Sensorless Sliding Mode Flux Observer for the Induction Motor with Input Speed Estimate and Gradient Descent Optimization

Sensorless methods for the induction motor (IM) typically need to estimate the speed, fluxes, flux magnitude and field orientation angle of the motor. Observers that estimate the fluxes and speed simultaneously are available but suffer from accuracy issues. This paper presents a sequential estimation approach: the speed is estimated first; then, the speed estimate obtained is fed into a flux observer. Owing to a special gain design, this yields flux estimates that are in phase with the real fluxes (despite the input speed signal which may be inaccurate). The estimated aft fluxes have correct phases; however, they have incorrect magnitudes. Using these fluxes, the tan1 function yields the correct field orientation angle; however, the calculated flux magnitude is inaccurate. To obtain fluxes with the correct magnitude and improve the accuracy, the observer is augmented with a self-adaptation mechanism based on the gradient descent method: this adapts the input speed signal in order to minimize a cost function. The cost function is constructed using the observer's internal variables.