IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 39, NO. 3, MAY/JUNE 2003 835 Position Estimation in Salient PM Synchronous Motors Based on PWM Excitation Transients Vladan Petrovic ´ , Member, IEEE, Aleksandar M. Stankovic ´ , Senior Member, IEEE, and Vladimir Blaˇ sko, Senior Member, IEEE Abstract—This paper presents a position and speed estimation algorithm based on magnetic saliency of permanent-magnet syn- chronous motors. The proposed method uses inherent high-fre- quency content of motor pulsewidth-modulation (PWM) excita- tion to measure position-dependent inductance parameters, which are then processed using a simple nonlinear observer to produce the position and speed estimates. To ensure persistent excitation at all operating voltages (speeds), a novel PWM algorithm with full voltage output is developed and presented. The efficiency of the modified excitation was evaluated and compared to the efficiency of the standard position-sensorless motor drive excitation. Experi- mental results are presented as well, showing good algorithm per- formance in a wide speed range, including zero, and for various load torques. Index Terms—Motor drives, permanent-magnet (PM) motors, position estimation, position-sensorless control. I. INTRODUCTION P ERMANENT-MAGNET synchronous motors (PMSMs) became increasingly popular in high-performance vari- able-frequency drives. In many applications PMSMs are the preferred choice due to their favorable characteristics: high effi- ciency, compactness, high torque-to-inertia ratio, rapid dynamic response, and simple modeling and control. To achieve proper field orientation in motion control of PMSMs, it is necessary to obtain the actual position of the rotor magnets. Although a position sensor mounted on the motor shaft (encoder, resolver, Hall-effect sensor, etc.) is typically used for this purpose, there is a significant interest in removing those sensors since they are often complex and rather fragile. Their removal thus results in reduced overall system cost (both in terms of parts and in maintenance), and in improved reliability. The industrial interest in position-sensorless operation of ac motor drives has triggered intensive research in this area in the past several years. Broadly speaking, there are two approaches Paper IPCSD 02–034, presented at the 2001 Industry Applications Society Annual Meeting, Chicago, IL, September 30–October 5, and approved for pub- lication in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Indus- trial Drives Committee of the IEEE Industry Applications Society. Manuscript submitted for review October 15, 2001 and released for publication February 24, 2003. This work was supported in part by the National Science Foundation under Grant ECS-9502636 and Grant ECS-9820977, and by the Office of Naval Research under Grant N14-95-1-0723. V. Petrovic ´ is with Aware Inc., Bedford, MA 01730 USA (e-mail: vpetrovi@aware.com). A. M. Stankovic ´ is with the Department of Electrical and Computer Engineering, Northeastern University, Boston, MA 02115 USA (e-mail: astankov@ece.neu.edu). V. Blaˇ sko is with Otis Elevator Company, Farmington, CT 06032 USA (e-mail: Vladimir.Blasko@otis.com). Digital Object Identifier 10.1109/TIA.2003.811776 to the rotor position estimation reported in the literature. The first approach concentrates on estimation of the motor back elec- tromotive force (EMF) and subsequent extraction of position in- formation from this signal [1]–[4]. The common characteristic of the back-EMF-based algorithms is that their performance de- grades at low speeds, since the back-EMF term vanishes close to standstill. To avoid this singularity, the second approach relies on position dependence of motor inductances due to magnetic saliency. The methods in this group usually involve injection of an auxiliary signal to probe the motor electrical subsystem, and use the response to such excitation to estimate the posi- tion [5]–[10]. Corley and Lorenz [5] propose the method sim- ilar in operation to a resolver and resolver-to-digital converter (RTDC), where the motor actually acts as the electromagnetic resolver. Sinusoidal injection is applied with the power con- verter and the current response is processed with an RTDC chip in hardware. In [6], a similar method with a lower injection signal frequency is used, but both the generation and processing of the test signals are performed without the additional hard- ware. The methods in [8] and [9] use motor saliency only for the startup, and continue with back-EMF-based algorithms at higher speeds. The first of those two methods uses pulse and the second sinusoidal signal injection. Another method that com- bines the two position sensing principles is presented in [10]. In this work, the authors use injected waveforms natural for in- verters (the so-called INFORM method) for position estimation at low speeds, and switch to the back-EMF method at higher speeds. In addition, a Kalman filter is used for the estimation of the mechanical variables and load torque. The majority of techniques operational at low speeds involve auxiliary signal injection. However, a different approach to system excitation for parameter estimation was proposed by Ogasawara and Akagi [11], [12]. The authors suggest the use of the high-frequency content of pulsewidth-modulation (PWM) voltage signal, instead of an auxiliary injection signal, for the system identification. A similar approach was adopted in [13] where the authors use PWM transients in a back-EMF-based position estimation with saliency-based startup. In our work, we start with a model discretization similar to [11] in order to exploit the current transients within a PWM period. Then, we use the current and voltage data and the PMSM electrical subsystem model to construct a novel least-squares problem intended for the estimation of the position-dependent motor parameters. Computationally inten- sive least-squares problem solution is avoided with a careful problem reformulation which reduces the online parameter estimation procedure to a simple matrix-vector multiplication. 0093-9994/03$17.00 © 2003 IEEE