engleski

Robust Self-Learning Fuzzy Logic Servo Control with Neural Network-Based Load Compensator

In this paper, a fuzzy-neural control scheme composed of a sensitivity model-based self-learning fuzzy logic controller (SLFLC) and a neural network-based (NN) load estimator consisting of two off-line trained feedforward neural networks is described. The outputs of the NN estimator have been used to generate a compensation signal whose aim is to increase robustness and to widen the operational range of the SLFLC. Accuracy of NN compensation depends on the correct value of a load compensation gain, but this gain varies with operating point transitions and consequent variations of gain coefficients in the feedforward control path (e.g. a power amplifier gain varies much). A composition of NN-based estimator and the SLFLC has resolved this problem, as a potential inaccuracy of estimation has been accomodated by the learning adaptability of the SLFLC. Experiments performed on the laboratory positioning servo system characterized by the presence of a gravitation-dependent load and fairly high friction have shown that upon applying the NN compensation signal to the output of the SLFLC, position responses have been significantly improved during start of learning and duration of learning has been much shorter than in case without NN compensation.

Self-learning fuzzy logic control, neural networks, nonlinear load compensation, intelligent control, servo systems

nije evidentirano