A Neural Network for Feedforward Controlled Smart Structures

Abstract

A neural network and adaptive algorithm suitable for driving non-linear feedforward control systems used with "smart" structures to actively control sound and vibration is presented. The neural network/algorithm combination can be viewed as an extension to the commonly used transversal filter/filtered-x LMS algorithm combination, in that the neural network is essentially a transversal filter with a non-linear hidden layer placed between the input delay chain and output ac cumulator. The algorithm accommodates the transfer function and time delay between the control signal output and error signal input. The arrangement is shown in simulation to be capable of sup pressing a primary distrubance, which is a non-linear function of the reference signal fed to the con trol filter, and of suppressing harmonics introduced into a system by a control actuator exhibiting non-linear performance characteristics. The results are compared to those similarly obtained using a linear adaptive control arrangement, which is incapable of providing attenuation in either case.