Distributed iterative learning control for networked dynamical systems

Abstract

Networked dynamical systems have found increasingly more applications during the last
few decades, thanks to the significant reduction in the cost of sensing, computing and actuating technologies. Among them, there exists a class of networked dynamical systems
working in a repetitive manner and requiring high control performance. As an example,
next generation advanced manufacturing contains a large number of subsystems working
together to perform a variety of manufacturing tasks repeatedly with high performance
requirements. For such systems, traditional control methods have significant difficulties
meeting the high performance requirements: centralised design does not scale well, while
distributed methods mainly focus on asymptotic behaviour. In addition, they all require
a highly accurate model which can be difficult/expensive to obtain in practice.
Recently, iterative learning control (ILC), which ‘learns’ from the input and error information of the previous attempts of the same task without requiring an accurate model
to generate the input, has been proposed as an alternative solution. However, most of
the existing ILC design for networked dynamical systems have poor scalability, limited
convergence performance, and also lack of the ability to deal with system constraints
and more general task, e.g., point-to-point (P2P) task. To address these limitations, this
thesis proposes novel distributed/decentralised optimisation-based ILC design methods.
This thesis considers three design problems for networked dynamical systems, i.e., consensus tracking, formation control and collaborative tracking. We propose optimisation
based ILC design methods using the idea of norm optimal ILC, and the proposed ILC
methods show appealing convergence properties and certain degree of robustness to
model uncertainties. Using the alternating direction method of multipliers (ADMM),
all the designs can be implemented in the distributed/decentralised manner such that
only local information is needed, allowing the proposed algorithms to be applied to large
scale networked dynamical systems and have great scalability for dynamically growing
network. These algorithms can also be extended to solve two unexplored problems in
ILC design for networked dynamical systems, namely, constraint handling and P2P task.
Numerical examples are given to illustrate the performance of the proposed algorithms

More information

Identifiers

ORCID for Bing Chu: orcid.org/0000-0002-2711-8717

Catalogue record

Export record