Modal Filters and Neural Networks for Adaptive Vibration Control

Abstract

An adaptive control algorithm is investigated for the vibration suppression of a space truss structure using modal filters for independent modal space control and a neural network for online system identification. The modal filters are computed off-line using measured frequency response functions and estimated pole values for the modes of interest. They are used to conduct transformation of response measurements from physical coordinates to modal coordinates. The time histories in the modal coordinates are then processed in real time by the neural network to extract estimates of modal parameters, namely, natural frequency, damping ratio, and modal gain. To examine the performance of the adaptive control approach, a controller was designed using the modal filters and implemented on a laboratory space truss using a single reaction-mass actuator and 32 accelerometers. The performance of the modal filter-based controller is compared to that of a local rate feedback controller using the same actuator. The applicability of the neural network to adaptive control was demonstrated by real-time estimation of the modal parameters of the truss with and without control. Because the modal filter control gain can be adjusted to maintain a desired closed-loop damping ratio, which is tracked by the neural network, adaptive control of individual modes in a time-varying system is possible. Eventually, this type of adaptive controller will help develop a control system that can maintain desired closed- loop performance characteristics under significant modal parameter variations.