162 Hanamoto et al. Table 1. Specifications of the tested motor Items Value Rated power (W) (rated/maximum) 86/355 Rated torque (Nm) (rated/maximum) 0.4/0.96 Rated current (Arms) (rated/maximum) 1.7/2.6 Inverter voltage (V) 200 Armature resistance () 1.89 d-axis armature inductance (mH) 93 q-axis armature inductance (mH) 36 Number of pole pairs 2 120 1800 time [s] 1200 600 1800 1200 600 0 w w w [min −1 ] w [min −1 ] Figure 6. Experimental results using speed sensor. 120 1800 time [s] 1200 600 1800 1200 600 0 w w [min −1 ] w [min −1 ] w Figure 7. Experimental results using proposed method. speed sensor. Fig. 7 shows the results using the proposed modified flux linkage observer when the same condition. From this figure, the modified flux linkage observer gives almost the same results as them measured by a sensor. As a result our proposed methods are effective to the sensorless control of Syn.RM. Fig. 8 illustrates the performance in the steady state. The measured and estimated rotor position of the middle speed and high-speed operation are shown. From the figure, the accuracy estimation results are obtained even if the high-speed region. Fig. 9 show the output of the estimation error correct function K y . From this figure K y is decreased when the effective current i rms is increased. And K y is also the function of the motor speed. II-2. Sensorless Control of Syn.RM Using Modified Flux Linkage Observer 163 6 4 0 2 0 0.04 0.08 w = 500 min −1 (a) time [s] [rad] q q q 6 4 0 2 0 0.04 0.08 w = 1500 min −1 (b) time [s] [rad] q Figure 8. Measured and estimated rotor positions in the steady state. 1.2 0.8 0.4 0 1.5 2 2.5 500 min −1 1000 min −1 1500 min −1 i rms [A] K y Figure 9. i rms vs. K Y . Conclusion In this paper, we proposed the modified flux observer with the estimation error correct function for the sensorless control method of Syn.RM. The validity of the proposed method is verified by experiments. References [1] T. Tamamura, Y. Honda, S. Morimoto, Y. Takeda, “Synchronous Reluctance Motor When Used Air-Condition Compressor Motor: A Comparative Study”, IPEC-Tokyo 2000, 2000, pp. 654–659. [2] T. Senju, T. Shingaki, K. Uezato, Sensorless vector control of synchronous reluctance motor withdisturbancetorqueobserver,IEEETrans.Ind.Electron.,Vol.48,No.2,pp.402–407,2001. [3] S. Shinnaka, Mirror-phase characteristics of synchronous reluctance motor and salient-pole orientation methods for sensorless vector control, Trans. IEE Japan, Vol. 121-D, No. 2, pp. 210–218, 2001 (in Japanese). [4] S. Saha, T. Iijima, K. Narazaki, Y. Honda, “High Speed Sensorless Control of Synchronous Reluctance Motor by Modulating the Flux-Linkage Angle”, IPEC-Tokyo 2000, 2000, pp. 643–648. 164 Hanamoto et al. [5] T. Hanamoto, T. Tsuji, Y. Tanaka, “Sensorless Speed Control of Cylindrical Type PMSM Using Modified Flux Observer”, IPEC-Tokyo 2000, 2000, pp. 2104–2108. [6] K. Yamazaki, S.P. Kommareddy, J. Liu, “Durable PC-Based Real-Time Control System for Servomotor Control in Windows NT Environment”, IPEC-Tokyo 2000, 2000, pp. 355–360. [7] T. Hanamoto, H. Ikeda, T. Tsuji, Y. Tanaka, “Sensorless Speed Control of Synchronous Reluctance Motor Using RTLinux”, PCC-Osaka 2002, Vol. 2, pp. 699–703, 2002. [8] http://www.rtlinux.org. [9] http://rtic-lab.sourceforge.net/. II-3. A NOVEL SENSORLESS ROTOR-FLUX-ORIENTED CONTROL SCHEME WITH THERMAL AND DEEP-BAR PARAMETER ESTIMATION Mario J. Duran 1 , Jose L. Duran 1 , Francisco Perez 1 and Jose Fernandez 2 1 University of M´alaga, Electrical Engineering Department, Plaza El Ejido S/N 29013 Malaga, Spain mjduran@uma.es, jlduran@uma.es, fmperez@uma.es 2 University of Jaen, Electrical Engineering Department, Alfonso X, 28, 23700 Linares (Ja´en), Spain jmoreno@ujaen.es Abstract. In this paper a novel scheme for vector control is presented that aims to improve some of the weaknesses of the sensorless vector control. Among indirect rotor-flux-oriented control (IRFOC), some of the aspects that can be improved are the low speed behavior, current control, and parame- ter detuning. The present scheme includes temperature estimation to correct the deviation in steady state, a new control scheme with skin effect estimation to improve the transient accuracy, and an overcurrent protection to be able to have a control on the stator current while allowing a good per- formance. The proposed scheme is designed from the Matlab/Simulink environment and is exper- imentally tested using a 1 kW induction motor and a TMS320C31 DSP proving its validity and usefulness. Introduction Sensorless vector control is a mature technology whose origins go back to the early seventies[1]. However, many high performance induction motor drives are still being pro- posed since some problems are still not solved. Sensorless operation mode has attracted much attention and twomain approaches can be considered: those based on the field orienta- tion principles to carry out the control FOC [1] and the direct torque control DTC [2] which is inherently a sensorless method. Both have their own weaknesses and a lot of research work has been done trying to solve them. Among the rotor field oriented schemes, the indirect approach is the most popular, but still presents problems concerning parameter detuning [3] and low speed performance [4]. In these sensorless schemes, it is necessar y to estimate the speed since no encoders are used, and this can be carried out directly from the motor model, or using other approaches such as MRAS [5] or Kalman filter methods [6]. In this work the speed is estimated from the motor model but including the mechanical equation, which allows including new phenomena such as the static friction. On the other S. Wiak, M. Dems, K. Kom ˛ eza (eds.), Recent Developments of Electrical Drives, 165–176. C 2006 Springer. 166 Duran et al. hand, parameter estimation is carried out to account for thermal and skin effects. Models for both effects are lumped parameter and simple enough to be included in the real-time application. Skin effect model must be calculated in every step while the thermal model could be implemented with a greater step size since the thermal constant time is higher. Parameters for both models are calculated experimentally, in the case of the motor heating, and considering the analytical case for the deep-bar effect [8]. Additionally, the problem of current control is considered proposing an overcurrent protection that allows transient currents above the rated value, which improves the drive performance. Control scheme As previously stated, speed estimation is required to carry out the control, and this can be done by using three from the four well-known RFOC equations [7], building a speed estimator from one of the rotor electrical equations. Nevertheless, in this paper a different proposal is made including the dynamical equation (1) into the estimator together with the four RFOC equations, obtaining a set of equations: T s pi d +i d = u d R s + ω mr T s i q − (T s − T s )p|i mr | T s pi q +i q = u q R s − ω mr T s i d − (T s − T s )|i mr | T r p|i mr |+|i mr |=i d (1) ω mr = ω + i q T r |i mr | P 2 L 2 m L r i q |i mr |=T m + Jpω 2 P + α f ω 2 P Instead of obtaining the speed from an electrical equation, it is obtained from the me- chanical one, including new parameters as the inertia or friction coefficient. It allows to include a variable friction that takes into account the static friction when the movement starts. From the set of equations (1) the motor speed can be calculated using an estimator whose inputs are the voltage components and the load torque. Since these are the real inputs of an induction machine, the estimator is further called simulated motor. Voltage components can be measured or reconstructed from the stator equations, but torque must be estimated since it is not a measured variable (Fig. 1). In order to estimate the load torque, and adaptive scheme is adopted considering that the motor torque is proportional to i q . This component can be obtained from the simulated motor as an estimated value and can also be measured. For the measurements of the currents a digital filter is used before considering the transformation matrix into the dq values. The difference between them is due to the fact that the information of the load torque that the estimator is using is not correct, and so a controller can be used to update this torque value. In Fig. 2 the complete control scheme is shown. II-3. Sensorless Rotor-Flux-Oriented Control Scheme 167 Figure 1. Evolution of the friction torque. The speed and flux target values are compared with the estimated ones obtaining target currentsthataretransformedintotargetvoltagesbyreconstructionfromthestator equations. These voltages are, together with the estimated torque, the inputs for the simulated motor and also the outputs to generate the PWM for the VSI inverter. Three controllers are involved in the control as it is usual in this kind of vector control, one for the flux comparison obtaining the direct component i d and two for the speed and torque comparison obtaining the quadrature component i q . The rotor flux reference decreases in inverse proportion to the speed of rotation in the field-weakening region, while it is constant and equal to rated rotor flux in the base speed region. Figure 2. Control scheme using the simulated motor as a speed estimator with adaptive load torque estimation. 168 Duran et al. Overcurrent protection Vector control provides high performance to drives,but to achieve a good transient response a high electrical torque is required, and it means high currents flowing in the machine. Apart from the achievement of decoupled and optimal control, it is also necessary to protect the motor against overcurrents. The control will need high currents to provide torque, and this can be necessaryforhigh accelerations or load torque.In Fig.3, it is shownthe stator current evolution reference speed is increased in a rampwise manner with high accelerations, and it is clear that in these transient, currents over the rated value are required. The inverter used is VSI type and so to carry out a proper control, target voltages are supplying the motor, but without anycurrent control. For this reason, it is convenienttoincludeacurrent protection in the software design. To have control over the currents in field oriented control is relatively easy compared with other schemes such as direct torque control (DTC), since it is only necessary to control the quadrature component of the stator space vector i q . The obvious solution is just to saturate this component in the control scheme. However, the aim here is to design a nonconservative protection that allows transient currents above the rated one. Traditionally, for steady-state operation, manufacturers provide the maximum time for a certain value over the rated currentso that the motor is not damaged. For a vector control ap- plication the motor works in transient state, but a protection based on energy considerations can also be used. The method proposed is to use an energy counter that starts to rise then the current is over the nominal value by integrating this current. When the energy counter is over a certain energy threshold, then the protection acts limiting i q to its nominal value. The method proposed is to use an energy counter that starts to rise then the current is over the nominal value by integrating this current. When the energy counter is over a certain energy threshold, then the protection acts limiting i q to its rated value. Figure 3. Three-phase currents during acceleration transient. II-3. Sensorless Rotor-Flux-Oriented Control Scheme 169 Figure 4. Proposed protection. Some considerations have to be made in order to make the system work: the integration for the energy counter must be limited and if the current is below the nominal value the integration must continue with a negative value until the energy counter is set to zero. This makes that, if a repetitive cycle occurs (Fig. 4), the protection finally acts even if the energy of the each cycle is below the limit. Parameter estimation In the proposed control scheme a speed estimator was built from the motor model equations, and some parameters were involved in this estimation. Since this parameters change with the operation conditions, the problem of parameter detuning common to all vector control remains the same. In order to overcome this problem, two of the main influencing factors are considered: thermal and deep-bar effects. In a previous paper [7] a thermal model is developed that takes electrical RFOC variables as inputs and provides stator and rotor representative temperatures. The model is simple enough not to be time consuming, but proves to be very accurate. Considering the copperlosses,hysteretic,andeddycurrent losses and taking into account only the stator, rotor, and environment, representative temperatures of the stator and rotor can be obtained making thermal balance. R s i 2 s + k Hs ω s + k Fs ω 2 s = G s θ s +C s dθ s dt + G sr (θ s − θ r ) (2) R r i 2 r + k Hr ω r + k Fr ω 2 r = G r θ r +C r dθ r dt + G sr (θ r − θ s ) Both conduction and convection are considered in the thermal conductances. G = G 0 (1 + b ·ω) (3) Model parameters are obtained from three tests: blocked shaft, DC, and AC tests. For the conductances and convection coefficients it is only necessary to consider steady-state values, while for the capacitances the thermal transient must be taken into account. 170 Duran et al. Table 1. Results for the thermal tests Simulation Tests and simulations Test temperature temperature Variable f (Hz) θ s θ r θ s θ r DC 1 0 61.9 58.6 62.2 58.2 2 0 79.4 74.0 79.8 74.3 AC Blocked shaft 1 4 51.1 63.4 50.0 60 2 5.5 71.6 91.0 70.8 88 AC 1 0 41.25 50.8 41.4 50.8 2 40 43.66 58.3 43.0 57.0 The results for the different test carried out to obtain the different parameters can be summarized in Table 1 showing the experimental and simulation steady-state results. To account for the deep-bar effect, FEM solutions are not possible for a real-time appli- cation, and both analytical [9] and lumped parameters [10] of previous solutions are only valid for rectangular rotor bars. In the present work the classical analytical solution [9] is generalized starting from the same wave equation but changing the contour condition in the upper part of the bar so that there is a contribution of the sides when using Amp `ere’s law. The same occurs when the Poynting’s flux is considered, and so there is also flux through the sides, and not only through the upper partof the bar, what is considered in the equations: P + jQ = EH (0)b(0) +2 h 0 EH (z)b(z) dz = I 2 r R r + j I 2 r L r 2 (4) All in all, a general analytical solution is presented whose main weakness is to be time- consuming due to hyperbolic functions. Because of that an approximate solution is considered that starts from a lumped param- eter π equivalent circuit (Fig. 5), and calculates the values of the different parameters by comparing the results with the previous analytical solution and minimizing the error shown in (5) using a Nelder-Mead direct search. E = (1/f cr ) · R an r − R calc r 2 + p RL (1/ f cl ) · X an r − X calc r 2 (5) being cr and cl weight coefficients that improve the solution performance at low or high frequencies, P rl a coefficient that allows a better adjustment of resistance or inductance. Figure 5. Lumped parameter π rotor equivalent circuit. II-3. Sensorless Rotor-Flux-Oriented Control Scheme 171 Figure 6. Global estimation scheme. Consideringacircuitwiththreesections,six resistancesandthreeinductancesareneeded. Weight values of cr = 2, cl = 2.5, and P rl = 100 have been chosen so that the estimation of both parameters is compensated and for a best adjustment at low frequencies, obtaining the following parameters: R 1 = 0.92 /cm, R 2 = 0.029 /cm, and R 3 = 0.11 /cm R 4 = 0.062 /cm, R 5 = 0.016 /cm, and R 6 = 0.123 /cm L 1 = 6.65 μH/cm, L 2 = 27.4 μH/cm, and L 3 = 5.1 μH/cm Including both the deep-bar effect and thermal model into the speed estimator, the scheme shown in Fig. 6 is finally obtained. The stator resistance is updated considering just the stator representative temperature, since the skin effect is neglected in the stator. The rotor resistance is updated thanks to the skin effect model, which already takes into account the rotor temperature changes since one of its inputs is the rotor representative temperature. It must be noticed that the motor heating influences the deep-bar model due to the electrical conductivity variation but the motor temperature is not practically affected by the additional losses caused by the skin effect. Experimental rig Intheexperimentalrig,there area1kWinductionmotor(AEG eAM90SY4Ex), aSemikron Skiip with integrated rectifier and VSI inverter, a dynamo and a bank resistor for load tests, and a digital signal processor (DSP) (TMS320C31) main control board (Fig. 7). The control design is carried out in Simulink (Matlab) and compiled to be executed in the DSP. Forthe acquisition data, twotypes can be considered: the control data and the verification data. Control data are the currents necessary to estimate the motor speed, which need PCBs specifically designed with Hall effect transducers to obtain proper voltages that can be introduced in the DSP thanks to 16-bit A/D converter. Moreover, the speed is also measured in order to have a verification tool, and so it is vital to carry out the measurement with high precision. For this purpose, a 1024 CPR encoder is used and the TTL pulses are filtered and counted into the DSP obtaining the shaft position. . Fernandez 2 1 University of M´alaga, Electrical Engineering Department, Plaza El Ejido S/N 29013 Malaga, Spain mjduran@uma.es, jlduran@uma.es, fmperez@uma.es 2 University of Jaen, Electrical Engineering Department,. Shinnaka, Mirror-phase characteristics of synchronous reluctance motor and salient-pole orientation methods for sensorless vector control, Trans. IEE Japan, Vol. 121-D, No. 2, pp. 210– 218, 2001 (in. 2002. [8] http://www.rtlinux.org. [9] http://rtic-lab.sourceforge.net/. II-3. A NOVEL SENSORLESS ROTOR-FLUX-ORIENTED CONTROL SCHEME WITH THERMAL AND DEEP-BAR PARAMETER ESTIMATION Mario J. Duran 1 ,