Professional Documents
Culture Documents
Crossmark: Control Engineering Practice
Crossmark: Control Engineering Practice
A R T I C L E I N F O A BS T RAC T
Keywords: In this paper a robust sensorless control for an Interior Permanent Magnet Synchronous Motor (IPMSM) is
Nonlinear sensorless control designed. The proposed control strategy uses a backstepping controller, whose robustness is improved by using
Interior Permanent Magnet Synchronous integral actions added at each step of the original algorithm, and by a Maximum-Torque-Per-Ampere strategy
Motor (IPMSM) (MTPA) to improve its energy efficient operation. Furthermore, to implement this controller in the framework of
Nonlinear observer
the mechanical sensorless control from the only measurements of the currents and voltages, an adaptive
Backstepping
interconnected high gain observer is developed for estimating the rotor speed, the position and the load torque.
Experimental results
Moreover, sufficient conditions are given to ensure the practical stability of the Observer-Controller system even
if bounded uncertainties occur. Finally, the performance and the effectiveness of the designed method are tested
experimentally throw a significant benchmark including different speed references and with significant
robustness tests. A comparative evaluation of the computational effort of our scheme with respect to classical
motor control is given.
1. Introduction 2013b). The one based on the high-frequency injection signal brings
high-frequency noise. In most cases, this noise happens to be at
The permanent-magnet synchronous motor (PMSM) has been frequencies higher that the bandwidth of the regulators which can be
widely used in the industry for variable speed applications due to its avoided by using analogical filter (Qiao et al., 2013) but has to be below
high performance reliability and its power density. Owing to the the PWM frequency.
progress in the permanent magnet materials, micro and power On the other hand, the observer design based on the sliding mode
electronics, fast digital signal processors and modern control technol- techniques is widely used (Qiao et al., 2013) thanks to its performance
ogies, the permanent magnet synchronous machines have become and its robustness (Ren, Liu, Wang, & Liu, 2015). Nevertheless, the
more widespread in the industrial applications, for example: auto- chattering phenomenon presents the principal drawback of this kind of
mobiles, robotics, aeronautics and aerospace domains. observers (Liu, Laghrouche et al., 2014; Wang, Li, Zhang, Yu, & Xu,
Furthermore, high-performance controls require continuous mea- 2013). Furthermore, due to its simplicity, the back electromotive force
surement of rotor position and speed. Then, due to the cost and is widely used for the rotor speed estimation (Lu, Lei, & Blaabjerg,
reliability constraints (Betin et al., 2014; Delpoux, Bodson, & Floquet, 2013a).
2014; Liu & Zhu, 2014), the elimination of shaft position sensors is Regarding the observation problem of PMSM, the observer design
desirable for industrial applications (Kshirsagar et al., 2012; Raggl, of a PMSM using the methods based on the back EMF is more sensitive
Warberger, Nussbaumer, Burger, & Kolar, 2009). to parametric deviations. More precisely, the information of position is
In the literature, several methods have been proposed to estimate found in the back EMF and in the inductances can be used to estimate
the rotor speed in order to achieve high performance sensorless the position and the speed. Nevertheless, the electromotive force
control. These methods are arranged in two principal classes of the becomes too small at very low speeds which make the speed estimation
non model-based approaches: the high-frequency injection (Kim, Ha, very difficult. To address this problem, the extended electromotive
& Sul, 2012; Liu & Zhu, 2014) and artificial intelligence methods force is proposed as an alternative strategy (Wang, Zhan, Zhang, Gui,
(Accetta, Cirrincione, Pucci, & Vitale, 2012; Lu, Lei, & Blaabjerg, & Xu, 2014). The rotor speed can be also obtained from the flux
⁎
Corresponding author.
E-mail addresses: assaad.hamida@hotmail.fr, hamida.assaad@univ-ouargla.dz (M.A. Hamida).
http://dx.doi.org/10.1016/j.conengprac.2016.11.012
Received 22 September 2015; Received in revised form 17 November 2016; Accepted 18 November 2016
0967-0661/ © 2016 Elsevier Ltd. All rights reserved.
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
linkage estimation (Xu, Zhang, & Liu, 2013). This method is easy to
implement. However, at low speeds is not sufficient to obtain a good
estimation when the parameters are not exactly known (Hamida,
Glumineau, & de Leon, 2012).
For most of the observers, the parameter uncertainties (Niapour,
Tabarraie, & Feyzi, 2014) and an unknown load torque represent one
of the main difficulties for the IPMSM observation and control.
Thus, in this paper, a nonlinear observer is implemented experi-
mentally to estimate on-line the rotor resistance and the load torque.
Then this estimation will be used for the sensorless control of the
IPMSM, which allows to increase the robustness of the proposed
controller-observer scheme under parameter variations.
It is worth mentioning that the most of the sensorless control
solutions proposed in the literature do not take into account this
difficulty. From this point of view, nonlinear robust controls as sliding
mode (Alwi & Edwards, 2014) and backstepping (Escareno,
Rakotondrabe, & Habineza, 2015) could be more efficient. In
Hamida et al. (2012), the classical backstepping control is robustified
by adding tracking errors integral actions. This controller has been
tested only in simulation. Moreover, in order to guarantee an energy
efficient operation of the IPMSM, this controller is combined with a
MTPA strategy maximize the output torque. Fig. 1. PMSM rotor permanent magnets layout: (a) Surface permanent magnets, (b)
Inset permanent magnets, (c) Interior permanent magnets (d) Flux concentrating.
The structure of the paper is as follows. In Section 2, the Permanent 1. Non-Salient Poles: the magnets are located in the rotor surface
Magnet Synchronous Motor (PMSM) and the nonlinear mathematical (Fig. 1a): Surface Permanent Magnet Synchronous Motor (SPMSM)
model of the Interior Permanent Magnet Synchronous Motor(IPMSM) 2. Salient Poles: the magnets are buried into the rotor (Fig. 1c and
are introduced. Furthermore, before to design an observer, the d): Interior Permanent Magnet Synchronous Motor (IPMSM).
observability of the system is firstly recalled in Section 3 and then, a
high gain adaptive observer is developed in Section 4. The convergence 2.1.2. Model of the IPMSM
of the proposed observer under parameter uncertainties is studied in The mathematical model of the IPMSM in the (d , q ) reference frame
Section 5. Furthermore in Section 6, a robust Backstepping Controller for sinusoidal stator current excitation is given by Pillay and Krishnan
is introduced. Next, the stability of the observer-controller scheme is (1989):
proved in Section 7. Experiments are carried out showing the effec- did R Lq 1 diq R L 1
= − s id + p Ωiq + vd = − s iq − p d Ωid − p ϕf Ω
tiveness of the sensorless control strategy and reported in Section 8. dt Ld Ld L d dt Lq Lq Lq
Finally, some conclusions are given in Section 9. 1 dΩ p f p 1 dθ
+ vq = (L d − Lq ) id iq − v Ω + ϕf iq − Tl =Ω
Lq dt J J J J dt (1)
2. Problem statement with
65
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
id , iq stator currents. ⎡ iq ⎤ ⎡ Lq ⎤
0 −L ⎥ ⎢ 0 p Ld iq ⎥
vd , vq stator voltages. A1 (·) = ⎢ q , A2 (·) = ⎢ ,
⎢⎣ ⎥ fv ⎥
Ω rotor mechanical speed. 0 0 ⎦ ⎢⎣ 0 − J ⎥⎦
θ rotor angular position. ⎡ ϕf ⎤
L
− p Ld Ωid − p L Ω ⎥ ⎡ 0 ⎤
J moment of inertia. g1 (·) = ⎢ q q , Φ = ⎢ 1⎥
fv viscous friction coefficient. ⎢ ⎥ ⎢⎣− J ⎥⎦
⎣ 0 ⎦
p number of pole pairs.
Tl load torque. ⎡ R
− L s id ⎤ ⎡1⎤ ⎡1⎤
g2 (·) = ⎢⎢ p d ⎥,
⎥ Φ1 = ⎢ Lq ⎥ , Φ 2 = ⎢ Ld ⎥,
p ⎢⎣ 0 ⎥⎦ ⎢⎣ 0 ⎥⎦
⎣ J ϕf iq + J (L d − Lq ) iq id ⎦
Definition 1. The IPMSM physical operation domain + is defined by
the set of values C1 = C2 = [10],
+ = {X ∈ 5| id ≤ Idmax , iq ≤ Iqmax , Ω ≤ Ω max , Tl ≤ Tlmax , Rs ≤ Rsmax } with X1 = (iq Rs )T , X2 = (id Ω )T are the states, u = (vd vq )T is the input, and
y = (id iq )T is the output of the IPMSM model. Furthermore, g1 and g2
are the interconnection terms.
with X = [id iq Ω Tl Rs]T and Idmax, Iqmax, Ωmax, Tlmax and Rsmax are
For each subsystem, the state of the other subsystem is supposed
the actual maximum values for currents, speed, load torque and stator
available. In the sequel, a nonlinear observer will be developed for each
resistance, respectively.
subsystem. Then, the nonlinear observer for the IPMSM is composed
from the interconnection of the two observers.
3. IPM synchronous motor observability analysis
Furthermore, the convergence of the interconnected observer for
the IPMSM model, represented by (2),(3) and (4), is ensured provided
In this section, the observability property of the IPM synchronous
that the following properties are satisfied (see Besancon & Hammouri,
machine is analyzed. This study is made to decide if it is possible to
1998 for details):
compute the estimates of the rotor position, the rotor speed, the load
torque and also the stator resistance parameter based only on the
1. The matrices A1 et A2 are only functions of the measurement y1 = iq
measured stator voltages and the stator currents (for more details see
and the observability property of each subsystem is ensured if the
Hamida, Leon, & Glumineau, 2014).
signal y1 = iq has sufficient information to ensure the observability
From the mathematical model of the IPMSM in the d-q reference
property.
frame (1), based on the observability analysis given in Hamida et al.
2. The term g1 (X2 ) is globally Lipschitz w.r.t. X2.
(2014), and for the IPM Synchronous Motor, i.e. (Lq ≠ L d ), it follows
3. The term g2 (X1, y2 ) is globally Lipschitz w.r.t. X1 and uniformly w.r.t.
that the rotor speed and rotor position of the IPMSM are observable for
y2.
all rotor speed values, including rotor speed equal zero. For the SPM
Synchronbous Motor, i.e. (Ld=Lq), the rotor speed and rotor position
Then, a nonlinear adaptive interconnected observer for (3) and (4) is
of the IPMSM are observable, if the rotor speed is different from zero.
given by:
This can be summarized as follows.
⎧ Z˙ = A (y ) Z + g (Z ) + Φ (u ) + S −1 C T (y − yl )
1. The Surface PMSM (SPMSM) is not observable at zero speed. ⎪ 1 1 1 1 1 2 1 1 1 1 1
O1: ⎨ S˙1 = −ρ1 S1 − A1T (y1) S1 − S1 A1 (y2 ) + C1T C1
2. The rotor speed and the rotor position of the Interior PMSM ⎪
(IPMSM) are observable at all speed ranges including at zero speed. ⎩ yl1 = C1 Z1 (5)
3. The stator resistance and the load torque of the Interior PMSM are
⎧ Z˙ = A2 (y1) Z 2 + g2 (Z1, y2 ) + Φ 2 (u ) + ΦTll + KC1T (y1 − yl1)
observable at all speed ranges except at standstill (the currents and ⎪ 2
the voltages are null). This particular case can be easily detected by ⎪ + (ϖΛS3−1 ΛT C2T + ΓS2−1 C2T )(y2 − yl2 )
⎪
the stator measurements. ⎪ Tl˙l = ϖS3−1 ΛT C2T (y2 − yl2 ) + B (Z1)(y1 − yl1) + B1 (Z1)(y2 − yl2 )
⎪
O2 : ⎨ S˙2 = − ρ2 S2 − A2T (y1) S2 − S2 A2 (y1) + C2T C2
4. Nonlinear interconnected observers design for IPMSM ⎪
⎪ S˙3 = − ρ3 S3 + ΛT C2T C2 Λ
⎪ ˙
In this section, a nonlinear interconnected observer is designed for ⎪ Λ = {A2 (y1) − ΓS2−1 C2T C2} Λ + Φ
⎪ yl
estimating simultaneously the rotor speed, rotor position, load torque ⎩ 2 = C2 Z 2 (6)
and the stator resistance. For this purpose the mathematical model of
the PMSM given in (1) can be represented in two subsystems, where T T
the dynamics of the load torque and the stator resistance are included with Z1 = [i^q mRs ] and Z 2 = [i^d Ω l ] are the estimated state variables
in each subsystem. These dynamics are assumed to be piece-wise respectively of X1 and X2. ϖ is a positive constant, which represents the
functions, which can be expressed as follows: gain of the load torque observer. Λ is the solution of dynamical system,
which works as a filter system. The ρi are positive constants, for
T˙l = KTl R˙s = K Rs, (2) i = 1, 2, 3. Furthermore, S1 and S2 are symmetric positive definite
with KTl and K Rs are positive constants. matrices (Besancon & Hammouri, 1996), with S3 (0) > 0 ,
B (Z1) = k J Φf i^q , B1 (Z1) = k J (L d − Lq ) i^q
p p
From (1) and (2), the IPMSM mathematical model can be decom-
posed into two interconnected subsystems (3) and (4) (Hamida et al., ⎡ − kc1 ⎤ ⎡ ⎤
2012): K = ⎢− k ⎥ , Γ = ⎢1 0 ⎥
⎢⎣ c2 ⎥⎦ ⎣0 α⎦
⎧ X˙1 = A1 (y ) X1 + g (X2 ) + Φ1 u
Σ1: ⎨ 1 1
⎩ y1 = C1 X1 (3)
with k, kc1, kc2 and α and are positive constants.
⎧ X˙2 = A2 (y ) X2 + g (X1, y ) + Φ 2 u + ΦTl From the structure of the proposed observer, the first subsystem (5)
Σ 2: ⎨ 1 2 2
represents an observer used to estimate the current iq and the stator
⎩ y2 = C2 X2 (4)
resistance Rs. This is a copy of the system plus a term depending of the
where error between the measured and estimated outputs. Furthermore, the
66
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
gain of this observer is given by S1−1 C1T , which depends on the solution ⎡ ⎛L ⎞ ⎤
of a matrix differential equation of S1 (differential Ricatti equation). ⎢ 0 pΔ ⎜ q ⎟ iq ⎥
ΔA2 (·) = ⎢ ⎝ Ld ⎠ ⎥,
The solution of the differential equation on S1 uses the following lemma ⎢ f ⎥
(see proof in Besancon, De Leon, & Huerta, 2006), which will be used ⎢⎣ 0 Δ ( Jv ) ⎥⎦
to ensure the convergence of the developed observer. ⎡ ⎛1⎞ ⎤
⎢ − Rs Δ ⎜ L ⎟ id ⎥
Lemma 1. Assume that the input v is regularly persistent for a given Δg2 (·) = ⎢ ⎝ ⎠d ⎥.
state affine system and consider the following Lyapunov differential ⎢ 1 1 ⎥
⎢⎣ − pΔ ( J
) ϕ i i
f d q − pΔ ( J
)(Δ L d − ΔL q q⎥
) i ⎦
equation
The Δ(.) are bounded uncertain terms with respect to the uncertain
S˙ (t ) = −θS (t ) − AT (v (t )) S (t ) − S (t ) A (v (t )) + C T C.
parameters, then there exist positive constant ϱi > 0 , for i = 1, …, 4 ,
such that:
with S (0) > 0 , then ∃ θ0, ∀ θ ≥ θ0, ∃ α > 0, β > 0, t0 > 0 :
ΔA1 ≤ ϱ1, ΔA2 ≤ ϱ2 , Δg1 ≤ ϱ3, Δg2 ≤ ϱ4 .
∀ t0, αI ≤ S (t ) ≤ βI
The estimation errors are defined as:
were I is the identity matrix.
From Lemma 1, it is clear that for subsystem (5), v = y1 and ϵ1 = X1 − Z1, ϵ′2 = X2 − Z 2, ϵ3 = Tl − Tll . (10)
S (t ) = S1. From Eqs. (5)–(6) and (8)–(9). Using the same idea as in Zhang
For the second subsystem (6), an observer is designed to estimate (2002), we introduce the following transformation ϵ2 = ϵ′2 − Λϵ3. The
the current id, the rotor speed Ω, and the load torque Tl. The current id estimation errors dynamics can be expressed as:
and the rotor speed Ω are estimated by using the observer, whose gain
is split into two parts. The first part is the gain ΓS2−1 C2T obtained from ϵ̇1 = [A1 (y1) − S1−1 C1T C1]ϵ1 + g1 (X2 ) − g1 (Z 2 ) + ΔA1 (y1) X1 + Δg1 (X2 ) (11)
the matrix differential equation of S2. Similarly, from Lemma 2, for
subsystem (6), it follows that v = y1 and S (t ) = S2 . The second part is ϵ̇2 = [A2 (y2 ) − ΓS2−1 C2T C2 − ΛB1 (Z1) C2]ϵ2 − [ΛB (Z1) C1 + KC1T C1]ϵ1
the term associated to the identification of the load torque, which gain + {Φ − ΛB1 (Z1) C2 Λ}ϵ3 + g2 (X1, y2 ) − g2 (Z1, y2 )
is given by ϖΛS3−1 ΛT C2T and it depends on the solution of the differential
+ ΔA2 (y1) X2 + Δg2 (X2 , y1) (12)
equations of S3 and Λ.
Remark 1. The load torque Tl is estimated using Eq. (6) where the ϵ̇3 = −B (Z1) C1 ϵ1 − [ϖS3−1 ΛT C2T + B1 (Z1)] C2 ϵ2 − [ϖS3−1 ΛT C2T + B1 (Z1)] C2
term B (Z1)(y1 − yl1) + B1 (Z1)(y2 − yl2 ) can be expressed, in terms of the Λϵ3. (13)
∼
electromagnetic torque Te and its estimated Te , as follows:
From Lemma 1, there exist t0 ≥ 0 and real numbers ηSmax > 0,
∼ i
B (Z1)(y1 − yl1) + B1 (Z1)(y2 − yl2 ) = k (Te − Te ). ηSmin > 0 , which are independent of ρi, such that V (t , ϵi) = ϵTi Si ϵi for
i
i = 1, 2, 3; satisfies
Furthermore, to estimate the rotor position, the following observer is
designed: ηSmin
i
ϵi 2 ≤ V (t , ϵi) ≤ ηSmax
i
ϵi 2 , ∀ t ≥ t0. (14)
dθl l + Kθ (iq − i^q ) Then, the following result can be established about the observer
=Ω
dt (7) convergence in spite of bounded parametric uncertainties.
where Kθ is the observer gain. Theorem 1. Consider the IPMSM dynamic model represented by (3)
and (4). System (5) and (6) is an adaptive interconnected observer for
system (3) and (4) with strongly uniformly practical stability of the
5. Observer convergence
estimation error dynamics.
The electrical machine parameters may vary which affects the Proof of Theorem 1. Consider the following Lyapunov function
operation of the machine. The influence of the parameter uncertainties
on the proposed will be studied in this section. The practical stability Vo = ϵ1T S1 ϵ1 + ϵT2 S2 ϵ2 + ϵT3 S3 ϵ3.
theorem introduced in Lakshmikantham, Leela, and Martynyuk (1990)
will be used to ensure the stability of the proposed observer under Then, taking the time derivative and replacing the suitable terms, it
parameter uncertainties (see the Appendix A for more details). From follows that
system (1), the IPMSM parameters are defined in the physical domain V˙o = −ρ1 ϵ1T S1 ϵ1 − ρ2 ϵT2 S2 ϵ2 − ρη ϵT3 S3 ϵ3 + ϵ1T S1 {g1 (X2 , u ) − g1 (Z 2, u )}
+ (Definition 1) with well-known bound values. The interconnected
IPMSM uncertain model is presented as follows: + ϵT2 S2 {[A2 (y1) − A2 (yl1)] X2 + g2 (X1, u , y2 ) − g2 (Z1, u, yl2 )
67
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
ψ V1 + V2 + V3 > V1 + V2 .
z Ω = Ω* − Ω + k′Ω ∫0 (Ω* − Ω ) dt
(28)
t
Then, the following inequalities hold with k′Ω ∫ (Ω* − Ω ) dt represents an integral action introduced to
0
guarantee the rotor speed tracking despite parameter uncertainties.
⎧ ρ > λ∼ φ ,
⎪ 1 1 1 In order to ensure the speed tracking, iq will be forced to track iq* by
⎪ ∼ means of the control law vq.
⎪ ρ2 > λ2 φ2 + λ1 + λ∼3,
∼
⎨ φ1 Computing the time derivative of (28), and replacing the current iq
⎪ ∼ by iq*, it follows that:
⎪ λ2
⎪ ρη > φ . p f p 1
⎩ 2 (22) z˙Ω = Ω˙* − (L d − Lq ) id iq* + v Ω − ϕf iq* + Tl + k′Ω (Ω˙* − Ω˙ ).
J J J J (29)
Furthermore, it follows that there exist positive constants, λ min (S ) and
λ max (S ), where S = diag (S1, S2, S3) and ϵ(t ) = (ϵ1T (t ), ϵT2 (t ), ϵT3 (t ))T , such To study the stability of the speed loop, the following candidate
1
that Lyapunov function is chosen as VΩ = 2 zΩ2 . The time derivative of the
above Lyapunov candidate function can be computed as follows:
λ min (S )∥ ϵ(t )∥2 ≤ Vo = ϵ(t )T S ϵ(t ) ≤ λ max (S )∥ ϵ(t )∥2 .
⎧ p f p 1 ⎫
Then, it is obtained V˙Ω = z Ω ⎨Ω˙* − (L d − Lq ) id iq* + v Ω − Φf iq* + Tl + k′Ω (Ω˙* − Ω˙ ) ⎬.
⎩ J J J J ⎭
V˙o ≤ −δVo + μψ Vo , ≤−δλ min (S )∥ ϵ(t )∥2 + μψ λ max (S ) ∥ ϵ(t )∥. (23) (30)
Taking γ such that 0 < γ < δλ min (S ), then the above inequality can be To ensure its negativity and then, the stability of the speed loop, iq* is
expressed as follows chosen as:
J ⎡ fv 1 ⎤
V˙o ≤ −γ ∥ ϵ(t )∥2 + ∥ ϵ(t )∥{(γ − δλ min (S ))∥ ϵ(t )∥ + μψ λ max (S ) }. (24) iq* = ⎢kΩ z Ω + Ω˙* + Ω + k′Ω (Ω* − Ω ) + Tl ⎥ .
p ⎣ J J ⎦
pϕf + (L d − Lq ) id
Then, using this inequality, it is easy to show that J
(31)
V˙o ≤ −γ ∥ ϵ(t )∥2 (25)
Then V˙Ω = −kΩ zΩ2 , with kΩ > 0 . Hence, the speed tracking is ensured.
μψ λmax (S )
if ∥ ϵ(t )∥ ≥ (δλmin (S ) − γ )
= δ*, and from Theorem 5.1 and Corollary 5.3 in Current iq loop. First the following tracking error is defined
[? ], if ∥ ϵ(0)∥ = δ*, there exists To, such that
z q = iq* − iq + z′q (32)
γ
λ max (S ) − t t
∥ ϵ(t )∥ ≤ ∥ ϵ(0)∥ e (2λmax (S ) , ∀ t < To with z′q is an integral action z′q = k′q ∫ (iq* − iq ) dt added to robustify
λ min (S ) (26) 0
the current tracking. The following Lyapunov candidate function is
and chosen to study the stability of the current loop as:
68
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
1 2 1 2 to ensure its negativity. The control law vd which make the d axis
Vq = VΩ + zq + zq ′.
2 2 (33) current loop stable is chosen as follows:
From the mathematical model of the IPMSM, the time derivative of Vq ⎡ t R Lq ⎤
can be computed as follows. To ensure its negativity and then the vd = L d ⎢kd (id* − id + k′d
⎣
∫0 (id* − id ) dt ) + id* + s iq − p iq Ω⎥ .
Ld Ld ⎦ (43)
stability of the current loop, the control law vq is chosen as:
The time derivative of Vd is then:
⎡ ϕf R diq* ⎤
vq = Lq ⎢kq z q + p Ω + s iq + ⎥.
V˙d = −(kd − k′d ) z d2 − k′d z d2′.
⎢⎣ Lq Lq dt ⎥⎦ (34)
From (32) and (34) it follows that By choosing Kd = min{(kd − k′d ), k′d }, it follows that
V˙d ≤ −Kd Vd .
V˙q = −kΩ zΩ2 − (kq − k′q ) zq2 − k′q zq2′ ≤ −Kq Vq (35)
Then, the control input vd force the d axis current to track its reference
where Kq = min{kΩ, (kq − k′q ), k′q }. The control law vq ensures that the
computed by MTPA strategy which allows to achieve the second control
current iq tracks the virtual input iq*. This virtual input is computed to
objective.
assure the speed tracking, i.e. (Ω → Ω*). From where, the first control
objective is reached.
Current id loop. 7. Observer-controller scheme stability analysis
The Maximum Torque Per Ampere (MTPA) strategy is proposed to
obtain the most of reluctance torque in the IPMSMs. In the last years, In the previous section, the control laws are computed using a
some effective proposals based on two schemes of the MTPA control perfect system state knowledge. Even if an observer is designed to
have been presented (Uddin & Rahman, 2007). Here, the strategy reconstruct the system state, in practice the observer cannot provides
adopted to obtain the current id reference is based on the mathematical ideal estimates due to disturbances and uncertainties. It is why, in this
model of the IPMSM. The IPMSM electromagnetic torque is given by: section, the stability of the whole system (observer-controller) will be
studied based on the practical stability theory (Lakshmikantham et al.,
Te = p {ϕf iq + (L d − Lq ) id iq}. (36) 1990). In the appendix (Section Appendix A) the necessary definitions
are recalled. To ensure the stability of the observer-controller scheme,
The second term of the electromagnetic torque expression in (36) is
the observation dynamics are taken into account. Following the proof
subject of several researches in order to guarantee an energy efficient
in Hamida et al. (2012), we consider the following candidate Lyapunov
control of the IPMSM. The MTPA strategy (Rahman, Vilathgamuwa,
function:
Uddin, & Tseng, 2003) is developed to provide a maximum torque/
current ratio. The MTPA strategy is now summarized. The stator phase 1 2 1 1 1 1
Voc = Vo + Vc = zΩ + zq2 + zq2′ + z d2 + z d2′ + ϵ1T S1 ϵ1 + ϵT2 S2 ϵ3
Ia current is composed from id and iq as follows: 2 2 2 2 2
Ia2 = iq2 + id2. + ϵT3 S3 ϵ3. (44)
(37)
Taking into account the observation dynamics in the control laws
From (36) and (37) the electromagnetic torque can be expressed in
and by adding ± vq (X ) and ± vd (X ), the time derivative of Vc is obtained
terms of id and Ia:
by:
Te = p [ϕf + (L d − Lq ) id ] Ia2 − id2 . (38) 1
V˙c = −kΩ zΩ2 − {kq − k′q } zq2 − k′q zq2′ − l ) − vq (X )}
z q {vq (X
The derivative of the electromagnetic torque with respect to the current Lq
id is given by: 1 l ) − vd (X )}.
− {kd − k′d } z d2 − k′d z d2′ − z d {vd (X
Ld (45)
∂Te −ϕf id + (L d − Lq )(Ia2 − 2id2 )
=p . Now, considering the following inequalities:
∂id Ia2 − id2 (39)
ξ j1 2 1 ξj2 2 1
From 31, the torque is maximum when
∂Te
= 0 , i.e. |zj | ϵ1 s1 ≤ ϵ1 s1 + |zj |2 |zj | ϵ2 s2 ≤ ϵ2 s2 + |zj |2 |zj
∂id 2 2ξ j1 2 2ξ j 2
ϕf ξj3 1
2id2 + id − Ia2 = 0. | ϵ3 s3 ≤ ϵ3 2
s3 + l ) − vq (X )
|zj |2 |vq (X
(L d − L q ) (40) 2 2ξ j 3
Then, by solving the second order equation (40) the d-axis current | ≤ L1 { ϵ1 s1 + ϵ2 s2 + ϵ3 s3 }|vd (X )
l − vd (X )
reference is given by: | ≤ L 2 { ϵ1 s1 + ϵ2 s2 + ϵ3 s3 } (46)
ϕf ϕf2 ∀ ξ j1, ξ j 2, ξ j 3 ∈ ]0 1[ ; for j=q, d; and by substituting equation (46) into
id* = − − + iq2 .
2(L d − Lq ) 4(L d − Lq )2 (41) (45), we obtain
Now, in order to guarantee an energy efficient operation of the IPMSM V˙oc ≤ −δVo + μφ Vo + ϑ1 ϵ1 2
s1 + ϑ2 ϵ2 2
s2 + ϑ3 ϵ3 2
s3 − ϑ4 zΩ2 − ϑ5zq2
via the MTPA strategy, the current id will be forced to track its − ϑ6 zq2′ − ϑ7z d2 − ϑ8z d2′ (47)
reference given by (41). The d axis current tracking error is defined as:
L ξL1ξq1 L1ξq2 L ξ L1ξq3 L2 ξd 3
where ϑ1 = 2Lq + 22Ldd1 , ϑ2 = 2Lq + 22Ldd 2 , ϑ3 = + , ϑ4 = kΩ ,
z d = id* − id + z′d , 2Lq 2Ld
L1 1 1 1
t ϑ5 = {kq − k′q } − 2Lq { ξ + ξ + ξ }, ϑ6 = k′q , ϑ7 = {kd − k′d }−
q1 q2 q3
the integral term z′d = k′d ∫ (id* − id ) dt is added to eliminate the L2 1 1 1
0 { + ξ + ξ }, 8 ϑ = k ′d.
current tracking error. 2Ld ξ d1 d2 d3
The following candidate Lyapunov function is defined to study the By taking ϑO = max(ϑ1, ϑ2, ϑ3) and ϑC = min(ϑ4, ϑ5, ϑ6, ϑ7, ϑ8), then
stability of the current id loop: the inequality (47) becomes:
69
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
70
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
200
Speed (rad/s)
150
100
50
0
−50
2 4 6 8 10 12 14
Time (s)
10
Torque (rad/s)
5
−5
−10
2 4 6 8 10 12 14
Time (s)
200 10
Ω obs
Ω real 8
150 6
100 2
0
50
−2
−4
0
−6
−8
−50
−10
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 5. Speed tracking: nominal case. (a) real and estimated speeds, (b) speed error.
a
0
−1
−2
Position (rad)
−3
−4
−5
−6
−7 θ real θ est
0 0.5 1 1.5 2 2.5 3 3.5 4
Time (s)
different to zero, but is bounded. After 10 s, the rotor speed takes a 8.2.1. Robustness study
high value, and then the estimation is improved. This response Now, the robustness of the sensorless control strategy is tested. The
confirms the effectiveness of the proposed observer in this specific torque estimation is shown in Fig. 9a, where one can see the good
operation condition (noted nominal case). estimation when an unknown load torque is applied (see the trajectory
Furthermore, the convergence of the estimates to the actual values of the load torque given in Fig. 4). The error torque observation is given
is ensured and the estimates are bounded. For high rotor speeds the in Fig. 9b (Fig. 10).
convergence of the position estimation to the actual values is verified. It is well-known that the machine parameters can be deviated from
In Fig. 8, the rotor resistance estimation is shown. This estimation the nominal values under temperature variations or magnetic circuit
is used to improve the robustness of the proposed observer-controller saturation. Then, taking into account these parametric deviations in the
scheme. observer-controller scheme, the proposed sensorless control strategy is
experimentally tested with the benchmark reference trajectories
71
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
−1
−2
Position (rad)
−3
−4
−5
−6
θ mes θ obs
−7
8 8.5 9 9.5 10 10.5 11
Temps (s)
1.41
1.405
Resistance (ohm)
1.4
1.395
1.39
0 5 10 15
Time (s)
10 2
8
1.5
6
1
Torque error (N.m)
4
Torque (N.m)
0.5
2
0 0
−2
−0.5
−4
−1
−6
Tl −1.5
−8
T l −obs
−10 −2
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 9. Torque estimation: nominal case. (a) real and estimated torques, latex (b) torque error.
(Fig. 4). More precisely, the proposed sensorless control strategy is All these results show clearly the effectiveness of the developed
tested under inductances variations. The system performances under sensorless control strategy.
+20% and −20% variations on the inductance values are respectively
shown on Figs. 11–14. As mentioned before, the inductances affect the 8.3. Comparative study and implementation
behavior to estimate the rotor position when the EMF methods are
implemented to estimate the position. For a more detailed robustness Now, to illustrate the performance of the proposed control scheme,
analysis, see Marwa Ezzat and Glumineau (2010). Then, we can a comparative study will be presented. Here, the Integral Backstepping
conclude that the proposed methodology performs well under these Control proposed in this paper and the High Order Sliding Mode
operating conditions. controller analyzed in Hamida, Glumineau, and de Leon (2013) are
The quality of the load torque estimation is clearly displayed in implemented.
Figs. 9b, 12b, 14b for the nominal and robustness test cases. The It is worth mentioning that the implementation of both control
quality of the tracking is also easy to check in Figs. 5b, 11b, 12b, 13b strategies have been made on the same motor set-up, using the same
also for the nominal and robustness test cases. hardware and software and under the same operation conditions and
72
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
6 400
5
300
4
200
Voltage (V)
Current (A) 3
id
100
iq
2
0
1
0 −100
ud
uq
−1 −200
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 10. Input signals (a) d − q voltages, (b) d − q currents.
200 10
Ω real
Ω obs 8
150 6
100 2
0
50 −2
−4
0
−6
−8
−50
−10
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 11. Speed tracking: robustness w.r.t. +20% Ld-Lq deviation. (a) real and estimated speeds, (b) speed error.
10 2
8 1.5
6
1
Torque error (N.m)
4
Torque (N.m)
0.5
2
0 0
−2 −0.5
−4
−1
−6
Tl −1.5
−8
T l −obs
−10 −2
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 12. Torque estimation: robustness w.r.t. +20% Ld-Lq deviation. (a) real and estimated torques, (b) torque error.
the industrial test trajectories. Sliding Mode Controller (see Hamida et al., 2013 for more details) in
From the experimental results obtained, we can remark that. closed-loop with either an Adaptive Interconnected Observer (given in
Hamida et al., 2012) or a HOSM Observer (see Hamida et al., 2014), is
(1) The Integral Backstepping Control shows good transient perfor- presented to illustrate that the proposed method is computationally
mance. satisfactory.
(2) For comparison the integral Backstepping Control is less sensitive The running time of each Algorithm (Controller +Observer) was:
to the noise than switching control as High Order Sliding Mode 38 μs for the Backstepping strategy and 58 μs for Sliding Mode strategy.
Controller even if the high frequency content is attenuated with On the other hand, the running time using PI+FOC strategy (28 μ) is
respect to the classical sliding mode control (Hamida et al., 2013). less than Backstepping strategy.
Notice that the computation time required by the Backstepping
In Table 2, a computation time comparison between a FOC-PI Controller is similar to a classical FOC-PI Controller which is widely
controller, an Integral backstepping Controller and a High Order used in the industrial applications with low-cost hardware. Moreover,
73
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
200 10
Ω obs
Ω real 8
150 6
Speed (rad/s)
100 2
0
50 −2
−4
0
−6
−8
−50
−10
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 13. Speed tracking: robustness w.r.t. −20% Ld-Lq deviation. (a) real and estimated speeds, (b) speed error.
10 2
8
1.5
6
1
Torque error (N.m)
4
Torque (N.m)
2 0.5
0 0
−2
−0.5
−4
−1
−6
−8 Tl −1.5
T l −obs
−10 −2
0 5 10 15 0 5 10 15
Time (s) Time (s)
Fig. 14. Torque estimation: robustness w.r.t. −20% Ld-Lq deviation. (a) real and estimated torques, (b) torque error.
74
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
Appendix A
This section is devoted to introduce some concepts and results of practical stability properties using in terms of Lyapunov functions
(Lakshmikantham et al., 1990).
Theory of stability is the basis of the control systems study. Moreover, the concept of practical stability allows to study the properties of a
nonlinear system when the state of this system is bringing close to a set instead of the equilibrium point. This is clear that, in practice, asymptotic
stability towards a domain whose the size has to be determined, for performance checking, is sufficient. If the behavior of the system can be bounded
by certain bounds, the notion of practical stability becomes useful.
Now, we introduce definitions which are useful for to guarantee the practical stability, in terms of Lyapunov functions (For more details see
Lakshmikantham et al., 1990).
Define the following class of function
W = {d1 ∈ C [R+, R+]: d1 (l ) is strictly increasing in l and d1 (l ) → ∞ as l → ∞}.
Let Br = {e ∈ Rn: e ≤ r}. Consider the dynamical system
e˙ = f (t , e), e (t0 ) = e0 , t0 ≥ 0, (50)
A result of the practical stability in terms of Lyapunov-like functions can be given from Lakshmikantham et al. (1990). Assume that
with α1 and α2 > 0, V0≔V (0, e (0)) ≥ 0 implies strong uniform practical stability of system (50).
The solution of equation (52) is of the form
α2
V (t , e) ≤ V0 e−α1t − t 0 + [1 − e−α1 (t − t 0) ], t ≥ t0.
α1 (53)
75
M.A. Hamida et al. Control Engineering Practice 59 (2017) 64–76
76