMF. 4. Extended EMF [7][8][9][10]

Pk |k −1 = Pk −1|k −1 + ( Fk −1 Pk −1|k −1 + Pk −1|k −1 Fk'−1 )Tc + Qd (2) Innovation step
xek |k = xek |k −1 + K k yk − h xek |k −1 Pk |k −1 = Pk −1|k −1 + Fk −1Pk −1|k −1 + Pk −1|k −1Fk′−1 Tc + Qd






(3) Kalman gain
′ HPk |k −1 H k ′ +R K k = Pk |k −1 H k




Choosing different state variables leads to different EKF observers: x = iα iβ ω θ ' [11]



IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

x = id



ω θ ]' [12]

A crucial step in Kalman filter design is the choice of the elements of the covariance matrices Q and R, as they will affect the performance, convergence and stability. (1) R is related to the measurement noise. Increasing the value of R will assume that the current measurements are more affected by noise and thus less reliable. Consequently, the filter gain K will decrease, yielding poorer transient response. (2) Q is related to parameter uncertainties. Large values in Q presume high model noise and/or parameter uncertainties. An increase of Q will result in faster filter dynamics but poorer steady-state performance. The filter convergence and stability is much more sensitive to the terms of the covariance matrix Q than R. Tuning of covariance matrices of the model and measurement errors is rough and requires skilled operators. This matrices parameter tuning and the potential filter instability under specific operating conditions resulting from the non-optimal EKF tuning are the biggest disadvantage of EKF. Another problem is that, if implemented in the α, β-reference frame, the Kalman filter may converge to the wrong solution (-ω, θ+π), and special measures should be taken into consideration [13]. For IPM motor, another big difficulty in designing Extended Kalman Filter is the complexity of the motor dynamic voltage model, which is more complicated than that of SMPM. This excludes from the very beginning the use of a dynamic model with parameters in the stator reference frame for IPM, as is used for SMPM. Kalman filter based algorithms are computationally intensive and time consuming. Heavy computation load from variance matrix makes EKF hard to apply in electrical drive industry. Paper [14] suggests a reducedorder Linear Kalman filter for PMSM speed estimation. The most advantage of this method is that the estimation error covariance matrix P is developed into a time invariant Riccati difference P by using orthogonal output y = [cos θ sin θ ]' . This method has greatly reduced the computation time. Yet due to its simplicity steps for calculation, the dynamic response is not as good as in the steady state. 2. MRAS Many articles have used MRAS approach to estimate rotor position. It makes use of the redundancy of two machine models of different structures that estimate the same state variable (rotor speed) of different set of input variables. The estimator that does not involve the quantity to be estimated is chosen as the reference model, and the other estimator may be regarded as the adjustable model. The error between the estimated quantities obtained by the two models is proportional to the angular displacement between the two estimated flux vectors. A PI adaptive mechanism is used to give the estimated

speed. As the error signal gets minimized by the PI, the tuning signal ω approaches the actual speed ω of the motor. Based on MRAS principle, paper [15] uses voltage model and current model to calculate stator flux, the error between the two results is used to estimate rotor speed. Though simple for application, the estimated result depends greatly on motor parameter accuracy. In order to overcome this problem, in [16][17] a combined method is suggested (Fig.1). The idea comes from HF injection method. In the proposed method a calibration signal ε containing estimated angle error is used for the calculation of stator flux using the voltage model. The author claims that the proposed combination of the two methods results in an observer having good steady-state accuracy and excellent dynamic properties over a wide speed range.


ud , uq
id , iq

ˆ d ,u , ψ ˆ q ,u ψ

ˆm ω

id , iq

ˆ d ,i , ψ ˆ q ,i ψ
Fig.1 MRAS based on stator flux estimation

Paper [18] presents a MRAS method based on stator current. Using PM motor itself as a reference model, another adaptive mechanism is established (Fig.2). This method is easy for application. The stability of the system is guaranteed by the Popov super stability theory. It is somewhat robust to parameter inaccuracy. Yet in the calculation, only a PI integration is used to calculate the estimated speed from current difference between the two models. The convergence speed and the steady estimation accuracy cannot be properly assured.



Coordinate Transform

id iq

Coordinate Transform


Adaptive Model

ˆd i

ˆq i

ˆ θ

ˆ ω

ˆq − iqi ˆd − ˆ = ∫ k1[id i ω


ˆq )]dτ (iq − i

ˆq − iqiˆd − ψ r (iq − i ˆq )] − ω ˆ (0) + k2 [id i Ld

Fig.2 MRAS based on stator current

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

3. Adaptive control In [19] a simple adaptive control method is proposed. For known magnitude of the permanent magnet ψ M ,ref = ψ M , the observer (Fig.3) can be seen as a rotor field angle observer.
ˆ off u

ˆ M ,ref ψ

ˆr ψ
u s − Rs is

Using the reconstructed back EMF, the rotor position and motor speed is estimated Z eqα ˆ sα ˆ = − tan −1 ( e θ ) = − tan −1 ( ) r ˆ e sβ Z eqβ Paper [22] proposed a new SMO method by properly choosing gain coefficient, aiming at full speed range operation especially for improving low speed range performance. But in the low speed range only 20% of rated speed was tested. The nearly zero speed range is still unverified. In [23] another sliding mode control method is proposed. In this method, not only the speed, but also stator resistor is online estimated using SMO.

e − jθ

Ls i s
Fig.3 Signal flow graph for adaptive control

ˆM ψ

Its advantage is that this method concerns compensating voltage offset (uoff), and a calibration loop (with rotor flux) is introduced, which makes the estimation somewhat more accurate. Yet the proposed method is based on motor fundamental equations, so it’s dependent on parameter accuracy. Its performance cannot be guaranteed for wide speed range operation, especially at low speed region. 4. Sliding mode observer A number of estimation techniques have been developed to achieve speed and position sensorless PMSM drives. Most of them suffer from variation of motor parameters such as the stator resistance, stator inductance and torque constant. Also, it is known that conventional linear estimators are not adaptive to variations of the operating point in a nonlinear system. Sliding mode observer illustrates the fancy idea to construct a speed observer based on motor equations but not merely confined to the basic structure. In SMO, the current error vector ΔIs is used to define the sliding hyperplane of the sliding mode compensator [20][21]. The magnitude of the estimation error ΔIs is forced to the vicinity of zero by a high-frequency nonlinear switching controller. The switched complex signal is directly used to exert a compensating influence on the machine model. The robustness of the sliding mode approach ensures zero error of the estimated stator current. SMO and the estimated back EMF: diα l R 1 = − iα + eα − 1 sign(iα ) dt L L L di β l R 1 = − i β + e β − 1 sign(i β ) L L L dt

SMO doesn’t need coordinate transformation, thus it won’t introduce error due to position estimation transient deviation. By comparing the real and the estimated stator current, it is simple and to ensure the accuracy speed estimation comparing to other methods where coordination transformation is involved. The sufficient high switching gain satisfies the necessary condition for SMO convergence in the high speed range. This gain of the error compensator increases when a sliding mode controller is employed for observer tuning. The problem is, good steady state result versus fast convergence rate is a conflict. In order to get accurate speed estimation result, the convergence rate cannot be properly assured, especially for wide speed range operation. In principal the sliding model observer is always switching due to the existence of sliding surface. This will introduce error to the estimated value in the steady state. Finally, in low speed range there will be big error in the back EMF estimation, because the back EMF is a normally small value at low speed.
C. Methods based on motor non-ideal property The common problem of motor fundamental equations based methods is that the performance of the rotor position estimation is critically dependent on the magnitude of back-EMF voltage, which is proportional to the rotor speed. In low speed region, where the magnitude of back-EMF voltage is so small, the performance is significantly degraded. As a consequence, the PM motors have to be first started in opened-loop with an arbitrary switching and brought up to a speed high enough to permit one to detect voltage components. This approach is not reliable in many applications because it does not necessarily lead to smooth starting in the right direction. Different methods have been proposed for sensorless control of PMSM in the low speed range: 1. HF injection[24][25][26][27][28] HF injection method is based on the magnetic saliency property. By injected voltage [25][26] or current [27] signals having other frequencies than the fundamental, the response signals can be used to detect spatial information. The injected signals may be

ˆα − iα ) ⋅ ω c ˆα = ksign(i e s + ωc

ˆβ − iβ ) ⋅ ω c ˆβ = ksign(i e s + ωc

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

periodic, creating either a high frequency revolving field, or an alternating field in a specific, predetermined spatial direction. Such signals can be referred to as carriers, being periodic at the carrier frequency with respect to space, or time. The carrier signals, mostly created by additional components of the stator voltages, get modulated by the actual orientations in space of the machine anisotropies. The carrier frequency components are subsequently extracted from the machine current waveforms. They are demodulated and processed to retrieve the desired information using position angle error signal Vinj Ldiff ~ ~ ~ f (θ r ) ≈ θ r = K errθ r 2ω h Lrdh Lrqh The results in [25] and [26] are similar. The difference is that in [26] the HF signal is injected in the d-axis, which avoids introducing torque ripple as that in [25]. Another advantage of [26] is that injecting signal in d-axis enables the utilization of flux saturation characteristic, thus can be used for those motors of whose saliency is not very obvious. But saturation might cause trouble for heavy load operation, where stator over current might occur. Paper [28] injects HF voltage signal in motor stationary frame

2 2 2 2 ~ ~ 3p ψ m Ic ~ 3p ψ m Ic ˆ cq = cos θ sin θ e sin(ω c t ) ≈ θ sin(ω c t ) Jω c Jω c

LF injection method is originally designed for the speed estimation of IPM motor at low speed. The saliency of the motor is a distortion for this method. Thus the basic idea is quite different from the high frequency methods, where the saliency is a necessary condition for the estimator. Typical injection frequency is between a few Hz to a few hundred Hz according to the ratio of p2/J. The position error signal is dependent on motor mechanical parameter J. If J is too high this estimator method cannot be applied. In most cases, no accurate value of J can be acquired; this will cause trouble to the observer. The slow dynamic response is another drawback of this estimator. 3. INFORM Rotor position estimation of permanent magnet synchronous machines (PMSM) at low speeds and standstill can be realized by utilizing position dependent inductance values of the stator windings which may be caused by saturation and/or reluctance effects. One of the methods is the so-called INFORM (“Indirect Flux detection by On-line Reactance Measurement”). This method is first introduced in [30]. The basic idea of INFORM is to measure the current response, which is evoked by voltage space vectors applied in different directions, and using this current to identify inductance variation. In detail, a sequence of voltage space vectors u s is applied to the motor via the inverter and the current reaction dis / dτ is measured [31]. The author defines

⎡Vαi ⎤ ˆ ⎡− sin(ωi t ) ⎤ Vi = ⎢ ⎥ = V i⎢ ⎥ ⎣ cos(ωi t ) ⎦ ⎣Vβi ⎦ The benefit of this method is that it avoids coordinate transformation using the estimated angle, thus less influenced by rotor position estimation error. HF injection methods are insensitive to parameter inaccuracy. Coordination transformation (projection stator current to the estimated q-axis) won’t cause error to position estimation. Another advantage is that, for a PM motor whose saliency is not very prominent, flux saturation effect can be utilized by using d-axis signal injection like that does in [26]. In the control procedure, certain amount of voltage needs to be injected into motor in order to get adequate stator HF current. Thus the available voltage for motor normal control is decreased. Also, digital filter is used for signal processing, this will cause phase shifting and magnitude distortion. Due to the application of filter, the dynamic performance is poor. If the saliency is not sinusoidally distributed and furthermore its shape and phase shift with respect to the rotor position will be load dependent. This will produce harmonics in the position signals and in turn causes angle estimation error.
2. LF injection LF injection [29] is another prospective method for motor speed estimation in the low speed range. In the proposed method, low frequency current is injected into the reference d-axis. The injected current will cause speed transient variation which indicates a speed estimation error. This error is then used to form a speed calibration value, and finally to estimate rotor position:

IN FO RM -ax is

us di s / dτ

Fig.4 Complex function yINFORM

The parameter l INFORM is the so called p.u. complex INFORM reactance containing the desired dependence on the rotor position,. . l INFORM is an 180° periodic function periodically changing with motor position. Its calculation is simple

l INFORM = l INFORM (2γ m − 2γ u )

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

and no need to use motor equations. Thus this method is parameter insensitive. Yet the calculation is based on the assumption that motor flux is sinusoidally distributed, so a flux distortion will cause speed estimation error, and the influence needs to be further checked. Another disadvantage is that in the steady state the applied testing voltages will cause current ripple. This is the biggest disadvantage of this method. D. Initial position detection In order to ensure that PM motor starts smoothly, rotor initial angle needs to be known first. Otherwise if the PM motor starts in an arbitrary position which is unknown, some sensorless control schemes may encounter starting failure. There are several motor initial position detection methods, and they have been proved effective in application. The basic principle of these methods is to use certain excitation to extract rotor position information. 1. Using pulse voltage Based on the phenomenon that inductance varies with the rotor magnet position, paper [32] suggests a simple method to estimate motor initial position. In the proposed method, certain amount DC voltage rectangular pulses are applied to motor stator windings at standstill. The peak values of the response currents are measured. And their relation wish rotor position is modeled as I u = I 0 + ΔI u = I 0 + ΔI 0 cos(2θ )

position estimated by this method is +/- 0.9375 elec. deg.





Fig.5 test voltage and current sampling

2π ) 3 2π I w = I 0 + ΔI w = I 0 + ΔI 0 cos(2θ + ) 3 Where I0=1/3(Iu+Iv+Iw) is the DC current component and ΔI0 is the amplitude of the fluctuated component. From above trigonometric identities, one can extract rotor position information as I v = I 0 + ΔI v = I 0 + ΔI 0 cos(2θ −
3 ( ΔI u − ΔI v ) 2 ΔI u − ΔI v − ΔI w The exact value could be found by calculating the inverse tangent and dividing the remaining angle by two. For small angles, an approximation of tg(2θ)=2θ, thus rotor position can be calculated directly. tg ( 2θ ) =
2. Using space vector test voltages To SMPM, if the voltage vector approaches the Npole, the d-axis current gradually increases due to the magnetic saturation of the stator core. Therefore, the initial rotor position can be estimated by detecting the maximum d-axis current response for the optimum voltage vector [33]. In the proposed method, voltage vectors in different direction are applied to motor each, and the response currents are measured. The applied voltage with the highest current indicates rotor position. Theoretically the maximum accuracy of the rotor

Fig.6 Voltage vectors for rotor position detection

3. Using HF injection test signal Paper [34] detects motor initial position using a HF current injecting signal. When motor is at standstill, the voltage equations can be written as , An injected HF current Ihcosωht results in a voltage response as
e e e vd = Rs idh + ( Ld cos 2 Δθ r + Lq sin 2 Δθ r ) ⋅ pi dh

r r r r r r vq = Rsiq + pLqiq vd = Rsid + pLd id

A demodulating technique can be utilized to extract the rotor position error signal from the q-axis reference voltage. Multiplication of the q-axis voltage by sinθh and low pass filtering gives 1 e* LPF (vq ⋅ sin θ h ) = ωh I h ( Lq − Ld ) sin 2Δθ r 4 Where θ h = ∫ ωh dt

1 e* e vq = − ( Ld − Lq ) sin 2Δθ r ⋅ pi dh 2

Magnet polarity can be detected by use of saturation effect of the stator iron. To describe the effect, first order Taylor approximation for the inductance is used as follows
e Ld ≅ Ld 0 ∓ Ld 1idh

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

Here “-” means a north pole is aligned with the coil and vice versa. Then the d-axis voltage can be deduced as
e e e e e vd = Rsidh + Ld 0 pidh ∓ Ld 1idh pidh

Therefore, the polarity of the magnet can be detected from the polarity of the second harmonic component of the d-axis voltage, given as 1 e* LPF (v d ⋅ sin 2θ h ) = ± Ld 1ω h I h2 4 The whole process is illustrated in Fig.7.

e vq


ˆ θ r

sin θ h
e vd

sin 2θ h
Fig.7 Initial position detection – HF injection

Injection HF voltage signals can get similar result [35][36][37]. If a HF voltage vi e jωi t is superimposed to the motor [35], the response current is
s iqdi = iip e j (θ i (t ) −π / 2) + iin e j ( 2θ r −θ i (t ) +π / 2)

range, the estimation result is not as good as that in high speed range. 3. Methods based on motor non-ideal properties have avoided directly calculating motor back EMF. Special excitation signal is used to induce extra response which carries rotor position information. These methods have proved to be effective in motor low speed range even down to zero speed. Yet in the steady state, still additional excitation signal is needed. Yet the injected signal will cause steady state transience, and will decrease the applicable voltage of the inverter for motor normal operation. Besides, the consequence of complicated signal processing is low poor estimation result during motor dynamic response, which is the biggest problem of these methods. In order to acquire a good PMSM speed sensorless control performance over wide speed range, the trend is to combine different method, certain switch schemes is need when motor is running at different speed. Here needs to point out that, all the sensorless control schemes are based on the assumption that rotor flux is sinusoidally distributed. A distortion of the flux will cause trouble to speed estimation inevitably. So researches on PMSM sensorless control still need to be furthered. REFERENCES
Naidu M, Bose B K. “Rotor position estimation scheme of a permanent magnet synchronous machine for high performance variable speed drive,” IEEE IAS 1992, vol.1, pp. 48-53. [2] Hoque M, A Rahman M. A. “Speed and position sensorless permanent magnet synchronous motor drives,” Canadian Conference on Electrical and Computer Engineering, 1994, vol.2, pp. 689-692. [3] A E Fitzgerald, C K Kingsley, S D Umans. Electric Machinery, Fourth Ed. New York McGraw-Hill, 1983, pp. 361-371. [4] A B Kulkarni, M Ehsani. “A novel position sensor elimination technique for the interior permanent-magnet synchronous motor drive,” IEEE Transactions on Industry Applications, 1992, 28(1), pp.144-150 [5] Wu R, Slemon G R. “A permanent magnet motor drive without a shaft sensor,” IEEE Industry Applications Society Annual Meeting, 1990, vol.1, pp. 553-558. [6] B K Bose. Modern Power Electronics and AC Drives. Pearson Education North Asia Limited and China Machine Press, 2002. [7] Zhiqian Chen, Tomita,M, Ichikawa S, Doki S, Okuma S. “Sensorless control of interior permanent magnet synchronous motor by estimation of an extended electromotive force,” Conference Record of the 2000 IEEE Industry Applications Conference, Oct. 2000, vol.3, pp. 1814-1819 [8] Morimoto S, Kawamoto K, Sanada M, Takeda Y. “Sensorless control strategy for salient-pole PMSM based on extended EMF in rotating reference frame” IEEE Transactions on Industry Applications, Volume 38, Issue 4, July/Aug. 2002, pp. 10541061. [9] Zhiqian Chen, Tomita M, Doki S, Okuma S. “An Extended Electromotive Force Model for Sensorless Control of Interior Permanent-Magnet Synchronous Motors,” IEEE Transactions on Industrial Electronics, Volume 50, Issue 2, April 2003, pp. 288-29. [10] Hasegawa M, Hatta H, Matsui K. “Adaptive Flux Observer on Stator Frame and Its Design Based on γ−Positive Real Problem for Sensorless IPM Drives,” 32nd Annual Conference of IEEE Industrial Electronics Society, 2005, 6-10 Nov, 2005, pp. 6. [1]

L ⎤ vi ⎡ ΔL ⎤ vi Where iip = ⎡ iin = ⎢ 2 2 2⎥ 2 ⎢ ⎦ ωi ⎣ L − ΔL ⎦ ωi ⎣ L − ΔL ⎥ The current vector induced by the carrier frequency voltage vector can be divided into two components: the first is a positive sequence component that rotates in the same direction as the injected voltage; the second is a negative sequence component that rotates in the opposite direction as the injected voltage. It can be clearly seen that the second part contains rotor position information. This information can be picked out using digital signal processing.
III. CONCLUSION This paper has introduced different speed and rotor position estimation schemes of PMSM motor. Each method has its advantages and disadvantages. In summary: 1. Open loop methods based on motor fundamental model are easy for application, and have quick dynamic response. But these methods are vulnerable to parameters uncertainty and stator current measurement noise. So they are directly used not in application. 2. Close loop observers are superior to open loop methods. As feed back calibration loop is used, the convergence property is guaranteed. Since most of these methods are based on motor fundamental equations, and motor back EMF which is relatively small in low speed

IEEE Vehicle Power and Propulsion Conference (VPPC), September 3-5, 2008, Harbin, China

[11] Bolognani S, Oboe R, Zigliotto M. “Sensorless Full-Digital PMSM Drive With EKF Estimation of Speed and Rotor Position,” IEEE Transactions on Industrial Electronics, Volume 46, Issue 1, Feb. 1999, pp.184-191. [12] Bolognani S, Tubiana L, Zigliotto M. “EKF-based sensorless IPM synchronous motor drive for flux-weakening applications,” IEEE Transactions on Industry Applications, Volume 39, Issue 3, May/June 2003, pp.768-775. [13] A. Bado, S. Bolognani and M. Zigliotto. “Effective estimation of speed and rotor position of a PM synchronous motor drive by a Kalman filtering technique,” PESC'92, June/July 1992, vol. 2, pp. 951–957. [14] Huang M C, Moses A J, Anayi F, Yao X G. “Linear Kalman filter (LKF) sensorless control for permanent magnet synchronous motor based on orthogonal output linear model,” International Symposium on Power Electronics, Electrical Drives, Automation and Motion, 2006. May, 23rd - 26th, 2006, pps.1381-1386. [15] G D Andreescu. “Position and speed sensorless control of PMSM drives based on adaptive observer,” EPE’ Proc. 1999, CD-ROM. [16] Piippo A, Luomi J. “Adaptive Observer Combined With HF Signal Injection for Sensorless Control of PMSM Drives,” IEEE International Conference on Electric Machines and Drives, 1518 May 2005, pp. 674-681. [17] Piippo A, Hinkkanen M, Luomi J. “Analysis of an Adaptive Observer for Sensorless Control of PMSM Drives,” 32nd Annual Conference of Industrial Electronics Society, IECON 2005. 6-10 Nov. 2005, pp.6. [18] Yan Liang, Yongdong Li. “Sensorless control of PM synchronous motors based on MRAS method and initial position estimation,” Sixth International Conference on Electrical Machines and Systems, Vol.1, 9-11 Nov. 2003, pp. 96-99. [19] Rasmussen H, Vadstrup R, Borsting H. “Adaptive observer for speed sensorless PM motor control,” Conference Record of the 38th IAS Annual Meeting. Vol.1, 12-16 Oct. 2003, pp. 599-603. [20] Changsheng Li; Elbuluk M. “A sliding mode observer for sensorless control of permanent magnet synchronous motors,” Conference Record of the 36th IAS Annual Meeting. Sept./Oct. 2001, Vol.2, pp. 1273-1278. [21] Salvatore L, Cupertino F, Cascella G L. “A new approach to sensorless vector control of SMPM with adaptive sliding-mode observer,” Proceedings of the 2002 IEEE International Symposium on Industrial Electronics, SIE 2002,Vol.2, 8-11 July 2002, pp. 489-494. [22] Song Chi, Longya Xu. “Position Sensorless Control of PMSM Based on a Novel Sliding Mode Observer over Wide Speed Range,” Power Electronics and Motion Control Conference, 2006. IPEMC '06, Vol.3, Aug. 2006, pp. 1-7. [23] Yoon-Seok Han, Jung-Soo Choi, Young-Seok Kim. “Sensorless PMSM Drive with a Sliding Mode Control Based Adaptive Speed and Stator Resistance Estimator,” IEEE Transactions on Magnetics, Vol.36, Issue 5, Part 1, Sept. 2000, pp. 3588-3591. [24] Ji-Hoon Jang, Seung-Ki Sul, Jung-Ik Ha, Ide K, Sawamura, M. “Sensorless Drive of Surface-Mounted Permanent-Magnet Motor by High-Frequency Signal Injection Based on Magnetic Saliency,” IEEE Transactions on Industry Applications, Vol.39, Issue 4, July/Aug. 2003, pp. 1031-1039. [25] Corley M J, Lorenz R D. “Rotor position and velocity estimation for a permanent magnet synchronous machine at standstill and high speeds,” Industry Conference Record of the 21st IAS Annual Meeting, IAS '96, Vol.1, 6-10 Oct. 1996, pp. 36-41. [26] Ji-Hoon Jang, Jung-Ik Ha, Ohto M, Ide K. Seung-Ki Sul; “Analysis of permanent magnet machine for sensorless control based on high frequency signal injection,” IEEE Transactions












on Industry Applications, Vol.40, Issue 6, Nov./Dec. 2004, pp. 1595-1604. Ribeiro L A, Degner M W, Briz F. “Comparison of carrier signal voltage and current injection for the estimation of flux angle or rotor position,” Industry Applications Conference, IAS 1998, Vol.1, pp: 452-459. Saltiveri D, Asher G, Sumner M, Arias A, Romeral L. “Application of the matrix converter for the sensorless position control of permanent magnet AC machines using high frequency injection,” European Conference on Power Electronics and Applications, 11-14 Sept. 2005, pp. 10. Eskola M, Tuusa H. “Sensorless control of salient pole PMSM using a low-frequency signal injection,” European Conference on Power Electronics and Applications, 11-14 Sept. 2005, pp. 10. Schroedl M. “Operation of the permanent magnet synchronous machine without a mechanical sensor,” 4th IEE Conference On Power Electronics and Variable-Speed Drives, 1990, pp.51-56. Schrodl M, Lambeck M. “Statistic properties of the INFORM method for highly dynamic sensorless control of PM motors down to standstill,” The 29th Annual Conference of the IEEE Industrial Electronics Society, IECON '03. Vol.2, 2-6 Nov. 2003, pp. 1479-1486. Boussak M. “Implementation and experimental investigation of sensorless speed control with initial rotor position estimation for interior permanent magnet synchronous motor drive,” IEEE Transactions on Power Electronics, Vol.20, Issue 6, Nov. 2005 pp. 1413-1422. Tanaka K, Yuzawa T, Moriyama R, Miki I. “Initial rotor position estimation for surface permanent magnet synchronous motor.” Proceedings of the Fifth International Conference on Electrical Machines and Systems, ICEMS 2001. Vol.2, 18-20 Aug. 2001, pp. 824-827. Dae-Woong Chung, Jun-Koo Kang, Seung-Ki Sul. “Initial rotor position detection of PMSM at standstill without rotational transducer,” International Conference of Electric Machines and Drives, IEMD '99 9-12 May, 1999, pp.785-787. Wang Limei, Guo Qinging. “Principles and implementation of permanent magnet synchronous motor zero-speed ensorless control.” 7th International Workshop on Advanced Motion Control, 3-5 July, 2002, pp. 247-250. Budden A S, Holliday D, Mellor P H. “Zero Speed Sensorless Position Detection for Permanent Magnet Synchronous Machines,” 36th IEEE Power Electronics Specialists Conference, PESC '05, 2005, pp. 2436-2441. Jeong Y, Lorenz R D, Jahns T M, Sul S. “Initial rotor position estimation of an interior permanent magnet synchronous machine using carrier-frequency injection methods,” IEEE International Electric Machines and Drives Conference, IEMDC'03. Vol.2, 1-4 June 2003, pp. 1218-1223.

Sign up to vote on this title
UsefulNot useful