A neural-network-based adaptive predictive controller is developed and validatedexperimentally. On-line nonlinear plant identification is performed using amultilayer perceptron neural network with tapped delay inputs. The performanceindex includes the squared value of plant response (which is desired to be zero forvibration suppression) and a weighted squared change in the control signal. Theone-step ahead prediction of plant response is used to minimize the performanceindex. Efficient algorithms are used for on-line plant identification andperformance index minimization to achieve real-time control of plantwith relatively fast response time. Piezoelectric actuators are employedto reduce the vibrations with sine wave and band-limited white noiseexcitation. Experimental results demonstrate the excellent performance of thedeveloped control system. Adaptive control is verified through similarperformances with changes in the plant dynamics and external excitation.
展开▼