Bayesian System Identification of Dynamic Structures for Active Vibration Control

Abstract

Bayesian system identification is a remarkable approach used in either modelling or estimation problems, and especially when addressing uncertainty in structural systems. With the increasing use of flexible structures in aerospace engineering, medicine and robot applications, control techniques are often used to carry out such tasks as unwanted vibration mitigation and damage detection. Therefore understanding the dynamics of the nonlinear structure from both a design and a control perspective is an important step.

To the author’s knowledge, limited cases have been made of the Bayesian approaches either in modelling or control in the context of data driven control for vibration control. This is an area with much potential for offering a new perspective on tackling vibration problems.

This thesis seeks to fill this gap in the literature. In doing so, several Bayesian systems identification methods have been proposed, ranging from incorporating well known identification structures, such as Wiener-Hammerstein model and NARX, with GP models to Bayesian state space model. These are then used to inform the design of a new kind of active vibration control, making use of linear and nonlinear structural systems. This thesis concludes by drawing attention to the feasibility of Bayesian methods in active vibration control.