ANN Based Flux and Torque Estimator for Vector Controlled Induction Drive with Dynamic Loading

Abstract

Vector control of induction motor drives is very useful technique to obtain high performance speed response. Implementation of vector control requires the knowledge of instantaneous magnitude and position of rotor flux and load behavior in the shaft for correct dimensioning of an induction motor. This paper demonstrates estimation of feedback signals in a three phase induction motor derive under dynamic load conditions using Artificial Neural Network (ANN). A instantaneous magnitude and position of rotor flux and load torque estimator, in the form of a two stage neural network is presented. The neural network is able to accurately estimate these parameters of an induction motor. The observations indicate that ANN based flux and load torque estimation may be a feasible alternative to other rotor flux and load torque estimation methods like programmable DSP kit. The comparative performance of both has been presented in this work with the help of a practical three-phase induction motor drive. Index Terms—Back propagation, DSP, FOC, Induction drive, Neural network, Vector control. I. INTRODUCTION In order to control an induction motor requiring high dynamic performance, an accurate knowledge of the instantaneous magnitude and position of rotor flux and load torque is necessary because these directly affect the mechanical power developed. Application of a motor depends upon the load behavior of the motor; therefore accurate estimation of these parameters is a difficult task for a increase the overall cost of the machine. However, with the introduction of indirect FOC the hardware requirements are much simpler, resulting in better overall performance. In order for such a scheme to work, the accurate estimation of instantaneous magnitude and position of rotor flux and load torque is vital. There are a few schemes available today, most of which are based on adaptive control (5)-(8), whereas others use digital signal processing (DSP) for their estimator implementation (9)-(14). In this paper an artificial neural network is described, as used for the estimation of the instantaneous magnitude and position of rotor flux and load torque for several operating points during line-start operation of an induction motor. The design of a two hidden layer neural networks is discussed in this work. The learning requirements of the design are evaluated by developing the back propagation learning technique for the instantaneous magnitude and position of rotor flux and load torque estimator.In balanced three-phase systems, the two axis (d-axis and q-axis) model is used for dynamic modeling of an induction motor (15). In implementatio n of vector control technique, also known as FOC shown in Fig.1, the three phase induction motor works as a two phase motor in which ids is in the direction of flux ψr and the iqs perpendicular to it. This alignment in stationary d-q model (regarded as d s -q s ), an induction motor can be expressed in either a stationary or a rotating reference frame. In stationary reference frame, the reference d and q axes are fixed on the stator. In synchronous rotating ( d e -q e ) frame, rotating at synchronous speed with respect to stationary frame (d s -q s ) at any instant, the angular position of the d e -axis with respect to the d s -axis is θe, where