You are on page 1of 25

Int. J. Appl. Math. Comput. Sci., 2001, Vol.11, No.

1, 105–129

Emmanuel DELALEAU∗, ∗∗ , Jean-Paul LOUIS∗ Romeo ORTEGA∗∗∗

This paper is devoted to the modeling and control of the induction motor. The well-established field oriented control is recalled and two recent control strategies are exposed, namely the passivity-based control and the flatness-based control. Keywords: induction motors, field oriented control, passivity based control, flatness based control, nonlinear control

1. Introduction
Induction motors constitute a theoretically interesting and practically important class of nonlinear systems. They are described by a fifth-order nonlinear differential equation with two inputs and only three state variables available for measurement. The control task is further complicated by the fact that induction motors are subject to unknown (load) disturbances and the parameters are of great uncertainty. We are faced then with the challenging problem of controlling a highly nonlinear system, with unknown time-varying parameters, where the regulated output, besides being unmeasurable, is perturbed by an unknown additive signal. Existing solutions to this problem, in particular the de facto industry standard field-oriented control (FOC), were not theoretically well understood. Consequently, no guidelines were available for the designer who had to rely on trial-and-error analysis and intuition for commissioning and high performance applications. These compelling factors, together with the recent development of powerful theoretical tools for analysis and synthesis of nonlinear systems, motivated some control researchers to tackle this problem. The main purpose of this paper is to review some of the main developments in the field, with particular emphasis on applications of passivity and flatness ideas. We start

∗∗ ***

Work partly supported by the European Commission’s Training and Mobility of Researchers (TMR) Contract # ERBFMRX-CT970137. A preliminary version of this paper was presented as a mini-course at the MTNS 2000. ´ ´ Laboratoire d’Electricit´ e, Signaux et Robotique, Ecole Normale Sup´ erieure de Cachan, 61, avenue du Pr´ esident Wilson, 94 235 Cachan cedex, France, e-mail: Emmanuel.Delaleau, Jean-Paul.Louis On temporary leave from Laboratoire des Signaux et Syst` emes with a financial support of the Centre National de la Recherche Scientifique. Laboratoire des Signaux et Syst` emes, CNRS–Sup´ elec, Plateau de Moulon, 3 rue Joliot-Curie, 91 192 Gif-sur-Yvette cedex, France, e-mail:


E. Delaleau et al.

by presenting the physical model of the motor adopting an innovative perspective that underscores the control aspects—in lieu of the classical electrical engineering viewpoint. We then review, again from a control theory perspective, the well-known FOC. Connections between this classical technique and passivity ideas have been revealed in the literature; in particular, it has been shown that passivity-based control schemes exactly reduce to FOC under some simplifying modeling assumptions. After reviewing these results, we present the recent developments which rely on the property of flatness of the motor. The control of induction drives gave rise to a large number of publications which are not possible to be reported in total here. An overview of various aspects can be found in the papers (Bodson et al., 1994; 1995; Chiasson, 1993; 1995; De Luca, 1989; Marino et al., 1993; 1996; 1998; Ortega et al., 1993b; Taylor, 1994). The paper is organized as follows: Section 2 is devoted to the physical modeling of the induction motor on standard assumptions but in the perspective of the control of the machine. Section 3 gives a novel presentation of the well-established field-oriented control. Section 4 exposes the passivity-based control. Finally, Section 5 introduces the flatness-based approach to the control of the induction motor.

2. Modeling
2.1. Physical Modeling1 The induction machine considered here has a three-phase stator and a squirrelcage rotor which can be represented by a short-circuited three-phase rotor winding (see Fig. 1(a)). We adopt the classical assumptions: linearity of the materials (no saturation), sinusoidal distribution of the field in the air-gap, balanced structure. ∆ The vectors relative to stator variables are denoted2 by xabcs = (xas , xbs , xcs )t = t xabc = (xa , xb , xc ) and vectors relative to rotor variables are denoted the xabcr = t t ∆ (xar , xbr , xcr ) = xABC = (xA , xB , xC ) . Fluxes, currents and voltage are denoted respectively by ψabc , ψABC , iabc , iABC , vabc and vABC . The fundamental physical equations of the machine are the relations between fluxes and currents: ψabc = ls iabc + msr (θ) iABC , ψABC = mrs (θ) iabc + lr iABC , with ls  ls =  ms ms
1 2

(1a) (1b)

ms ls ms

 ms  ms  , ls

lr  lr =  mr mr

mr lr mr

 mr  mr  , lr

See (Chatelain, 1983; Semail et al., 1999) for details. Recall that v t is the transpose of a vector v .

dt (2a) (2b)  msr (θ) = mrs (θ)t = Mo   cos(np θ − γ )  cos(np θ) cos(np θ + γ ) cos(np θ − γ ) cos(np θ) cos(np θ + γ ) cos(np θ − γ )  cos(np θ + γ )  . dt The final equation is given by the expression of the electromagnetic torque 0 = Rr iABC + τem = it abc ∂ msr (θ) iABC . where np is the number of pairs of poles. but we prefer first to eliminate the terms relative to the “zero-sequence” . (1) and (3) is the coupling matrix msr (θ) which describes the electromechanical conversion. (b) Park transformation.Modeling and control of induction motors 107 Rotor axe © ¡¢ ¡ ¨§ ¥§×¦ b c ¨§© ¡ ¡ § ¥ ¦ × C ¨§© ¡ ¡ ¥§×¦ ¡ ¨§© B ¡ n¡ £¤ θ£ A ¨§a© ¥§×¦ ¨ © § ¡ ¡ ¡ ¥ ·¦ § ¥ צ ¨§© § ¡ ¡ ¥§·¦ a B ¨§© A ¡ ¥§·¦ ¨§© ¡ ¡ C ¥§·¦ ¡ ¨§© ¨§© ¡ ¡ c¥§¦ ¥§·¦ · p   Stator axe ¨§© ¥§×¦ βr q -axe  ¡¢ ¡  ¡ ¡ ¡ ¡  ¡ ¡    ¡ ¡        βs    d-axe        ξ    ξ      α      n θ  r s p r αs b (a) (b) Fig. (a) Transverse section. 1. θ signifies the mechanical position of the rotor and γ = 2π/3. The second system of equations is composed of the voltage equations: vabc = Rs iabc + dψabc . (3/2)Mo enp θ and (3/2)Mo e−np θ . Algebraic Properties of the Coupling Matrix msr (θ) The most important term in eqns.2. the other parameters (inductances) are constant. Its eigenvalues are 0. We must detail some of its algebraic properties. ∂θ (3) 2. cos(np θ)  dψABC . This matrix is diagonalizable.

3. 2. . we have for the voltages: vαβs = Rs iαβs + 0 = Rr iαβr + dψαβs . Concordia Transformation Examining eqns. T32 = √ √  . xβr )t = T32 xABC . with: Ls = ls − ms and Lr = lr − mr . it is useful to define the Concordia transformation applied to all electric variables (voltages v . we conclude that we do not have six unknown variables. (5) and (6). ψαβr = M P (−np θ) iαβs + Lr iαβr . where M = (3/2)Mo . (4)  3 3 3  sin(ξ ) cos(ξ ) 0 + − 2 2 (5a) (5b) (6) τem = np M it abc T32 P np θ + π 2 t T32 iABC . Then we will use a real transformation. dt dψαβr . Thus the fundamental equations (1) and (3) can be rewritten as t ψabc = Ls iabc + M T32 P (np θ ) T32 iABC .108 E. t t We have3 T32 T32 = I2 and the “factorization” msr = M T32 P (np θ) T32 . For this. (7a) (7b) (8) 3 Recall that In denotes the n-dimensional identity matrix. Then the equations of fluxes and torque (5) and (6) can be rewritten as ψαβs = Ls iαβs + M P (+np θ) iαβr . ∆ ∆ t t currents i) by xαβs = (xαs . xβs )t = T32 xabc and xαβr = (xαr . fluxes ψ . Therefore. define the planar rotation matrix P and the Concordia sub-matrix T32 as follows:  t 1 1 1 − − cos(ξ ) − sin(ξ ) 2 2  2  P (ξ ) = . Delaleau et al. t ψABC = M T32 P (−np θ) T32 iabc + Lr iABC . τem = np M it αβs P np θ + Furthermore. Furthermore. component (associated with the eigenvalue equal to zero) because they are almost always zero and do not participate in the energy conversion. T32 diagonalizes the matrices like ls and lr : ls T32 = Ls T32 and lr T32 = Lr T32 . t t but only four are given by T32 iabc and T32 iABC . dt (9a) (9b) π 2 iαβr .

0)t . but the so-called magnetizing current iµr . This new variable. fluxes): P (−ξs ) (xαs . Lr with the definition of the following new parameters: the dispersion coefficient: σ = 1 − M 2 /Ls Lr . We can rewrite the other variables (torque. stator fluxes. 0)t . xq ) . 0)t . and we will have to consider only two equations relative to the torque τem and to the magnitude of the rotor flux ψr . Choice of “Useful Variables” 109 Equations (7) and (8) show that we now have four unknown variables to determine the components of the stator and rotor currents. and the magnetizing current (magnitude and polar angle). ξs = ξr + np θ. This leads to the following torque equation which has the simplest form τem = np Lm iµr iq . and the magnetizing inductance Lm = (1 − σ )Ls = M 2 /Lr . rotor currents) with the help of these two vector variables: τem = np Lm iµr it αβs P np θ + ξr + π 2 (1. (10) Then it appears that it will be natural to choose the following “useful variables” (physical signification of the future state variables which will be defined in a later section): the stator currents. iαβr = M iµαβr − P (−np θ) iαβs . 2. (13) t t t t (12a) (12b) (12c) This transformation is. and its polar angle ξr are defined by the transformation: ψαβr = M iµαβr = M ∆ iµr cos(ξr ) iµr sin(ξr ) = M iµr P (ξr ) (1. xβr ) = (xD . . which are measurable. in fact. we decide to use not exactly the rotor flux. xQ ) .Modeling and control of induction motors 2. (11a) (11b) (11c) ψαβs = N1 iαβs + Lm iµr P (np θ + ξr ) (1. P (−ξr ) (xαr . Park Transformation Examination of (11a) indicates that it will be much simpler to write it if we make the following transformation for the stator variables (voltages. currents. a rotation of the axes and the d-axe is given by the direction of the rotor flux.5. For convenience. its magnitude iµ . the leakage inductance N1 = σ Ls = Ls − M 2 /Lr . xβs ) = (xd .4. It is known as the Park transformation (see Fig. 1(b)).

The problem has now two aspects: (a) What are the currents needed to impose the torque and the magnitude of the rotor flux? That is. Vas. 1994. Delaleau et al. 1999). we see that we have 6 unknown variables (the currents iabc and iABC ) and only 2 equations given by τem and ψr . 1990. The state equations which determine these state variables are respectively given for the equations deduced from the stator voltages: did 1 = (vd − ed ). 1997. Extensions are given in (D´ el´ emontey. Jelassi. 1991. 1990). State Variables and State Equations E. Graf von Westerholt. dt Tr dξr 1 iq = . . 1972. where Tr = Lr /Rr .110 2.. (12a)). 1995. (14a) dt N1 diq 1 = (vq − eq ). 2. Caron and Hautier. 1998. 1993. Controlling the electromagnetic torque τem . (1) and (3) which are algebraic equations? Examining (1) and (3). dt N1 with the following definitions for the back electromotive forces ed and eq : e d = R s id + Lm 1 iq (id − iµr ) − np ω + Tr Tr iµr 1 iq Tr iµr N1 i q . Field-Oriented Control4 The designers have chosen the following two criteria to have a good control of the IM: 1. Mendes. Robyns. and for the equations deduced from the rotor voltages: diµr 1 = (id − iµr ) . The most practical state variables are the stator currents in the Cartesian representation id and iq (cf. De Fornel. dt Tr iµr (16a) (16b) 3. 1995. 1997. how to “inverse” eqns. Jacquot. 1993). Our approach (“inversion of models”) is mainly deduced from the point of view of (Grenier et al. Louis and Bergmann. (14b) (15a) (15b) e q = R s iq + np ω + (N1 id + Lm iµr ) . Controlling the magnitude of the rotor flux ψABC : ψr . how to inverse (2) which are differential equations? Furthermore. Grellet and Clerc. and the magnetizing current in the polar form (10). 4 The classical field orientation is presented in (Blaschke. Leonhard. we must protect the motor against excessive magnitudes of stator currents. (b) What are the voltages which can create the appropriate currents? That is.6. 1995.

Closed-Loop Control and Introduction of Physical Constraints For the design of controllers.Modeling and control of induction motors 3. dt Tr dξ r 1 ıq = . dt dt (18a) diµr d2 iµr + N 1 Tr = vd − ed . (16a) and (16b): t ıdq = P (−ξs ) T32 iabc . τem = np Lm ıµr ıq .. 3.e. We want to impose the dynamics and the steady-state behaviors of the two variables of interest: the torque.. it is necessary to determine the variables which are not measurable. The symbol denotes the corresponding estimated variables. we have to solve two problems: 1. i. dt dt2 diµr = id . ξs . (17a) (17b) (17c) (17d) (17e) dıµr 1 = (ıd − ıµr ). iαβs = T32 iabc . τem . with the t help of the measurable variables. and the second for the magnetizing current): dτem 1 1 = (τem )ref − τem = np Lm iµr (Iqref − iq ) . stator currents.1. dt Tr ıµr ξs = n p θ + ξr . iµr + 2 ξ diµr 1 d2 iµr + 2 = iµrref ωn dt ωn dt2 (19a) (19b) . and the amplitude of the magnetizing current. ω and θ. The equations of the estimator are deduced from eqns. dt τc τc .e. iµr . ξr and iµr . dt (18b) (18c) iµr + Tr We choose arbitrarily the following dynamic models which have the lowest order physically realizable (the first for the torque. and mechanical variables. i.2. For this we will apply input-output linearization by state-feedback using the differential equations dτem = n p Lm dt N1 diµr diq iq + iµr . Estimator 111 If we want to impose a dynamic with the help of a state feedback.

We have to limit the variations of the stator currents for security during excessive movements. we control the torque and the magnitude of the component iq of the stator current: vq = kq (Iqref − ıq ) + eq − N1 ı q N1 (ıd − ıµr ). Eqn. Control scheme. kµ . the two-loop structure is given by Idref = kµ (Iµrref − iµr ) + iµr .112 Cref   1 np Lmˆ ıµr ¡ ¨        § ¤ © ¥©   ¦  ¡ max min Iqref + − ˆ ıq ˆ ıµr     E. Algebraic computations give the following results for the control law: (a) On the q -axis. given by (19b)). (17) © ¥¨©   § ¤¦ ¡ ¡ ¢ ¢ ¨© ¤¦   § © ¥ + + v q2   £   vq iabc θ Iµref ¢ ¨© ¤¦   §¦ © ¥   − + kµ      §¢ ¨ ¢ ¦ § ¨ © ©¥       ¦   ¤¦ ¤ © ¥©    ˆ ıd + + max − + min Idref d-axe comp.e. Figure 3(b) shows that the id component of the . Thus. with kµ = kd = (ωn Tr ) . Examples of Transients Figures 3(a) and (b) show transients which prove that this approach gives a complete inversion of the dynamical model. 3. 2. and additive compensators like vd2 . the magnetization. kd . with kq = . we want to have a control structure which contains internal loops on the stator currents. Tr ıµr τc (20) (b) On the d-axis we control the rotor flux ψr = M iµr and the magnitude of the component id of the stator current: first. 2. Tr We observe that this structure makes use of proportional controllers like k q . as indicated in Fig.. (22) + + v d2 vd kd v d1 Fig. Eqn.3. i. vq2 and Idref 2 = iµr . vd = kd (Idref − id ) + ed . Delaleau et al. (20) Estimator Eqn. v q1 kq q -axe comp. 2. Figure 3(a) shows three responses: First. 2ξωn Tr − 1 2 (21) (22) with N1 (2ξωn Tr − 1) . the response of the stator flux to a step reference (dynamics of the second order.

Modeling and control of induction motors 113 1.000325 1.5 0 0 0.6 0.4 1. 3.4 1.8 1 td n 1.8 1 td n 1.6 1. (a) Flux.4 0.898831 (b) Fig.5 2 i d1 n I d_max i q1 n I q_max I d_sat I d_max I q_sat I q_max 1 0 1 1.2 0. .5 C max Ω n Ω maxi ψ r n M .4 0.2 1.6 1.5 1 1.2 1.6 0. torque and speed responses.8 2 1.898831 (a) 1. (b) Responses of id and iq .001511 1.8 2 1.2 0.5 1 C n 0.6 0 0 0.I µ _ref C ref n C max 0 0.

Thus.114 E. Model In Section 2. Passivity-Based Control 4. N= I2 0 . 4. we neglect the effect of . (24) with Rs . but without overshoot: this is a protection effect due to the internal loop.3 we have derived the standard two phase αβ -model of an np pole pair squirrel cage induction motor with uniform air-gap. respectively. given by (19a).1.1. For simplicity. Control Properties of the Induction Motor Model In this section we establish some properties of the model (input-output and geometrical) that will be instrumental for further developments. Particularly useful for further developments is the following relationship between rotor fluxes and rotor currents: ˙ r + R r ir = 0 ψ The model is completed by computing the electromagnetic torque as 1 t ∂L(θ) 1 ∂L−1 (θ) i i = − ψt ψ 2 ∂θ 2 ∂θ and replacing it in the mechanical dynamics τem = ¨ = τem − τL . (23) L(θ) = Lt (θ) > 0 being the 4 × 4 inductance matrix of the windings. The inversion of the model is completed. Jθ (26) (25) (27) where J > 0 is the inertia of the rotor and τL signifies a term of load torque which we will assume to be constant but unknown. Rr > 0 being the stator and rotor resistances. 4. Delaleau et al. The electrical dynamics are defined by the voltage balance equation ˙ + Ri = N u. After the magnetization we see the response of the torque to the steps reference: the dynamics is of the first order. For convenience in the sequel. iαβr ] . L(θ) = Ls I2 M e−J np θ M e J np θ Lr I2 .1. set t t t t ψ = [ψαβs . ψαβr ]t and i = [it αβs . where we have defined the (skew-symmetric) rotation matrices J = −J t = P (π/2) and eJ np θ = P (np θ) (see (4)). ψ R= R s I2 0 0 R r I2 . stator current reaches a maximum value (denoted by idsat ). The speed ω has quasi-linear responses (the torque is well controlled). We observe that the decoupling between the two axes is perfect: the magnitude of the flux remains constant when the torque has large variations. the relation between the flux and the currents has the form ψ = L(θ)i.

1. the motor model can be decomposed as a feedback interconnection of two passive operators with storage functions He and Hm . which for the induction motor is given as ˙2 . 1994) the ab-model was used. as shown by Espinosa and Ortega (1994). Finally. as shown in Ortega et al. while the developments of Ortega and Espinosa (1993) and Ortega et al. the work of Nicklasson et al.3. For instance.1. its workless forces. and their corresponding storage functions. ˙ θ) = 1 ψ t L−1 (θ)ψ + 1 J θ H(ψ. 1993a) relied on the dq -model. θ.. −τL ]t → [it s . Geometric Properties We now exhibit an invertibility property of the induction motor model which is essential for obtaining an explicit expression of the PBC.e. 4. θ ] is passive. (1998). (1997) was carried-out in the original αβ frame. From (26) and (23). with storage function H. respectively. it can be derived in any reference frame chosen for model representation. Remark.. We have neglected the capacitive effects in the windings of the motor and considered a rigid shaft.Modeling and control of induction motors 115 friction but. as a by-product. From the integration of the equation above we obtain the energy balance t t H(t) − H(0) = − stored energy it (s) R i(s) ds + 0 dissipated 0 ˙ it s (s)u(s) − θ (s)τL ds supplied−extacted (28) Furthermore. 1998). we see that the torque can be written as J np θ ir . 2 2 He Hm ˙) denote the electrical energy and the mechanical kinetic where He (i. 4. (Ortega et al. This is easily established from the system total energy. These passivity properties. Input-Output Properties The cornerstone of the passivity-based design philosophy is to reveal the passivity property of the system and identify. in (Espinosa and Ortega. We established in (Espinosa and Ortega. The rate of change in the energy (the system work) is given by t ˙ ˙ = it H s u − θτL − is Ris .2. it can be easily accommodated into our analysis. respectively. are the basis for two different PBCs studied in (Ortega et al. θ) and Hm (θ co-energy.. hence the potential energy of the motor is zero. (29) . τem = np M it sJ e ˙t which proves that the mapping [ut . 1995) that PBC is coordinate independent. i.

1997). ∀ x ∈ 2 ) have been used. 2 em (33) From this equation we conclude that if τem and ψr are fixed to constant values. the rotor flux rotates at a constant speed. The two equations above will be instrumental in the next section for the derivation of the PBC. This fact becomes clearer if we evaluate the angular speed of the rotor flux vector with respect to the rotor fixed frame (the slip speed ) as ρ ˙ = = ˙ r 2 ψr 1 − ψ r 2 ψ ˙ r1 d 1 ψ arctan(ψr2 /ψr1 ) = 2 dt 1 + (ψr2 /ψr1 )2 ψr 1 1 ψr 2 ˙ t J ψr = ψ r Rr np ψ r τ . Equation (32) also shows that the zero dynamics of the motor with outputs τem and ψr are periodic. where the fact that J and eJ np θ commute (J eJ np θ = eJ np θ J ). explicitly solve this equation as ˙ r = τem Rr J ψr . Now. that is. solving (25) for ir and substituting it into (29) gives   τem = np M t J np θ i Je ψr . The underlying fundamental assumption for the machine to be Blondel-Park transformable is that the windings are sinusoidally distributed. This is a key expression that allows us to invert the systems dynamics. this means that the magnetomotive force can be suitably approximated with the first harmonic in a Fourier approximation. For a practical machine. giving a sinusoidal air-gap magnetomotive force and sinusoidally varying elements in the inductance matrix L(θ). This expression also shows that the torque can be controlled by controlling the rotor flux norm and slip speed. Delaleau et al.116 E. we assume that the machine is Blondel-Park transformable to ensure this invertibility property. For this class of machines the application of PBC is still an open issue. as is well-known in the drives community. Rr r and then substituted into (30) to give τem = (31) where (25) has been used again. where we study the model of the generalized rotating machine. In (Nicklasson et al. and that the skewsymmetry of J (J t = −J → xt J x = 0. ψ ψ r np (32) where · is the Euclidean norm.. Lr s (30) Finally. (23) can be solved for is as is = 1 J np θ e (ψr − Lr ir ) M np ˙ t ψ J ψr . . Examples of machines in which higher-order harmonics must be taken into account are square-wave brushless DC motors and machines with significant saliency in the air gap.

If. instead. There are at least three motivations for this approach: firstly. of course. more direct form. 1998) for a robotics problem. Although for both controllers we can prove global asymptotic speed/position tracking. Namely. while preserving the global stabilization property. we use the notation (·)d .) A very interesting property of the resulting scheme. Secondly. (In simple applications. and Col is an outer-loop speed controller which generates the desired torque5 τem d . To overcome this obstacle. and replace the acceleration by its approximate differentiation. there is typically a time-scale separation between the electrical and the mechanical dynamics. since the nested-loop configuration is the prevailing structure in practical applications. we denote it by (·)∗ . Col is just a PI around speed error. it is generated by the controller.. where Cil is an inner-loop torque tracking PBC. 1998 that for electromechanical systems the PBC approach can be applied in at least two different ways. cascaded) scheme. Finally. The so-designed controller will be called the nested-loop PBC. we can in some important cases establish a clear connection between our PBC and current practice. using this feedback-decomposition leads to simpler controllers which in general do not require observers. We go here through these additional complications to provide a complete proof of stability. Controller Structure In this section we solve the speed-position tracking problem adopting a nested-loop (i. Another route stems from the application of a passive subsystems decomposition to the electromechanical system. a PBC is designed for the whole electromechanical system using as the storage function the total energy of the whole system. we show that (under some reasonable assumptions) we can decompose the system into its electrical and mechanical dynamics.” We then design a PBC for the electrical subsystem using as the storage function only the electrical part of the system’s total energy.. and this in turn implies measurement of acceleration. 4.. where the latter can be treated as a “passive disturbance. which is further elaborated below. In the first. but here is a simple pole-placement) is then added to regulate the mechanical dynamics. leading to different controllers. the controller exactly reduces to the well-known 5 We will adopt throughout the following notation convention.e.2. This is the way PBCs are typically defined for mechanical and electrical systems. Nested-Loop Passivity-Based Control 117 It is shown in Ortega et al. We will show in this section that Col may be taken as an LTI system that asymptotically stabilizes the mechanical dynamics.Modeling and control of induction motors 4. is that if the inverter can be modeled as a current source and the desired speed and rotor flux norm are constant. An outer-loop controller (which can also be a PBC. If a signal is explicitly given as an external reference.2. . The main technical obstacle for its design stems from the fact that Cil requires the knowledge of τem ˙ d . for the sake of brevity we present here only the torque tracking version of the nested-loop PBC. and it is usually referred to as PBC with total energy shaping.1. we proceed as in (Ortega et al.

ψ τem = np ˙ t ψ J ψr . 0 ]. 4. After differentiation we get ψ be replaced in the first two equations of (36) to get6 ˙ sd + u=ψ 6 I2 0 Rs L−1 (θ)ψd . hence providing a solid theoretical foundation to this popular control strategy. In this subsection we derive a torque tracking PBC from the perspective of a system’s inversion. 2 (t) em ∗ n p β∗ ρd (0) = 0. the PBC is a “copy” of the electrical dynamics of the motor (34). Rr rd (36) (37) defines the desired values t t where τem ∗ is the torque reference and ψd = [ψsd . and refer the reader to Ortega et al. The where ψrd (0) = [β∗ (0). Torque Tracking PBC Implicit and explicit forms. we have repeated (31). Typically. Rr r (34) (35) where. (38) ρ ˙d = Rr τ . (35) with an additional damping injection term that improves the transient performance. To simplify the presentation. we define the PBC in an implicit form as ˙ d + RL−1 (θ)ψd . and β∗ (t) is a (time-varying) reference for last equation can actually be solved as ψrd = eJ ρd β∗ (t) 0 .118 E. we will omit the damping injection here. using (23).2.2. for ease of reference. An explicit realization of the PBC above is obtained by “inversion” of (37) as ˙ rd = ψ 1 2 (t) β∗ Rr ˙ ∗ (t)β∗ (t)I2 ψrd . we rewrite (24) and (26) as ˙ + RL−1 (θ)ψ = N u. . indirect field-oriented control. (39) ˙ rd in The description of the controller is completed by the replacement of ψrd and ψ ˙ sd which can the last two equations of (36) to get ψsd . For that purpose. Thus. An explicit state space description is given in Proposition 1. τem ∗ J + β np ψr . Nu = ψ τem ∗ = np ˙ t ψ J ψrd .. 1998. Delaleau et al. ψrd ] for the fluxes.

the signal τem ∗ will now be generated by an outer-loop controller Col . ˙ . ψ ˜ We have shown above that ψ rd ˙ rd is is bounded by construction.3. for some α > 0. The proposition below shows that Col can be taken as a linear filter. ψ ˙ → 0. Col must be chosen with care and a new argument should be invoked to complete the proof. let us turn our attention to the torque tracking error τ ˜em = τem − τem∗ . 1996a) to give an “implicit observer” interpretation of the PBC controller. ψ ˜ = ψ − ψd are the flux errors. Stability. since it will be generated by Col . Global convergence can be easily established where ψ by considering the storage function7 Hψ = 1 ˜t −1 ˜ ψ R ψ≥0 2 whose derivative satisfies ˜t L−1 (θ)ψ ˜ ≤ −αHψ ˙ ψ = −ψ H To illustrate the second difficulty in the stability analysis of the nested-loop scheme. .Modeling and control of induction motors 119 The expression above causes a difficulty in the implementation of the nested-loop ˙ sd . Hence. Speed Tracking PBC Main result. whose proof can be found in Ortega et al.2. which in turn depends on τ scheme. The error equation for the fluxes is obtained from (34) and (36) as ˙ + RL−1 (θ)ψ ˜ ˜ = 0. ψ ˜ → 0 (exp. from (35) and (37) we get τ ˜em = np Rr ˙t ˜ ˙t ˜ ˜ ˙ ˜ ψ r J ψr + ψr J ψrd + ψrd J ψr .. In position-speed control. A globally stable speed tracking PBC is presented in the proposition below. Proposition 1.). Therefore. We will see in Proposition 1 how to overcome this which will generally depend on θ obstacle with the use of a linear filter. 4. ˜ → 0 exponentially fast. and consequently. On the other hand. 1998. see (38). Also. as the control law depends on ψ ˙em∗ . τem ∗ is not a priori bounded. After some simple operations. The nonlinear dynamic output feedback nested-loop controller ˙ rd +Rs isd ˙ sd + M eJ np θ i ˙ rd + np M J eJ np θ θi u = Ls i ˙ sd ψ 7 (40) This function was used in (Martin and Rouchon. we cannot prove that ψ bounded. Unfortunately. unless τem ∗ is bounded. Let us now analyze the stability of the closed loop.

when placed in the closed loop with (23) and (24). ˜ lim θ t→∞ lim | ψr − β∗ (t)| = 0 for all initial conditions and with all internal signals uniformly bounded. That is. We can extend the result given in Proposition 1 to the case of unknown but linearly parameterized load. (42) (43) ˙ ˜ z ˙ = −az + bθ ˙ ˜ ˙−θ ˙∗ and a.120 with 1  M  id =    ˜ i= ˜ is ˜ ir =  ˙∗ Lr β 1+ R r β∗ − E. provides a solution to the speed and rotor flux norm with θ =θ tracking problem. It is easy to see that choosing the desired torque in the controller above as ¨∗ − z − f θ ˜ + τL τem d = J θ (44) yields global asymptotic position tracking for all positive values of a.   . (41) is − isd ir − ird ¨ ∗ −z + τL .  ˙∗ τem d β  J + I ψ 2 rd 2 n p β∗ R r β∗ . b > 0. θ measurable regressor. eqns. τL = η t φ(θ. ˙ ).   Adaptation of load torque. where  Lr J np θ I2 + τ J e ψrd  2 em d n p β∗  . τem d = J θ and controller state equations ˙ rd = ψ ˙∗ Rr β τem d J + I2 2 n p β∗ β∗ ψrd . Extensions. Delaleau et al. (26) and (27) ensure t→∞ ˙ = 0. b. f . θ ˙) is a where η ∈ q is a vector of unknown constant parameters and φ(θ. The proof of global asymptotic rotor flux norm and position tracking follows verbatim from the proof of the main result above.   Position control.

indeed. Structural Properties of the Model 5. i = i ˜ ˜ We thus have ψ r = ψ and is = i . Therefore. we mark the variable with a tilde when it is given in the rotating frame. we prefer to work with a complex8 model instead of the real one given by (7)–(9). r re se r s s In this notation the expression for the electromagnetic torque (8) becomes τem = 8 np M m is ψ ∗ . (46b) (47a) r ˜ + Rr i (47b) r = 0. r Lr (48) We denote by  the pure imaginary number satisfying 2 = −1. m (x).Modeling and control of induction motors   121 Integral action in stator currents. 1998 shows that. Flatness-Based Control 5. the imaginary part and the conjugate of a complex quantity x are respectively denoted by e (x).1. The advantage of considering complex variables will clearly appear in the sequel.1. The real part. ψ = ψ ˜ enp θ . 5. otherwise. It is interesting to note that the global tracking result above still holds in this case. The experimental evidence presented Ortega et al. Complex Form of the Model For simplicity. np θ np θ ˜ enp θ . To simplify the proof of flatness. eqns. It is common in applications to add an integral loop around the stator current errors to the input voltages. ψ = L i + M enp θ i (46a) s s s r ˜ = M e−np θ i + Lr i ˜ ψ s r. and changes of frames are simply accomplished by multiplying complex variables by an appropriate complex exponential. r Equations (9) become dψ s dt ˜ dψ + Rs is = us . it is referenced in the fixed frame which is the natural frame for considering the variable of the stator.1.. To this end. It reduces the number of equations. In order to distinguish the value of a given variable between being referenced in the fixed frame and in the rotating frame. it is useful to consider some variables in the frame rotating at the speed np ω which is the natural frame to consider variables of the rotor. this robustifies the PBC by compensating for unmodeled dynamics. dt respectively. (45) where we set • = s or • = r depending on whether one considers a stator or a rotor variable. for any variable x introduce the notation x• = xα• + xβ • . . and x∗ . (7a) and (7b) take on the forms ˜. respectively.

u. . Flatness of the Model E. 1 ˜ 2 Lr d ˜ ˜ ∗ ˜∗ ˜ is ψ r ∗ = i (|ψ r | + (ψ ) ψ ).1. . Delaleau et al. .. x = B (y. ψ r . ˙ . this flat output has a physical meaning which will simplify the control design: θ (or its first derivative ω ) is the variable to be controlled. y (r) ) and u = C (y. 3. . ˙ − np ω is the slip speed. 1996b). ˙ .. τem = np ρ2 α/R ˙ r . ). u. ˙ . the components of y are differentially independent. . . So the mechanical equation of the induction motor becomes ω ˙ = np 2 f 1 ρ α ˙ − ω − τL . This leads to a huge collection of industrial applications (Fliess et al. y. y. s ψr = M Rr dt r r and finally. y = A(x. . . . s = M r Rr dt r Thus. It is useful to express the electromagnetic We thus have ψ r r torque produced by the motor in terms of ρ and α: using (46b) and (47b) leads to 1 ˜ Lr d ˜ ˜ i (ψ + (ψ )). . but not on other variables (is . . u). usually only defined on constant speed Notice that α ˙ = δ operations. ˜ = ψ e−np θ = ρeα . 2000) for flatness-based control regarding other kinds of motors. A flat output of the induction motor is y = (θ. therefore δ is the angle of the rotor flux with respect to a fixed frame. 1995) for an introduction to this subject. 1999) and (Hagenmeyer et al. . 1995). See (Boichot et al. the flat output has a clear physical meaning with respect to the control objective.122 5. is (differentially) flat if there exists a set of variables y = (y1 . . JRr J J (49) Hypothesis: The torque load is an unknown function of time which can possibly depend on θ or its derivatives. The concept of (differential) flatness was introduced in 1992 and we refer to (Fliess et al. The flatness of the model of the induction motor was established in (Martin and Rouchon. where x is the n-dimentional state and u the m-dimentional input. . We recall the proof in the present notation. As usual. α). ir . y (r+1) ) for an appropriate integer r. and α is the angle of the rotor flux with respect to a frame rotating at speed np ω (recall that np ω is called the synchronous speed). Recall that a (nonlinear) control system x ˙ = f (x. A set of variables y with these properties is called the flat output. Set α = δ − np θ. u(q) ) for an appropriate integer q . Set ρ = |ψ r | and define δ as the angle such that ψ r = ρeδ . . 2.. A strong interest in flatness stems from the fact that it allows a straightforward solution to the motion planning problem: in practice. .2.. ym ) such that: 1. .

α. α. Tr Tr s (51a) (51b) (51c) ρ ˙ + αρ ˙ = − d dq M 1 ˙ i dq + 1 u dq . α(3) . . τ˙ L ). θ(3) . dt s (50e) (50f) (50g) (50h) (50d) Accordingly. = a(θ. . . L ). Stationary Operation The most useful frame to study stationary operations of the motor is certainly the frame of the flux. . ˜ = ρeα = b(θ. τ˙ i (ψ ) = c(θ. . . . r = d(θ.e. ¨ α. we obtain 1 d ˜ ¨ θ(3) . xdq = xe−δ = x ˜e−α . i.1. . α ¨ . δ = np θ + α. ρ satisfies ρ= ¨ + fθ ˙ + τL ) Rr (J θ ˙ θ. α ¨ . τL . . JRr J J 1 M dq ρ+ i . usually called the dq -frame. . . τL . Consequently. We denote by xdq the value of the variable x in this frame. From (49). . Then the state-variable complex model reads as follows: ω ˙ = np 2 f 1 ρ α ˙ − ω − τL . ˙ θ. α. . . . r = − Rr dt r enp θ ˜ (3) ˜ ψ r − Lr i . (i ) = ( − np ω )ψ dq − a + δ s r dt s σLs Lr Tr σLs s with a= 1 σLs Rs + M 2 Rr L2 r . τ˙ L ). τL . ˙ τL ) np α ˙ 123 (50a) (50b) (50c) which is a function of the flat output and its two first derivatives. . is = us = Rs is + d (ψ ) = f (θ. . θ(4) . . . α) is a flat output of the induction motor.Modeling and control of induction motors It is obvious that ˙ ω = θ. α. ..3. . . . θ M ψ s = Ls is + M enp θ ir = e(θ. . . r Continuing the calculations using successively (46b). ψ ˙ τL ). 5. ˜ ˙ α ¨ . τL . . . α. . τ¨ L ). . ¨ α. and (47a). y = (θ.

this variable gives a degree of freedom in order to perform a complementary control task. So. Tr ρ A stationary operation at constant speed ωo with constant load τL = τL o is obtained when θ = ωo t + θo . For the second component α of the flat output. using (51b). at least when τL = 0. the induction motor has to be fed by sinusoidal voltages.. In conclusion.e. these expressions allow us to calculate the control us .and right-differentiable and t → α is everywhere 3-times left. Delaleau et al..124 E. we have d 1 M (ψ ) = (− + np ω )ψ r + i . the induction motor can only follow trajectories such that t → θ is everywhere 4-times left. we are able to make the reverse analysis.e. and finally. As θ(4) and α(3) appear in the expression (50h) of the control us .2. (51a) implies that ρ is constant ρ = ρo and thus. For the first component θ of the flat output. where α1 and αo are constant (i. for a known mean value τLo of τL or for an estimated value τL . tf ]. 5. The choice of the reference trajectories of θ and α is made in order to fulfill the constraints on all system variables. the ˙ = np ωo + α1 = δ1 is constant. with the flatness properties. In particular. However. Trajectory Generation By (50a)–(50h) we obtained the expressions for all system variables in terms of the flat output components and the disturbance τL . dt r Tr Tr s . is and us are periodical functions of time with pulsation δ1 = np ωo + α1 . This corresponds to a known function of time t → θd on a given time interval [ti . r dq dq In turn. i. slip speed is constant). Notice that usually the stationary operation is analyzed by imposing us to be sinusoidal under constant load and deducing that all electric and magnetic quantities are periodic and finally. For example. so is us dq = us dq o . with (51c). (51b) can be splitted into ρ ˙=− 1 M ρ+ e is dq Tr Tr and α ˙ = M m is dq . α = α1 t + αo .and right-differentiable. ψ r . beginning with the variable to be controlled and deducing the control. (51b) implies that is = is o is constant. In this case δ As a consequence. eqn. Notice that as ρ and α are real variables. it is possible to minimize the copper losses in the stator at every constant speed with an appropriate choice of the value of α ˙ : We get ψ r = ρeδ = ρe(np θ+α) . the choice of the desired trajectory t → αd is not so obvious because the value of α does not correspond to a clear control objective. ψ dq = ρo . Here. that the speed is constant. the trajectory is often designed with respect to the control objective. To run at a constant speed with constant load.

Between two time intervals on which ω is constant.3. δψ (53a) s s s ˙ − np ω )ψ + Rr i = 0. −np θ ˜ = ψ e−np θ . is and ir remain bounded everywhere). Lr (54a) (54b) (53b) ψ s = L s is + 9 This is an important industrial problem as mentioned by Bartos (1998). For example. dt r which is the expression of the electrical equation of the rotor in the fixed frame. . (δ r r respectively. |is |2 = Tr 2 Rr (f ωo + τL o ) np M2 1 + α 1 Tr . α 1 Tr Tr ρo M Tr 2 2 ρo M2 1 + α1 e(np θ+α) . (47a) and (52) are ˙ + Rs i = u . Stabilization around Desired Trajectories In this section. Lr M ψ r − Mis . i ˜ Coupling ψ and (47b). ρ = ρo and α ˙ = α1 are both constant and therefore is = |is |2 = Thus. it is possible to write r = ir e r r d (ψ e−np θ ) + Rr ir enp θ = 0. 1996 for a detailed planning of the reference trajectories of θ and α in order to start the motor from rest to a nominal speed without any singularity9 (us . Tr 2 The magnitude of is is minimum if α1 = 1/Tr (see (Chelouah et al. dt r d (52) (ψ ) − np ωψ r + Rr ir = 0. We use a singular perturbation approach due to the good separation of the time scales. As ψ s = Ls is + M ir and ψ r = M is + Lr ir . 5. Tr 1 + α2 1 . 1996) for more details). we have ir = 1 ψ r − M is . dt r d (ψ )e−np θ − np ωψ r enp θ + Rr ir enp θ = 0. we refer to Chelouah et al.. t → αd can be chosen as a function of ω . we present a tracking feedback law which is designed by studying the stationary operation of the system..Modeling and control of induction motors 125 At a constant speed ω = ωo . The stationary modes of eqns.

Experimental Results We conclude with the presentation of some experimental results10 of a flatness-based control scheme. pp. + (δ s Lr (55a) (55b) ˙ ) = MRr /Lr Zr (ω.126 Equations (53) and (54) lead to ˙ s i + δ ˙ Rs + σ δL s −M Thus ˙) u . δ ˙ d) np M m Z r Rr i + Lr s M ψ = us . References Bartos F. ψ r = Zr (ω. Delaleau et al. . + (δ Lr Lr Ls ω ˙ d − κ(ω − ω d ) . We observe a good tracking for the acceleration motion (the experimental trajectory is hardly distinguishable from the reference one) and a small overshoot. (1998): Motor starting and protection moves closer to the process. δ Z Rr ˙ − n p ω ) + M R r δ ˙M . ˙d is the reference trajectory of the angular speed and δ d = αd − np ω d where ω d = θ is the reference trajectory of the slip speed.. the control law is given by |us |2 = JLr ∗ (ω. δ s ˙) u .104–109. France. 4(a)) consists in starting the motor from rest to its nominal speed. This control does not necessitate a flux observer. Figure 4(b) shows the braking of the motor from its nominal speed to rest. Zs (ω. — IEEE Contr.J. 5. is = Zs (ω. Rr ˙ − np ω ) ψ = 0. Ls r E. Nantes. Mag. IRCyN. Syst. δ ˙ d ) Zs (ω. The first experiment (Fig. δ Z and ˙ s Z = Rs + σ δL Finally. 10 Implemented on the experimental setup of the GDR Automatique. δ s where ˙ ˙ ) = Rr /Lr + (δ − np ω ) .4.

Vol... Automat. pp. (1989): Design of an exact nonlinear controller for induction motors.. 4. (1972): The principle of field orientation as applied to the new transvector closed loop control system for rotating field machines. pp. — IEEE Trans. China. (1995): A new approach to dynamic feedback linearization control of an induction motor.42.39. De Fornel B. — Th` ese de Doctorat.5. pp.80–85.76 s (b). (1994): High-performance induction motor control via input-output linearization. 34th IEEE Conf.. New Orleans. Vol. Electron. (1995): Mod´ elisation et commande de la machine asynchrone. Chiasson J.25– 33. L´ evine J. — Proc. 1-st CESA Conf. — Prep. No. Caron J.4. — Siemens Rev. Mechanical speed w [rad/s] (solid line) and the reference trajectory (dash-dot line) for T5% = 0. Chiasson J. France.2173–2178. (1990): Alimentation des machines asynchrones. and Novotnak R.. Chatelain J.. pp. Chiasson J.P. (1983): Machines ´ electriques.4.1–7.. — Georgi: Dunod. — IEEE Trans. (1995): Contribution a ` la commande des entraˆ ınements asynchrones de forte puissance: application au probl` eme de traction. and Novotnak R. — Paris: Technip.84 s (a) and T5% = 0. De Luca A. No. (1999): Modelling and control of a high-precision positionning system actuated by a linear synchronous motor.217– 220. Techn.34. (1993): Dynamic feedback linearization of the induction motor. pp. Contr. . D´ el´ emontey B. LA. Delaleau E. Institut National Polytechnique de Lorraine.. Bodson M. Lille.. Chiasson J.1588–1594. and Parzy E. and Rouchon. Martin P. Beijing.10. Diep. Contr. (1995): Nonlinear speed observer for highperformance induction motor control.. Syst. and Hautier J. (1996): Differential flatness and control of induction motors.P.V.14. Vol.1304–1307.. Delaleau E. Boichot J. Bodson M. Vol. 14th IFAC World Congress. N. pp. No.38..Modeling and control of induction motors 160 160 140 120 100 100 80 80 60 60 40 40 20 0 −20 0 127 140 120 20 0 0 5 Time (s) 10 15 2 4 6 Time (s) 8 10 12 (a) (b) Fig. No. — IEEE Trans. — Proc. — IEEE Trans. Contr. Automat. pp. Blaschke F. Chelouah A. — Techniques de l’ing´ enieur.. France. P. Decision and Control. Vol. Indust.

Vol. No.P.315–323. — IEEE Trans. (1996a): A simple flux observer for induction motors.1 (A. Etude et mise en œuvre de commandes num´ eriques a ` flux orient´ e et adaptatives. — Berlin: Springer. pp.D3644. Marino R.). pp. and Ortega R. pp. Leonhard W. filtrage ´ etendu du vecteur d’´ etat. (1996): Output feedback control of current-fed induction motors with unknown rotor resistance. France. and Louis J.. (1995): Conception. Vol. Automat. — Th` ese de Doctorat. Yala S..J. Delaleau et al.138–143. (1997): Passivity-based control of a class of blondel-park transformable machines. Phys. and Rouchon P. Espinosa G.. pp. and Rouchon P. Institut National Polytechnique de Toulouse. Paris.76–79. (1991): Positionnement d’une machine asynchrone par la m´ ethode du flux orient´ e.5. J. — Berlin: Springer. Contr. Fliess M. Louis J.. and Espinosa G. (1999): Commande num´ erique des machines synchrones. J. — Th` ese de Dectorat.4. (1996b): Two remarks on induction motors. Martin P. Jacquot B. Vol. Espinosa G.2. 2nd Ed. — Colloquium Th´ eorie et Applications des Observateurs Non lin´ eaires.439–451. Lille. (1993): Contribution a ` l’´ etude de servomoteurs a ` induction. Phys. No. Universit´ e Paris 6. — IEEE Trans. and Delaleau E. Contr. Martin P.. pp. (1998): Adaptive output feedback control of currentfed induction motors with uncertain rotor resistance and load torque. Lamnabhi-Laguarrigue and W. Vol. Institut National Polytechnique de Toulouse. (1994): State observers are unnecessary for induction motor control. In: Nonlinear Control in the Year 2000. and Tomei P.. Marino R. Jelassi K. Syst. Ortega R. Peresada S. L´ evine J.4. Grellet G. Peresada S. Peresada S.. — IEEE Trans. ´ etude de la num´ erisation de la commande. — Automatica.. Automat. — Techniques de l’ing´ enieur. France. (1993): Adaptive input-output linearizing control of induction motors. Hagenmeyer V... Vol. (1998): D´ efinition d’extensions de la transformation de park pour la commande de moteurs synchrones a ` aimants permanents mont´ es en surface. and Clerc G. (1994): Commande non lin´ eaire d’une machine asynchrone. and Ortega R.38. Contr. Contr. — Europ. Eds.42. Vol. France. Respondek. Graf von Westerholt E. F. pp. Appl.. Institut National Polytechnique de Toulouse. ´ Mendes E. and Rouchon P. mod` eles. Martin P. Contr. Vol. ´ etude et r´ ealisation des algorithmes de commande des syst` emes de traction asynchrone pour les TGV de nouvelle g´ en´ eration. No. Automat. Isidori. Contr. — IEEE Trans. Vol. No. commande. France.23..629–647. (1995): An output feedback globally stable controller for induction motors.. Nicklasson P.61. 1-st CESA Conf.1. (1995): Flatness and defect of non-linear systems: introductory theory and examples. France. — Proc. Vol. No. . and Bergmann C.208–221. (1997): Actionneurs ´ electriques. Grenier D. Vol. pp. and Tomei P.617–624. pp. sensibilit´ e aux variations param´ etriques. — Paris: Eyrolles. Techn. — Th` ese de Doctorat.. (1997): Control of Electrical Drives. principes.336–347. — Syst.40. and Valigi P.. — Th` ese de Doctorat.. contrˆ ole de la vitesse sans capteur m´ ecanique. pp.1-22. Marino R. — Int..34. Kohlrausch P. (2000): Flatness based control of the separately excited DC drive. Lett.5.1327–1361.128 E.5.P. pp.

(1993b): Nonlinear control of induction motors: Torque tracking with unknown load disturbance. and Sira-Ram` ırez H. — Oxford: Oxford Science Publications.. (1999): Computation of induction machine inductances for extended analytical modeling accounting for saturation. Contr.. pp. Appl..1675–1680. (1994): Nonlinear control of electric machines: An overview. — Europ..5. Phys. Vol. France.14. (1993): Commandes num´ eriques ´ evolu´ ees d’actionneurs a ` courants alternatifs. Contr. — IEEE Trans. — IEEE Trans. .. — Automatica. Automat.1675–1679. pp. Mag.J. Semail B. J.38. Vol. No. Canudas de Witt C.3. pp. — IEEE Contr. Ortega R. (1990): Vector Control of AC Machines. No. Phys. Vol.. Ortega R. (1993): Torque regulation for induction motors.29.G.257–267.38. Vol.P. Loria A. pp. Automat. and Seleme S.. Taylor D. Syst. pp.11. Canudas C. Vas P. Nicklasson P.11. and Seleme S. Robyns B. Universit´ e Catholique de Louvain-la-Neuve.. — Th` ese de Doctorat. — Berlin: Springer-Verlag.. Vol. Ortega R. and Bouillault F..I.Modeling and control of induction motors 129 Ortega R.621–633. (1998): Passivity-based control of Euler-Lagrange Systems.41–51. Louis J. No. (1993a): Nonlinear control of induction motors: Torque tracking with unknown load disturbances. and Espinosa G.