TLS EXIN based neural sensorless control of a high dynamic PMSM
A. Accetta
a,
n
, M. Cirrincione
b
, M. Pucci
a
a
ISSIA (Institute of Intelligent Systems for the Automation)—CNR (National Research Council), via Dante 6, Palermo, Italy
b
Universite
´
de Technologie de Belfort-Montbe
´
liard, 90010 Belfort cedex, France
article info
Article history:
Received 31 May 2011
Accepted 18 March 2012
Available online 10 April 2012
Keywords:
PMSM
Sensorless control
Neural networks
TLS EXIN
abstract
Sensorless vector control of the Permanent Magnet Synchronous Motors (PMSMs) has been a very
challenging subject for many years. In general, the absence of the encoder in the drive permits to obtain
high dynamical performance by exploiting increased reliability and also reduced cost. Among the
different methodologies proposed in the literature, a model-based approach has been proposed here. In
particular, the space-vector equations of the PMSM have been re-elaborated in a matrix form to permit
the use of a Least Squares technique for the estimation of the speed of the PMSM. The problem has been
then faced-up with the so-called TLS EXIN neuron, which is the only linear neural network able to
solve the TLS problem on-line in a recursive form. Experimental tests have been performed on an
experimental test set-up based on a fractional horsepower permanent magnet machine.
& 2012 Elsevier Ltd. All rights reserved.
1. Introduction
Sensorless control, also called encoderless control, of both
induction motors (IMs) and Permanent Magnet Synchronous
Motors (PMSMs) has been a challenging subject for the last
decade (Holtz, 2002, 2006; Vas, 2003). It is still an up-to-date
subject, as witnessed by the amount of recent scientific publica-
tions. The absence of the encoder, or speed sensor, permits an
increased reliability, a reduced cost and a reduced axial dimen-
sion of the drive. With specific regard to PMSMs, two main
categories of encoderless control exist: model based techniques
and saliency based ones.
1.1. Saliency based techniques
Saliency techniques are based on the idea of properly exciting
some magnetic saliencies of the machine, depending on the
machine speed, and then retrieving the machine speed/position
information by demodulating the resulting signal produced by the
carrier excitation. Two basic saliency techniques exist: one is
based on the injection of a high-frequency carrier (Holtz, 2006;
Raca et al., 2010; Zhu & Gong, 2011), the other is based on the
selection of proper inverter switching patterns (test vectors)
(Raute et al., 2010; Robeischl & Schroedl, 2004). Particularly,
carrier based techniques can be divided into two further cate-
gories: rotating voltage carrier (Corley & Lorenz, 1998; Jansen &
Lorenz, 1995; Kim, Harke, & Lorenz, 2003) and pulsating voltage
carrier (Cupertino, Pellegrino, Giangrande, & Salvatore, 2011;
Ferreira & Kennel, 2006; Kennel, Ferreira, & Szcupak, 2006;
Linke, Kennel, & Holtz, 2002). All these last techniques, which
permit the drive to work with high performance even at very low
and zero speed, present two main drawbacks:
they fail to work above a certain threshold speed;
they do not properly work if the saliency ratio of the machine is
poor (as in the typical case of Surface Mounted Synchronous
Machine).
Model-based techniques, on the contrary, do not suffer from these
limitations.
1.2. Model-based techniques
Model based encoderless techniques can exploit the back
electromotive force (EMF) or the flux linkage estimation, adopting
e.g. a state observer or an extended Kalman filter, to retrieve the
rotor position information (Bolognani, Oboe, & Zigliotto, 1999;
Chan, Wang, Borsje, Wong, & Ho, 2008; Cheng & Li, 2011; Delli
Colli, Di Stefano, & Marignetti, 2010; Ertugrul & Acarnley, 1994;
Hasegawa, Yoshioka, & Matsui, 2009; Kim & Sul, 1995; Luo, Chen,
Ahn, & Pi, 2010; Morimoto, Kawamoto, Sanada, & Takeda, 2002;
Tomei & Verrelli, 2011; Wu & Slemon, 1991; Yousfi, Azizi, & Saad,
2000). Otherwise, the error between the electrical signals mea-
sured from the drive and those estimated from the model can be
used to compute the rotor position (Bruguier, Champenois, &
Rognan, 1995; Harnefors & Nee, 2000; Matsui, 1996). Fundamen-
tal magnetomotive force (MMF) based methods are applied
Contents lists available at SciVerse ScienceDirect
journal homepage: www.elsevier.com/locate/conengprac
Control Engineering Practice
0967-0661/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conengprac.2012.03.012
n
Corresponding author.
E-mail addresses: angelo.accetta@unipa.it (A. Accetta),
maurizio.cirrincione@utbm.fr (M. Cirrincione), pucci@pa.issia.cnr.it (M. Pucci).
Control Engineering Practice 20 (2012) 725–732