On robustness of motion control systems

Robustness and intelligence are becoming increasingly important in motion control systems. In multi-mass electromechanical systems, the estimation of damping capability and design of robust controllers are among very important aspects. In this thesis, conventional control methods coupled with fuzzy logic and neural networks are used to address these issues. First, damping capability of multi-mass electromechanical systems is estimated. The maximal damping and complete damping cases are determined using the generalised model for a multi-mass electromechanical system. To eliminate the load variation influence and reduce elastic vibrations, robust modal control is proposed with observer-based state feedback and feedforward compensation. The use of fuzzy logic dealing with uncertainties is investigated. Good transient performance is obtained, even in the case of changing plant parameters, by fuzzy tuning of the proportional-integral (PI) controller parameters. It is shown that PI controllers with fuzzy tuning can be used in cascade control in a two-mass system. Fuzzy tuning schemes, based on expert knowledge, can be applied to sliding mode control to accelerate the reaching phase and reduce chattering for robustness enhancement. In robust modal control , taking into account uncertainties in the plant parameters and disturbance rate of change, an improvement of observer robustness is achieved via a fuzzy tuning scheme of the predictive coefficient. Insensitivity to load variations is enhanced by continuously tuning the feedforward compensation coefficients. These fuzzy tuning schemes can be applied to robust modal control of multi-mass systems in the presence of uncertainties. Since tuning is a continuous process, exponential membership functions are used. However, with Gaussian or sigmoidal membership functions, similar results can also be obtained. Observer robustness achieved by fuzzy tuning is demonstrated to be suitable for incipient fault detection in dynamic systems. Neural network-based techniques to the problem concerned are also presented. It is shown that a neural net controller can replace the role of a feedforward controller or a fuzzy logic controller. Moreover, a neural net-based controller can be used as a classifier for recognising the error and derivative-of-error patterns, and providing an appropriate control action to improve tracking performance. The proposed controller can be used in a two-mass system without a priori knowledge of the plant. Neuro-fuzzy approach is introduced with a feedforward compensation from an observer based control loop and robust enhancement from a neural network model. Tuning is a human experience to increase robustness. Fuzzy tuning is shown to be efficient thanks to the possibility of adopting this experience. Neurotuning with learning capability will be a subject for further research.