Identification and Speed Control Design of Resonating Mechanical Systems in Electric Drives

Ac electrical machines supplied with a frequency converter have been increasingly selected for torque actuators in modern motion control applications. These applications often contain several moving or rotating masses connected together with mechanical transmission components. This configuration results in mechanical resonances and, if the speed-feedback loop is delayed, instability can even occur. To overcome the resonance problems, several speed control methods and different tuning approaches have been proposed: They range from the gain decreasing of a simple proportional integral (PI) controller to the use of complex nonlinear controllers. This dissertation provides methods to analytically tune the speed controller of electrical drives. The tuning rules are given for both a rigid single-mass system and a resonating two-mass system. The speed controller is designed to give both the robust regulation performance as well as accurate reference tracking. The controller gains are always parametrized in the parameters of the mechanical system and in some design specifications, which results in the possibility of applying gain scheduling if required. The dissertation demonstrates on an experimental level that an effective and robust classical speed controller can be designed for both the rigid single-mass system as well as the resonating two-mass system. However, the controlled mechanical-system model must be known and the controller must be tuned using the known model. Finding a suitable mechanical model and its parameters may sometimes be problematic. Not all of parameters are always available or else the datasheet values may not be accurate. To overcome the problems of finding the mechanical parameters, this dissertation provides methods to identify the mechanical load and its parameters. The identification can be done both in the open-loop and closed-loop operations and it is based on using discrete-time polynomial models instead of frequency-domain methods. The experiments established that the polynomial-model based, discrete-time method is a better choice, than the frequency-response-based method. This is mainly because of the absence of the time domain to the frequency domain conversion, which adds an additional computational burden and numerical inaccuracy to the identification method. The experiments further demonstrate that the proposed identification method can be successfully applied both in the open-loop and in the closed-loop configurations.