Neurocontrol of industrial motion systems with actuator nonlinearities.

摘要

The past decade has witnessed a growing body of experimental work showing that neural networks (NN) and fuzzy logic systems can result in improved performance in complex feedback control problems. Since it is already well established that neural and fuzzy systems are adept at solving pattern recognition problems, as well as at providing approximate solutions to complex optimization problems, they are particularly suited for improving performance in complex nonlinear systems in the presence of large uncertainties. In applications such as VLSI circuit design and elsewhere, the requirements on the speed and precision of motion are increasing. Such electro-mechanical systems are characterized by complex dynamics having unmodeled nonlinearities, time delays, high-frequency actuation dynamics, disturbances, friction, and actuators with deadzone and backlash. The control problems associated with such complex systems are not easy, as they do not satisfy most of the assumptions made in the controls literature, including linearity in the parameters, feedback linearizability, and the model matching condition.; Different techniques for compensation of these nonlinearities have been developed in this dissertation. Standard techniques such as proportional-derivative controllers result in limit cycles if the system has deadzone or backlash. Adaptive control schemes require additional and unrealistic assumptions on actuator nonlinearities that do not always hold. NN controllers are adaptive learning systems, but they do not need usual assumptions made in adaptive control theory such as linearity in the parameters and availability of a known regression matrix.; This dissertation presents a series of controllers based on NN for compensation of nonlinearities appearing in industrial and automotive motion systems, including friction, deadzone, and backlash. The standard multilayer perceptron NN has been modified in order to approximate piecewise continuous functions of the sort that appear in friction, deadzone, backlash, and other motion control actuator nonlinearities. A rigorous mathematical proof of the approximation property for piecewise continuous functions is given as well as simulation of the modified NN.; NN compensators for industrial systems with actuator nonlinearities have been shown extremely effective in correcting motion inaccuracies arising from these nonlinearities. As “intelligent control systems” they are capable of learning and adapting on-line to more effectively control the system. It is shown how to design NN controllers for systems with actuator nonlinearities, demonstrate that they give closed-loop stability and guaranteed performance, and show the simulation results on different dynamical systems. In the case of backlash compensation, two different tuning algorithms are presented: modified backpropagation tuning law and modified Hebbian tuning law.; Nonlinear system identification using radial basis function NN is presented. The state estimation error is proven to converge to zero asymptotically. Parameters of the identifier converge to the ideal parameters provided that persistency of excitation condition is fulfilled. The multiple models identification structure, and its application to the multimodel failure detection is analyzed.