Iterative learning control implemented on a multi-axis system

Abstract

This thesis concerns the implementation and comparison of different Iterative Learning Control (ILC) strategies on a multi-axis gantry robot.  The majority of ILC research focuses on developing new algorithms for different classes of plant, then providing, by undertaking rigorous mathematical and simulation based studies, that the new algorithm will meet performance and stability requirements. The work presented here strictly concerns the performance of different ILC strategies on a physical plant by experimental methods alone, demonstrating that ILC can successfully be implemented in industrial applications.  A test facility consisting of a three axis gantry robot and associated peripheral hardware is designed and built for this purpose.  Four tests are developed to investigate key issues which are of particular importance to ILC implementation, minimum achievable tracking error, long-term stability, robustness to initial state error and robustness to plant modelling error. A rigorous framework for measuring quantitatively the performance of different algorithms is established to allow fair comparison.  The experimental analysis is divided into two categories:  basic and model-based algorithms.  The work relating to basic algorithms initially uses a standard three-term PID controller to establish a benchmark performance level, to which the ILC performance can be compared.  Combining a basic Proportional (P-type) ILC algorithm with the conventional PID controller to form a hybrid is found to increase significantly the performance, but at the expense of long-term stability.  To remedy this, a logical progression of different filtering techniques, band-stop, low-pass, zero-phase-low-pass and a new frequency aliasing method are applied to the hybrid controller to steadily improve long-term stability and subsequently tracking performance. The work relating to model-based algorithms compares the performance of three previously developed, optimal ILC algorithms:  adjoint ILC, inverse ILC and norm-optimal ILC.  These algorithms operate on the plant input signal alone, without the requirement for an additional feedback controller.  The three algorithms are found to produce significantly different tracking performance in response to disturbances and modelling error.  The thesis concludes with an analysis of the tracking performance generated by each algorithm and a general discussion summarising algorithm attributes.  For the plant used in this experimental work, it is found that the basic P-type algorithm has a slower convergence rate, but can achieve tracking error levels similar to more advanced model-based techniques.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata