Neural network-based structural vibration control system

Abstract

This dissertation proposes three types of neural network-based structural vibration control systems; active modal neuro-control system, semiactive modal neuro-control system, and hybrid neuro-control system. Especially the semiactive modal neuro-control system using a Magnetorheological (MR) damper for vibration suppression of the building structure subjected to ground motion has been experimentally investigated by a shaking table test. Also the hybrid neuro-control system, which combines neuro-controller and base isolation system, is applied to a base-isolated benchmark problem adopted by American Society of Civil Engineering (ASCE) control committee. The modal neuro-control system uses modal coordinates as inputs of neuro-controller. Therefore the modal neuro-control scheme is more simple and convenient than conventional neuro-control scheme which needs troublesome task for determining specific state variables as inputs in neuro-controller. Through numerical example employing hydraulic actuator as a control device, the performance of the modal neuro-control system for building structure is investigated. A Kalman filter is included in a control scheme to estimate modal states form measurements by sensors, and a low-pass filter is employed to eliminate the spillover problem. The semiactive modal neuro-control system which combines the modalneuro-control algorithm and semiactive control device (MR damper) is developed and verified in experiments with scaled shear building model. The clipped algorithm which induces the MR damper to generate approximately the desired control force is adopted in the proposed system. The performance of the semiactive modal neuro-control system is compared with that of passive-off, passive-on and passive-optimal systems. The hybrid neuro-control system contains semiactive neuro-controller using MR damper and base isolation system. Seismic base isolation is a technique that mitigates the effects of an earthquake by essentially is...