Diz 59 en
Distributed Estimation and Control With Applications to Spatially Distributed Camping Systems
Author: Xueji Zhang
Lightweight structures are increasingly installed in aerospace and automotive industries, stimulated by economic constraints and stringent emission standards. These lightweight structures are usually lightly damped and therefore prone to vibrating. Structural vibrations are detrimental to the safety and life-cycle of the mechanical structures, calling for effective vibration reduction techniques. In parallel, rapid advances and integration of computing, communication and smart sensing technologies have motivated deployment of small-size, low-cost sensing devices equipped with embedded processors and communication capabilities. The emerging networked control systems (NCS) provide a promising design paradigm for vibration control algorithms. In this envisioned paradigm, decisionmaking process is delegated to intelligent agents which are facilitated by an array of actuators and sensors deployed throughout the structures. This dissertation is dedicated to adapting the networked or distributed control concept specifically for vibration reduction of spatially distributed damping systems, i.e. flexible structures. Four decentralized approaches for distributed/cooperative observers over directed graph topology are developed. The first approach assumes that the information of graph topology is perfectly known to each agent. The second approach works independent of any specific graph topology and is only for locally detectable systems. The third approach does not require the exact information of graph topology as well, and can work for both locally detectable and undetectable systems. In the fourth approach, an observability decomposition is firstly applied locally at each agent, and parameters of the cooperative observers are designed in observable and unobservable subspaces, respectively. This approach can also be developed without the exact information of the graph topology. In particular, compared with centralized approaches in the literature, the four decentralized approaches have a few appealing features. Firstly, it is robust, to a certain degree, against graph reconfigurations. Secondly, the decentralized approaches have flexibility in integrating redundant sensors into the network. Thirdly, the computational complexity in designing the variables is reduced. The performance of proposed cooperative observers and controllers is examined with numerical simulations, where a smart flexible beam is considered. An experimental case study with a piezoelectric actuated composite plate is presented to validate the developed algorithms in real-time. This dissertation also performs a preliminary research on distributed homogeneous sensor fusion over the network. Inspired by the Bayesian-based fusion, variances of measurement noises are incorporated into the edge weight design of the network. The fusion performance is examined with numerical experiments. The steady-state expected values and variances of the state estimates of all the nodes in the network agree well with the analytical results.
- Xueji Zhang, mailto:zhangxue@fel.cvut.cz