Sensorless Saliency Extraction using Quadratic-Regression-based Current Derivative Estimation

Abstract

Stable field oriented control of induction machines in the proximity of zero electrical frequency relies on the extraction of machine saliencies for rotor flux/positon acquisition. To obtain such saliency information, voltage step excitation methods can be used. They excite the machine with a voltage step caused by the inverter and calculate the resulting phase current derivative, which contains several terms including the superposition of saliency components. Multiple strategies have been used to calculate the current derivative, such as FFT, neural networks, or linear regression. However, they do not take into account the influence of a curvature of the current response. This paper proposes using a least-square quadratic regression to calculate machine saliency information. In this sense, the linear response of the phase current can be accurately isolated from the inherent curvature. It will be proved by experimental measurements that the different saliencies' components are observed in both the second order and also first order term of the quadratic regression function. A performance comparison between linear regression and quadratic regression will be shown in terms of saliency acquisition.