System Identification of Smart Structures Using Neural Networks

Abstract

A composite smart structure embedded with piezoelectric sensor/actuator is fabricated and its piezoelectric effectiveness is validated by static test and modal testing. Instead of identifying the structural parameters of the smart structure, system identification is conducted by estimating the connective weights of a backpropagation neural network with an adaptive learning rate. Experimental verification shows that the [6-7-2] neural network, a neural network with 6 input neurons, 7 hidden layer neurons, and 2 output neurons, is capable of representing the system dynamics both in time domain and in frequency domain. In addition, it is fault tolerant. All simulations are validated by experiments. Integration of smart structure and neural network not only avoids the complexity introduced by other traditional analytical or computational methods but also lays the corner stone for effective neural controller design.