Torque ripple reduction of asynchronous motors by neural-fuzzy networks

Induction motors have complex mathematical models with high degree of nonlinear differential equations including speed and time dependent parameters. Depending on this complex model, there is a coupling effect between the flux and torque of the motor because formation of the flux and torque is a function of the voltage and frequency applied to the motor. The speed and torque response of an induction motor controlled by scalar control methods is poor due to the effect of this coupling. For the control of the motor, mathematical model represented by three phase axis variables is reduced to a model which is represented by two axis variables after some analytical analysis. Using the two axis mathematical model of the motor, the flux and torque can be easily decoupled to make the motor?s flux and torque control possible. With the development of Vector Control (VC) method, it has been noticed that the poor response of the scalar control method does not stem from the structure of the motor. Performance of the motor depends on the control method applied to the drive system.VC method is achieved by aligning any of the stator, rotor or airgap flux with a predetermined two axis reference frame, and therefore this control method is called in various ways depending on which flux of the motor is oriented. Among these various approaches, orientation of the stator flux which is called Direct Torque Control (DTC) has the simplest structure. In the DTC method, the flux and torque of the motor are controlled directly using the flux and torque errors which are processed in two different hysteresis controllers. Therefore by using the DTC method, high performance is obtained by means of the speed and torque. However, because of the hysteresis control, high ripples exist especially in the torque of the motor. Besides that, the switching frequency of the inverter is variable depending on the speed, load and bandwidth of the hysteresis controllers.In the study of this thesis, it is a aimed to reduce the high torque ripples of a direct torque controlled three phase induction motor caused by the hysteresis control using a Neural-Fuzzy Controller (NFC) and keep the switching frequency constant. For this purpose, two-input Sugeno type Neural-Fuzzy Torque Controller (NFTC) is designed using Neural-Fuzzy Network (NFN). The input variables of this controller consists of the torque error and change of the error. Speed and torque control of the induction motor were carried out experimentally using hysteresis based DTC and NFTC based DTC methods at the same sampling period, under the same operating conditions to demonstrate performance of the NFTC. The obtained experimental results are presented comparatively. The control algorithm used in this experimental study were realized by using the controller card, dSPACE DS1103.Parameters of the control system for methods such as the hysteresis bandwidths, sampling period, reference speed and reference flux and the user inputs are introduced into the control system via the software, Control Desk Developer (CDD). For different speed and load conditions, robustness of the NFTC was examined, and based on the obtained results, high performance were observed for both transient and steady states. In addition with this proposed control structure switching frequency of the inverter was kept constant and therefore the problem, varying switching frequency was removed. In the proposed NFTC, bandwidth of the torque ripples of the motor was reduced about % 38.69 and % 47.69 under no load and load conditions respectively, compared to hysteresis based control structure. Also, depending on the effectiveness of the proportional flux controller used for flux control, bandwidth of the flux ripples of the motor was reduced about % 9.30 and % 11.45 under no load and load conditions respectively, compared to hysteresis based control structure.