INVESTIGATIONS ON SENSORLESS VECTOR CONTROL OF MLI FED INDUCTION MOTOR DRIVE

Abstract

Variable speed operation of induction motor drives can be generally classified as scalar control and vector control. Scalar control is used for low performance drives where only the magnitude and frequency of the stator voltage or current is regulated. The most commonly used scalar control technique is the constant Volts/Hertz (V/f) control, which offers moderate dynamic performance and is therefore used in applications where high speed precision is not required such as fans, pumps and elevators. These control methods result in poor torque and flux response when a high dynamic performance operation is required. High performance induction motor drives can be implemented by using vector control. This method realises the speed control of IM in a similar manner as in DC motor, where the flux and torque are controlled independently by the field current and armature voltage. In vector control of induction motor, the stator current is resolved into the direct axis and quadrature axis current components. The former is used to control the flux in the motor and the latter to control the torque. The field angle calculation is critical for implementation of the vector control and it is obtained from the rotor angle and slip speed angle, which is useful for decouple control of torque and flux. The vector control scheme is classified as Direct and Indirect vector control depending on how the field angle is obtained. The closed loop control of vector-control of IM drives requires accurate information of speed or position. This information is provided by the speed/position sensor. However the use of speed/position sensor has many drawbacks such as, higher cost, lower reliability, increase in weight and size and difficulty of use in harsh environment. The drawbacks of sensors can be eliminated by using sensorless speed estimation algorithms for vector control. Sensorless techniques are classified based on the machine model and signal injection methods. Machine model methods are widely used in most of the AC motor drive, but their performance during starting is poor, whereas signal injection methods are having a better starting operation usually machine specific which limit their industrial applicability especially to induction machine. Several techniques such as Model Reference Adaptive System (MRAS), Luenberger and Kalman filter observers, sliding mode observers and Artificial Intelligence (AI) techniques are employed to estimate the rotor flux and speed. Most sensorless algorithms are based on the flux and speed estimations which are obtained from the IM voltage equations, but they are sensitive to machine parameters. Among various speed sensorless estimation techniques MRAS schemes offer simple implementation and require less computational effort compared to other methods. Various MRAS observers have been introduced based on the error quantity of the adaptation scheme as rotor flux, back e.m.f and reactive power MRAS. ii Back e.m.f based estimator improves the low speed performance but has stability problems, as well as at high speeds it suffers from noise and is more sensitive to stator resistance variation. The reactive power based MRAS estimation scheme suffers from the inherent instability in the low speed during regenerative mode, poor starting operation and sensitive to the rotor resistance variation. The rotor flux based MRAS is the most popular MRAS strategy compared to other MRAS schemes. The low speed performance of the rotor flux based MRAS is poor due to the presence of integrator in the reference model and is more sensitive to the motor parameter variation. Several methods have been proposed to reduce the effects of the integrator. Low-pass filters (LPFs) with low cut-off frequency have been proposed to replace the integrator. However, this method introduces phase, gain errors and delays in estimated speed relative to the actual, which may affect the dynamic performance of the drive in addition to inaccurate speed estimation below the cut-off frequency. To overcome this problem a programmable cascaded low pass filter is used to replace the integration by small time constant cascaded filters to attenuate the DC offset decay time. A nonlinear feedback integrator for drift and DC offset compensation has been proposed. Another technique has been proposed in literature where the rotor flux is estimated by defining a modified integrator having the same frequency response as the integrator at steady state. A rotor flux based MRAS speed estimator has been proposed with drift and inverter non linearity compensation for stator flux oriented vector control scheme. An artificial neural network (ANN) based adaptive speed estimator has been presented using MRAS with instantaneous and steady state reactive power. This adaptive system performs satisfactorily at very low speed with ANN to overcome stability problems. Vector control of IM when used in high power and medium voltage applications, the fundamental output voltage and corresponding power level of the inverter are to be increased to a high value. Two-level voltage source inverter, have a limitation at high voltages due to maximum voltage withstand capability in switching devices and high stress in the windings as they cannot operate at high switching frequencies. High performance of IM drive systems at increased power levels requires high quality inverter output with low harmonic loss and generating low torque ripple. Multi level inverter (MLI) is an emerging technology used in high power applications because they provide high quality inverter output voltage with low harmonic distortion, low torque fluctuations and avoid the limitations of two-level voltage source inverter in vector control and direct torque controlled drive system. There are three main topologies, namely Diode Clamped Multi Level Inverter (DCMLI), Flying Capacitor Multi Level Inverter (FCMLI) and Cascaded H-bridge Multi Level Inverter (CHBMLI). The DCMLI, particularly the three-level, has drawn much interest in IM drive applications because it needs only one common voltage source. Vector control can be iii naturally being extended to MLI fed drives with the modulation method extended to multi level Pulse Width Modulation (PWM). The direct torque control on the other hand cannot be easily extended for multi level inverter due to the large number of possible voltage vectors available for selection in PWM. Space Vector Modulation (SVM) is the most promising switching algorithm for MLI, because it offers great flexibility in optimizing switching pattern design, it can maximize the output voltage, suits well for digital implementation and drives application. Extensive review in the area of sensorless vector control technique and advantage of using a multi level inverter in induction motor drives explores the various gaps such as poor low speed performance due to speed estimation owing to its sensitivity to machine parameter variation like stator resistance variation with temperature, the presence of integrator in the reference model results in DC drift and initial value problems, the nonlinearity in inverter produces a difference between the actual machine terminal voltage and the reference model voltage. At low speeds distortion in voltage due to nonlinearity of inverter becomes more as the stator terminal voltage is low and gives an error in the flux estimation which results in the inaccurate speed estimation. MLI in sensorless vector control of induction motor improves the performance of the drive by providing high quality of inverter output voltage used in speed estimation. AI techniques have attracted much attention in the past few years as powerful tools to solve many control problems. The commonly used AI strategies are neural networks, fuzzy logic and genetic algorithms (GA). The main purpose of this work is to improve the performance of the rotor flux based MRAS at low speed operations with a proposed modified voltage model to estimate speed, NN based estimator to estimate stator resistance for an induction motor drive fed by three-level DCMLI with space vector modulation technique. In the present work the performance of rotor flux based MRAS is improved by modifying the reference model. A compensating voltage is added to the reference model dq axis rotor flux components by modifying the integrator, which uses the reference rotor flux of an indirect vector control scheme for obtaining the compensating voltage based on nonlinear feedback method. The compensating voltage is used to reduce the DC drift problem and is the combination of drift voltage and offset voltage. The offset voltage is obtained from the back-e.m.f of motor whose value is within limits of the reference rotor flux of the controller, drift value is a small voltage which improves the convergence value of the offset voltage and it is computed from proportional integral (PI) controller. PI controller is used in this scheme to obtain the drift value, which is useful in making equal voltage and current model dq axis rotor flux components. The speed tuning signal of the adaptation scheme in rotor flux based MRAS must be zero to obtain the accurate rotor speed estimation. The dq axis rotor flux iv components of reference model are made equal to an adaptive model for accurate rotor speed estimation. Fuzzy logic and neural network controllers are introduced in the adaptation scheme and modified reference model in place of the PI controller with an aim to further enhance the performance. The fuzzy logic controller (FLC) is implemented using the Mamdani type controller and Neural Network Controller (NNC) is of (2-12-1) multi-layer feedforward type for estimate the rotor speed and (2-8-1) is for obtaining the drift value in the modified reference model. The low speed performance of the rotor flux based MRAS speed estimation methods i.e. conventional rotor flux based MRAS, modified voltage model MRAS, modified voltage model MRAS using FLC, modified voltage model MRAS using NNC as adaptation schemes are compared with simulation studies performed on MATLAB/Simulink platform and validated with experimental results by the developed prototype model. The precise calculation of stator resistance is of crucial importance for accurate speed estimation of a sensorless drive in the low speed region, since any mismatch between the actual value and the value used within speed estimator may not only lead to a substantial speed estimation error but affects the stability. In general, all methods rely on stator current measurement and require information regarding stator voltages. Various parameter estimation techniques for estimating the stator and rotor resistance and magnetizing inductance have been presented in the literature using rotor flux based MRAS. For testing the variation of stator resistance of the drive, rotor flux based MRAS speed estimation scheme is implemented for two and half hours and its variations with time are recorded. A new stator resistance estimation method is proposed to improve the low speed performance of the rotor flux based MRAS. The stator resistance estimation block is replaced with NN based estimator which is more robust compared to earlier methods. The data obtained from speed sensorless indirect vector control method based on rotor flux based MRAS is simulated first and its data are saved in the workspace of MATLAB, the same data is used for NN based stator resistance estimation. It also eliminates the PI controller in the stator resistance estimator. Detailed simulation and experimental tests are carried out to investigate the performance of the drive with modified voltage model for rotor flux based MRAS, modified voltage model for rotor flux based MRAS using FLC in adaptation scheme modified voltage model for rotor flux based MRAS using NNC in adaptation scheme, stator resistance estimator and NN based stator resistance estimator are compared with the conventional rotor flux based MRAS. A prototype model of the rotor flux based MRAS observer for estimating the speed and stator resistance is developed in the laboratory using IGBT based Intelligent Power Module PEC 16DSMO1 of rating (25A-1200V). A speed sensor (HEDS 5605) is used only for comparing the actual rotor speed with estimated rotor speed from the proposed v speed observers. The real time implementation for validation of simulation study of proposed speed estimation methods for an 1 HP SEIMENS induction motor is performed with prototype model developed using dSPACE DS-1104 R&D control board for a two-level inverter fed induction motor drive at various operating conditions (both high and low speed). The results obtained show the improvement in the overall performance of the MRAS observer at low speed operation with faster settling time and lower peak overshoot and undershoot for NN based methods