Neural-network-based adaptive predictive control for vibration suppression of smart structures

A neural-network-based adaptive predictive controller is developed and validated experimentally. On-line nonlinear plant identification is performed using a multilayer perceptron neural network with tapped delay inputs. The performance index includes the squared value of plant response (which is desired to be zero for vibration suppression) and a weighted squared change in the control signal. The one-step ahead prediction of plant response is used to minimize the performance index. Efficient algorithms are used for on-line plant identification and performance index minimization to achieve real-time control of plant with relatively fast response time. Piezoelectric actuators are employed to reduce the vibrations with sine wave and band-limited white noise excitation. Experimental results demonstrate the excellent performance of the developed control system. Adaptive control is verified through similar performances with changes in the plant dynamics and external excitation.