An Optimal Neural Network Design Methodology for Fast Convergence and Vibration Suppression

Abstract

An optimal neural network design is to select the network configuration and the learning rule for fast convergence and best system performance. This paper develops a network design methodology so that the optimal design can be determined in a systematic way. The methodology combines the Taguchi method of quality engineering and the back-propagation network with an adaptive learning rate for their advantages in implementation feasibility and performance robustness. Vibration suppression experiments of a composite smart structure with embedded piezoelectric sensor/actuator validate that the methodology provides an efficient neural controller design, including the plant order, the number of hidden layer neurons, the number of training patterns, and the coefficients of adaptive learning rate.