INVESTIGATION ON NONLINEAR MODELING TECHNIQUES FOR SWITCHED RELUCTANCE MOTORS AND ULTRASONIC MOTORS

Abstract

The Switched Reluctance Motor (SRM) has high nonlinear dynamic characteristics. To minimize torque ripples ? an efficient algorithm is needed to model this nonlinear behavior accurately. Nonlinear fitting of flux linkage is performed and eventually extended to torque. This nonlinear model involves nonlinearity in its parameters, so existing linear methods not applicable. Curve fitting Levenberg Marquardt method, Neural Network and a combination of the two, Parallel Network, are used to model the flux-linkage characteristics. The novel parallel network achieved better results in terms of computation speed and mean square error for flux modeling. However, for the purposes of torque estimation for torque control of the SRM, the neural network is found to be the most suitable. Parallel network does not fare better due to the large errors in torque derivation in the Marquardt method. Interpolation Property of Neural Network is also tested. The back-propagation algorithm in Neural Network has good interpolation property. The ultrasonic motor is modeled to achieve a closed-loop position control with a precision of lOOnm. A quadratic equation is used to model the voltage-velocity relationship. This open-loop procedure gave a precision of 100//m. This does not give the required accuracy and needs further investigations.