Active Control of Structures Using Neural Networks

Abstract

A new neural-network-based control algorithm has been developed and tested in the computer simulation of active control of a three-story frame structure subjected to ground excitations. First, an emulator neural network has been trained to forecast the future response of the structure from the immediate history of the system's response, which consists of the structure plus an actuator. The trained emulator has been used in predicting the future responses and in evaluating the sensitivities of the control signal with respect to those responses. At each time step of the simulation, the control signal has been adjusted to induce the required control force in the actuator based in a control criterion. A controller neural network has been trained to learn the relation between the immediate history of response of the structure and actuator, and the adjusted control signals. The trained neurocontroller has been used in controlling the structure for different dynamic loading conditions. Results of this initial study indicate that the neural-network-based control algorithms have the promise of evolving into powerful adaptive controllers after further research.