Robust Adaptive Speed Control of Induction Motor Drives

This thesis concerns speed control of current vector controlled induction motor drives (CVC drives). The CVC drive is an existing prototype drive developed by Danfoss A/S, Transmission Division. Practical tests have revealed that the open loop dynamical properties of the CVC drive are highly dependent of the operating point, which is characterised by the speed and load. If the requirements to the controller performance is large, then it is difficult to maintain specified controller performance with a fixed controller, because of the open loop variations. An auto-tuner based on least squares, (LS) identification and generalized predictive control (GPC) has been implemented and tested on the CVC drive. Allthough GPC is a robust control method, it was not possible to maintain specified controller performance in the entire operating range. This was the main reason for investigating truly adaptive speed control of the CVC drive. A direct truly adaptive speed controller has been implemented. The adaptive controller is a moving Average Self-Tuning Regulator which is abbreviated MASTR throughout the thesis. Two practical implementations of this controller were proposed. They were denoted MASTR1 and MASTR2 respectively. Both designs performed well in simulations, but in practice problems with drifting parameters, difference between achieved and designed closed loop dynamics and internal instability occured. MASTR2 was more favourable than MASTR1, because it at least remained stable in the initial experiments. Therefore, only MASTR2 was treated in the rest of the thesis. An analysis of MASTR2 was performed with the purpose of examining its self-tuning properties and the choice of observer dynamics. Practical tests and simulations revealed that the output noise i.e. quantization error and measurement noise in general, were the major reasons for the drifting parameters. Two approaches was proposed to robustify MASTR2 against the output noise. The first approach consists of filtering the output. Output filtering had a significant effect in simulations, but the robustness against the output noise was little in practice. The seccond approach was only to update the controller parameters when excitation in load occurred. This was achieved by incorporating a dead zone in the estimator. This approach had significant effect on the robustness against output noise both in simulations and in practice. A supervisor for MASTR2 was designed. The supervisor was able to handle setpoint changes, lacking excitation and instability. To be able to detect when excitation vanished or when instability occurred, a load disturbance detector and an instability detector was designed. The functionality of the supervisor was validated through practical tests.