Stabilization of an induction machine drive

Abstract: Modern electric trains almost exclusively use inductionmachines fed by voltage source inverters for propulsion. Toreduce external as well as internal disturbances, an LC-filteris inserted between the input terminals of the drive and theinverter. In connection with efficient torque control of themotors, these filters however potentially make the driveunstable as excitations of the filter resonance are amplifiedby the torque control. The solution to such stability problemsusually is to modify the torque reference according tooscillations in the inverter input voltage. Although thisstrategy may be motivated from simple models of the closed loopdrive, structured tuning of this feed-forward compensation ismore difficult. Experience shows that the compensation must notonly be adapted to the varying dynamics of the drive, but alsoproperties of the torque control and applied inverter pulsepatterns must be considered.This thesis derives an expression for the feedforwardcompensation stabilizing the drive in terms of the operatingconditions. These conditions include the motor speed, operatingpoint torque and flux as well the actual DC-link voltage andtime delays of the torque control. From an equivalent feedbackrepresentation of the drive, the stabilization problem is firstinterpreted as appropriately shaping the inverter inputadmittance. The exact shape of the compensation meeting thedesign requirements is then derived using linear models of thedrive, including control. The derived linear models are alsoused to obtain tuning rules for the torque and flux controllerparameters, given requirements on bandwidth and stabilitymargins. It is assumed that torque is controlled using themethod Indirect Self Control.Stability of the closed-loop drive with the proposedstabilization is validated from measurements. Using modelsobtained from frequency domain system identification, stabilityof the non-linear closed-loop drive is verified by combiningstability results for linear systems with the small gaintheorem for the non-linear model errors.