Robust Parameter and Signal Estimation in Induction Motors

This thesis deals with theories and methods for robust parameter and signal estimation in induction motors. The project originates in industrial interests concerning sensor-less control of electrical drives. During the work, some general problems concerning estimation of signals and parameters in nonlinear systems, have been exposed. The main objectives of this project are: - analysis and application of theories and methods for robust estimation of parameters in a model structure, obtained from knowledge of the physics of the induction motor. - analysis and application of theories and methods for robust estimation of the rotor speed and driving torque of the induction motor based only on measurements of stator voltages and currents. Only contimuous-time models have been used, which means that physical related signals and parameters are estimated directly and not indirectly by some discrete-time approximation. All methods and theories have been evaluated on the basis of experimental results obtained from measurements on a laboratory setup. Standard methods have been modified and combined to obtain usable solutions to the estimation problems. The major results of the work can be summarized as follows: - identifiability has been treated in theory and practice in connection with parameter and signal estimation in induction motors. - a non recursive prediction error method has successfully been used to estimate physical related parameters in a continuous-time model of the induction motor. The speed of the rotor has been used as a third input. - the rotor speed and the driving torque of the induction motor have successfully been estimated based on measurements of the terminal quantities only. The following methods have been applied: a recursive prediction error method and a method based on a steady state model of the induction motor. - the recursive prediction error method has been robustified against parameter variations by extending the method with a selective forgetting of past data. - bias free estimates have been obtained by using a output-error model structure. - the output error model structure has been used to overcome the larger number of time varying parameters in the innovation model.