II-9. Diagnosis of Induction Machines 243 2β r flux sensor ls L Ls Figure 2. Measured flux. If thesensor hasn turns and it has a rectangular shape of S area, and considering that l r, then the h,k flux components regarding to the external flux density ones, are expressed by: h,k (r) = nS sin(hpβ) hpβ ˆ bn a h,k (r) cos kωt (31) Then,toobtainthecomponentatthekω angular frequency,it is necessarytoaddthedifferent terms which have this same frequency. k (r) = +∞ h=−∞ h,k (r) (32) The terms which compose k (r) are very numerous but many do not contribute significantly. Actually,relation(30)showsthatonlythetermswithasmallvalueofh willhaveaprominent effect. This property is enhanced by the quantity sin(hpβ) hpβ which appears in expression (31). The data given by the coil is a voltage which results of the Lenz law. Consequently, the e k (r) components of the corresponding measured e.m.f. are obtained by the following relation: e k (r) =− d k (r) dt e k (r) =−kωnS +∞ h=−∞ sin(hpβ) hpβ K h (r) ˆ b g h,k sin kωt (33) It can be noticed that the electro-motive force spectra amplify the high frequencies, what permits to distinguish the slot spectral lines more easily. 244 Thailly et al. Figure 3. Experimental device. Principle of measurements The synoptic of the Fig. 3 shows the experimental device that allows performing the mea- surement of the radial magnetic field component. The principle is to pick up the sensor signal with an analyzer via an acquisition card. A small winded flux sensor, with nS = 30 × 10 −3 , has been realized. It is located at a distance r equals to 100 × 10 −3 m. The tests are made consideringa star connected4 poles, 4kW,380/660 V, 50 Hz squirrel- cage induction machine with 36 stator slots and 44 rotor bars. For this machine, Rs int = 60 × 10 −3 m, Rs ext = 90 × 10 −3 m. The stator is supplied by 380 r.m.s. phase to phase voltage. The supply current is equal to I =1.5 A. In order to neglectthemagneticeffects generated bytherotor,theslipiscloseto0. Results In the measured spectrum of the e.m.f. only the spectral lines corresponding to kr = 0 (fundamental) and kr =±1 will be considered for the comparison with the theory. Concerning the fundamental, it is measured 5 × 10 −3 V (r.m.s. value). As the machine is underfed, it will be supposed that the maximum of the flux density in the air-gap is 0.6 T. In that case, assuming that the flux density component at 50 Hz is mainly composed of one p pair pole wave (h = 1, ˆ b g 1,1 = 0.5 T), μ r have to be equal to 600. This low value can be justified taking into account that the frame is of cast iron and that the relative permeability of this material is about 300. For the harmonics, Table 1 gives the relative magnitudes of the frequencies of the couple at [1 ± N r (1 − s)] f regarding to their fundamental. Table 1. Relative magnitudes of the frequencies of the couple at [1 ± N r (1 −s)] f regarding to the fundamental [1 − N r (1 − s)] f [1 + N r (1 − s)] f Calculation without K h correction coefficient −5.49 db −5.93 db Calculation with K h correction coefficient −31.72 db −22.31 db Calculation with K h and with prominent components −31 db −22.7bd Practical spectral lines −27.06 db −18.84 db II-9. Diagnosis of Induction Machines 245 –60 1000 1050 [1 – N r (1-s)]f [1 + N r (1-s)]f 1100 1150 1200 Practical values Theoretical values with K n correction coefficient Theoretical values without K n correction coefficient f(Hz) –40 –20 0 db Figure 4. Comparison between practical and theoretical toothing spectral lines in relative values regarding to the fundamental. The two first table lines correspond to computation results where kr and ks vary from −5to5andhs varies from 1 to 13. The third table line gives the computation results where only the prominent components depending on their number of pole pairs are considered. For the spectral line at [1 − N r (1 − s)] f frequency (kr =−1), the lowest number of pole pairs is obtained for hs = 1 and ks = 1; in that case, equation (8) gives |h|p = 2. For this at [1 − N r (1 − s)] f frequency (kr =+1), ks =−1 and hs =−5 leads to con- sider a prominent component such |h|p = 6. The latest table line gives the relative magnitudes of this couple of frequencies which are obtained by experimentation. The results show that considering only the prominent component which composes one spectral line is sufficient in the study of the external magnetic field. Fig. 4 gives a graphic illustration of the results. It can be observed that when only the analyticalexpressionofthe electro-motiveforceis computed,withouttake the K h correction coefficient into account, the frequencies of the couple at [1 ± N r (1 − s)] f have similar magnitudes which are high toward the magnitude of the fundamental. When the K h coefficient is introduced to modify the magnitudes of the components which compose the spectral lines, it can be noticed, in one hand, that the frequencies of this same couple do not have yet the same magnitudes, and on the other hand, that the difference with practical results is reduced (the error is only about 4 dB). Let us also precise that the b g expression given by (7) shows that current harmonics are induced in the stator windings and the rotor bars. These currents have been neglected in our 246 Thailly et al. study but some of them contribute to define the magnitude of the considered spectral lines. That can partially justify the deviations. Conclusion It is shown that all the flux density components do not evolve, through the stator, in the same way according to the pole pair number hp of this component. So, it is determined the correction to apply on the calculated air-gap flux density components in order to appraise the measured spectrum outside of the machine. The presented method permits to analyze a practical spectrum toward a reference one obtained by calculation, that is specific for each machine. Thisquantitativeapproachisinterestingiffaults are introduced in the modeling of the air- gap flux density. In fact, the effects of the faults in the external magnetic field, considering the pole number of the flux density components which are generated, can be precisely predetermined. Acknowledgment This work is part of the project “Futurelec 2” within the “Centre National de Recherche Technologique (CNRT) of Lille.” References [1] W.T. Thomson, M. Fenger, Current signature analysis to detect induction motor faults, IEEE Ind. Appl. Soc. Meet. (IAS Magazine), Vol. 7, No. 4, pp. 26–34, 2001. [2] M.E.H. Benbouzid, Bibliography on induction motors faults detection and diagnosis, IEEE Trans. Energy Convers., Vol. 14, No. 4, pp. 1065–1074, 1999. [3] H. H´enao, T. Assaf, G.A. Capolino, “Detection of Voltage Source Dissymmetry in an Induction MotorUsingthe MeasurementofAxial LeakageFlux”,Proceedings ofInternationalConference of Electrical Machines (ICEM 2000), Espoo Finland, August 2000, pp. 1110–1114. [4] D. Belkhayat, R. Romary, M. El Adnani, R. Corton, J.F. Brudny, Fault diagnosis in induction motors usingradial field measurement with an antenna, Inst. Phys. Meas. Sci. Technol., Vol.14, No. 9, pp. 1695–1700, 2003. [5] D. Roger, O. Ninet, S. Duchesne, Wide frequency range characterization of rotating machine stator-core laminations, Eur. Phys. J. Appl. Phys., pp. 103–109, 2003. [6] J.F. Brudny, Etude quantitative des Harmoniques du Couple du Moteur asynchrone triphas´e d’Induction, Th`ese d’Habilitation, No. H29, Lille, France, 1991. [7] J.F. Brudny, Mod´elisation de la Denture des Machines asynchrones. Ph´enom`ene de resonance, J. Phys. III, pp. 1009–1023, 1997. [8] K.J. Bins, P.J. Lawrenson, Analysis and Computation of Electric and Magnetic Field Problems, Oxford: Pergamon press, 1973. [9] Ph.L. Alger, The Nature of Induction Machines, 2nd edition, New York: Gordon & Breach Publishers, 1970. II-10. IMPACT OF MAGNETIC SATURATION ON THE INPUT-OUTPUT LINEARIZING TRACKING CONTROL OF AN INDUCTION MOTOR D. Dolinar 1 , P. Ljuˇsev 2 and G. ˇ Stumberger 1 1 Faculty of Electrical Engineering and Computer Science, University of Maribor, Smetanova 17, SI-2000 Maribor, Slovenia dolinar@uni-mb.si 2 Ørsted DTU Automation, Technical University of Denmark, Elektrovej, Building 325, DTU, DK-2800 Kgs. Lyngby, Denmark pl@oersted.dtu.dk Abstract. This paper deals with the tracking control design of an induction motor, based on input- output linearization with magnetic saturation included. Magnetic saturation is accounted for by the nonlinear magnetizing curve of the iron core and is used in the control design, the observer of state variables, and in the load torque estimator. Experimental results show that the proposed input-output linearizing trackingcontrolwith the included saturationbehaves betterthantheone without saturation. It also introduces smaller position and speed errors, and better motor stiffness. Introduction The magnetic saturation phenomenon, which occurs in the induction motor (IM) iron core, has been recognized from the very beginning of IM use. The models of saturated induction machines presented in the literature [1–3] show that magnetic saturation is most frequently considered by nonlinear variable mutual inductance, while the additional magnetic cross- couplings are neglected. In general, leakage flux path saturation is rarely included in the models, although its role in the squirrel-cage IM with closed rotor slots can be rather high. The level of idealization, introduced by neglecting magnetic cross-saturation, depends on the type of model, i.e. on the selected state variables [4]. Although magnetic saturation of the IM is very important for the performances of the controlled drive, it has hardly ever been properly considered in control design because of its mathematical complexity. Pioneering efforts to address the perturbing effects of magnetic saturation on the field oriented control are found in [5]. Saturated induction motor models found in the literature [1,3] are used primarily for the analysis of an induction machine operation. Some exceptions in which linear cascade rotor S. Wiak, M. Dems, K. Kom ˛ eza (eds.), Recent Developments of Electrical Drives, 247–260. C 2006 Springer. 248 Dolinar et al. field oriented control is used can be found in [6–8] but they present an incomplete solution for the control algorithm. An observer for state variables that includes magnetic saturation is found in [9]. Some nonlinear control approaches, based on modified IM models that include saturation, are given in [10–13] as well. In this paper an input-output linearizing tracking control design is presented. It is based on the mixed “stator current vector i s , rotor flux linkage vector Ψ r ” saturated IM model introduced by [3]. The effect of magnetic cross-saturation is totally included in the main flux path saturation. The saturation of the stator and rotor leakage flux path is neglected. In addition, a state variable observer and a load torque estimator were designed with included magnetic saturation. An i s , Ψ r saturated IM model is presented in this paper. The input-output linearization based on the introduced modelis carried out,with respect to the rotorposition. Toobtain the required unmeasurable state variables, an observer based on the inverse saturated IM model was designed. The experimental results of the proposed input-output linearizing tracking control of IM with included magnetic saturation show better dynamic performances of the drive than the classical control, where magnetic saturation is not considered. The main improvements are the smaller rotor position and speed errors, as well as higher stiffness and better load torque rejection, which results in a smaller stator current when the motor is loaded with a step change in the load torque [14]. Saturated induction machine model The general approach to IM modeling incorporating magnetic saturation with different selections of state variables is presented in [3]. The standard two-phase model of an IM in the general reference frame with stator current vector i s and rotor flux linkage vector Ψ r as state variables, is given by a nonlinear model (1) u = A(x) dx dt + B(x)x J dω r dt = ( t e − t l ) − f ω r (1) where u = u sd u sq 00 T x = i sd i sq ψ rd ψ rq T A(x) = ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ L l − L 2 rl L dd − L 2 rl L dq 1 − L rl L dd − L rl L dq − L 2 rl L dq L l − L 2 rl L qq − L rl L dq 1 − L rl L qq 0010 0001 ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ II-10. Impact of Magnetic Saturation of Induction Motor 249 B(x) = ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ R s −ω g L l − L 2 rl L r 0 −ω g L m L r ω g L l − L 2 rl L r R s ω g L m L r 0 −R r L m L r 0 R r L r −ω sl 0 −R r L m L r ω sl R r L r ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ t e = p L m L r (i sq ψ rd −i sd ψ rq ) u sd , u sq and i sd , i sq are the d- and q-axis stator voltages and currents, ψ rd and ψ rq are the rotor flux linkages, R s and R r are the stator and rotor resistances, L m is the mutual static inductance, L s and L sl are the stator self-inductance and the stator leakage inductance, L r and L rl are the rotor self-inductance and the rotor leakage inductances, L l = L sl + L rl is the total leakage inductance, L dq is the cross-coupling inductance, L dd and L qq are the inductances along d- and q-axis, ω g is the angular speed of the general reference frame, ω r is the rotor angular speed, ω sl = ω g − ω r is the slip angular frequency, J is the drive moment of inertia, f is the coefficient of viscous friction, t e and t l are the electrical and the load torque, and p is the number of pole pairs. Inductances L dd , L qq , and L dq are given as follows: 1 L dd = 1 L rl + L cos 2 ρ m + 1 L rl + L m sin 2 ρ m 1 L qq = 1 L rl + L m cos 2 ρ m + 1 L rl + L sin 2 ρ m , m = | Ψ m | 1 L dq = 1 L rl + L − 1 L rl + L m cos ρ m sin ρ m , i m = | i m | (2) L m = m i m , L = d m di m , cos ρ m = ψ md m , sin ρ m = ψ mq m where i m is the modulus of the magnetizing current vector. L represents the so-called dynamic inductance, ψ md and ψ mq are the d- and q-components of the magnetizing flux linkage vector Ψ m , ρ m is the angle of magnetizing flux linkage vector Ψ m , while m is the modulus of Ψ m . L m and L are given by equations (2) and are obtained from the measured magnetizing curve presented in Fig. C1, Appendix C. Two-phase i s , Ψ r model (1) can be written in the state-space form as: dx dt =−A(x) −1 B(x)x + A(x) −1 u = Cx +ω g Zx + ω r Wx + Du (3) dω r dt = p J L m L r (i sq ψ rd −i sd ψ rq ) − 1 J t l − f J ω r 250 Dolinar et al. where: C = ⎡ ⎢ ⎢ ⎣ c 11 c 12 c 13 c 14 c 21 c 22 c 23 c 24 c 31 0 c 33 0 0 c 42 0 c 44 ⎤ ⎥ ⎥ ⎦ Z = ⎡ ⎢ ⎢ ⎣ z 11 z 12 z 13 z 14 z 21 z 22 z 23 z 24 000z 34 00z 43 0 ⎤ ⎥ ⎥ ⎦ W = ⎡ ⎢ ⎢ ⎣ 00w 13 w 14 00w 23 w 24 00 0 w 34 00w 43 0 ⎤ ⎥ ⎥ ⎦ D = ⎡ ⎢ ⎢ ⎣ d 11 d 12 00 d 21 d 22 00 0000 0000 ⎤ ⎥ ⎥ ⎦ Elements of matrices C, Z, W, D are given in Appendix A. Input-output linearization The two-phase IM model (3) is transformed to the stationary reference frame αβ where ω g = 0. It is written in the compact form (4) ˙x = f(x) +Gu = f(x) +g α u sα + g β u sβ (4) where x = θ r ω r i sα i sβ ψ rα ψ rβ f(x) = ⎡ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ ω r p L m L r 1 J i sβ ψ rα −i sα ψ rβ − 1 J t l − f J ω r c 11 i sα + c 12 i sβ + c 13 ψ rα +c 14 ψ rβ + w 13 ω r ψ rα + w 14 ω r ψ rβ c 21 i sα + c 22 i sβ + c 23 ψ rα +c 24 ψ rβ + w 23 ω r ψ rα + w 24 ω r ψ rβ c 31 i sα + c 33 ψ rα + w 34 ω r ψ rβ +c 42 i sβ + c 44 ψ rβ + w 43 ω r ψ rα ⎤ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ (5) G = g α g β = 00d 11 d 21 00 00d 12 d 22 00 T θ r is the rotor angle, u sα ,su sβ and i sα , i sβ are the stator voltages and currents and ψ rα and ψ rα are the rotor flux linkages. Coefficients c (·) , w (·) , and d (·) are given in Appendix A. The output vector y = θ r 2 r T is chosen by (6), where 2 r = ψ 2 rα + ψ 2 rβ y = φ(x) = φ 1 (x) φ 2 (x) T = θ r 2 r T (6) Since the load torque t l is not the system control input and cannot be directly measured, it is excluded from the nominal part of the motor model and is later considered as an external disturbance. Theload torque isobtained by a simple load torque estimator described in [14]. The input-output linearization technique is based on exact cancellation of the system’s nonlinearities in order to obtain a linear relationship between the control inputs and the system outputs in the closed loop. The nonlinear control law is deduced to a successive differentiationof each outputuntil at leastone input appearsin the derivative.The derivative II-10. Impact of Magnetic Saturation of Induction Motor 251 of the first output φ 1 (x) = θ r is: ˙ y 1 = ∂φ 1 (x) ∂x dx dt = L f φ 1 = ω r (7) The second and the third derivatives are given by (8) and (9). ¨ y 1 = ∂ ∂x ˙ y 1 dx dt = ∂ ∂x (L f φ 1 ) dx dt = L 2 f φ 1 (8) = p L m L r 1 J i sβ ψ rα −i sα ψ rβ − 1 J t l − f J ω r y ··· 1 = ∂ ∂x ¨ y 1 dx dt = L 3 f φ 1 + L gα L 2 f φ 1 u sα + L gβ L 2 f φ 1 u sβ (9) Lie derivatives L 3 f φ 1 , L gα L 2 f φ 1 , and L gβ L 2 f φ 1 are given in Appendix B. The input voltages u sα and u sβ appeared in y ··· 1 , therefore, the relative degree of the first subsystem is three. The second output to be differentiated is the square of the rotor flux linkage modulus 2 r . Its first derivative is given by equation (10). ˙ y 2 = ∂φ 2 (x) ∂x dx dt = L f φ 2 (10) = 2R r L m L r i sα ψ rα +i sβ ψ rβ − 2R r L m L r ψ 2 rα + ψ 2 rβ The second derivative of 2 r is given by (11). ¨ y 2 = ∂ ∂x ˙ y 2 dx dt = ∂ ∂x (L f φ 2 ) dx dt (11) = L 2 f φ 2 + L gα L f φ 2 u sα + L gβ L f φ 2 u sβ Lie derivatives L 2 f φ 2 , L gα L f φ 2 , and L gβ L f φ 2 are given in Appendix B. The input voltages u sα and u sβ appeared in ¨ y 2 , therefore, the relative degree of the second subsystem is two. Consequently, the total relative degree of the system is unequal to the system order n = 6, which reveals the existence of uncontrollable internal dynamics. It is easy to prove (see Ref. [14]) that the dynamics of the third subsystem is stable if the third output φ 3 is selected as φ 3 = arctan ψ rβ /ψ rα , as in [11]. In the next step the input-output linearization of the nominal system is done [11]. The nominal part of the IM model is written in the form of higher derivatives of outputs y 1 and y 2 y ··· 1 ¨ y 2 = L 3 f φ 1 L 2 f φ 2 + E(x) u sα u sβ = D(x) +E(x)u (12) where E(x) = L gα L 2 f φ 1 L gβ L 2 f φ 1 L gα L f φ 2 L gβ L f φ 2 (13) is the decoupling matrix. The decoupling matrix E(x) is nonsingular, except at the motor start-up, where ψ 2 rα + ψ 2 rβ = 2 r [14]. 252 Dolinar et al. The obtained model (12) is still nonlinearand coupled. The linearized and decoupled IM model (15) is obtained by an appropriate selection of the control input u = u sα u sβ T (see equation (14)). u sα u sβ = E −1 (x) − L 3 f φ 1 L 2 f φ 2 + v α v β (14) Note that v = v α v β T in equation (15) is the new system input. y ··· 1 ¨ y 2 = D(x) +E(x) E −1 (x)D(x) + E −1 (x)v (15) = D(x) +E(x)u = v α v β = v The input-output behavior of the system (15) is linear, but the relationship between the control input v and the states x is still nonlinear. This nonlinear relationship is eliminated by selecting a new set of state variables, z =[z 1 z 2 z 3 z 4 z 5 ] introduced by the nonlinear transformation z = T(x), defined as: y = T(x) = ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎢ ⎣ θ r ω r p L m L r 1 J i sβ ψ rα −i sα ψ rβ − 1 J t l − f J ω r 2R r L m L r i sα ψ rα +i sβ ψ rβ − 2R r L m L r ψ 2 rα + ψ 2 rβ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎥ ⎦ (16) Theblockdiagramofthedecoupledand linearizedIMmodelisgiveninFig.1.Thissystem is linearized, decoupled and unstable. Both stabilization and tracking can be achieved without any concern about the stability of the internal dynamics using the linear tracking controllers designed by the pole placement [14]. The reference output vector y ∗ is given by (17) y ∗ = y ∗ 1 y ∗ 2 = ∗ r 2∗ r (17) where ∗ r and 2∗ r represent the reference trajectories of the rotor position and the square of rotor flux linkage modulus. The differences between the reference and the actual values of the controlled outputs are tracking errors (18). e 1 = y ∗ 1 − y 1 , e 2 = y ∗ 2 − y 2 (18) . v = α y 1 v= β y 2 z 1 (0) z 2 (0) z 3 (0) z 4 (0)z 5 (0) z 5 =z 4 . z 3 =z 2 . z 2 =z 1 z 1 Θ=y 1r z 4 =y=Ψ r 2 2 = Figure 1. Linearized and decoupled IM model. . the analysis of an induction machine operation. Some exceptions in which linear cascade rotor S. Wiak, M. Dems, K. Kom ˛ eza (eds.), Recent Developments of Electrical Drives, 247 260 . C 2006. modulus of the magnetizing current vector. L represents the so-called dynamic inductance, ψ md and ψ mq are the d- and q-components of the magnetizing flux linkage vector Ψ m , ρ m is the angle of. Science, University of Maribor, Smetanova 17, SI-2000 Maribor, Slovenia dolinar@uni-mb.si 2 Ørsted DTU Automation, Technical University of Denmark, Elektrovej, Building 325, DTU, DK-2800 Kgs. Lyngby,