Bridging model-free and model-based optimisation approaches in iterative learning control

Abstract

Systems that operate in a repetitive mode are found across multiple application areas. Iterative learning control (ILC), a high-performance control method, is designed to enhance the tracking performance of such systems by iteratively updating the control input using data collected from previous executions of the tasks.

Existing ILC designs can be classified as model-free and model-based. Model-free ILC methods avoid explicit system modelling, but achieving good convergence performance typically requires careful parameter tuning. Model-based ILC methods can achieve superior convergence performance by utilising a system model, yet obtaining an accurate system model can be difficult or expensive in practice. This thesis addresses this trade-off by developing an optimisation-based ILC framework that aims to recover model-based convergence behaviour under increasingly limited model information.

Within this framework, a stochastic zeroth-order (ZO) optimisation-based ILC algorithm is first proposed for the scenarios where the system model is unavailable. The developed method is shown to achieve asymptotic convergence of the tracking error without the need for deliberate parameter tuning. Motivated by the limited convergence rate in the complete absence of a model, a nominal model-based design is then developed using modifier adaptation (MA). A feasible-side globally convergent modifier adaptation (FS-MA) based ILC is proposed, guaranteeing monotonic convergence to the minimum achievable tracking error, global convergence of the control input to the plant optimal control, and satisfaction of input constraints in the presence of plant-model mismatch. The framework is further extended to handle nonlinear systems via a QP-based ILC implementation, and an accelerated FS-MA ILC variant is proposed to improve the convergence speed while preserving the appealing convergence properties. For each algorithm, a rigorous convergence analysis is provided, followed by simulation examples that demonstrate the effectiveness of the proposed methods.

More information

Identifiers

ORCID for Haonan Shen: orcid.org/0000-0002-9839-0865

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

ORCID for Srinandan Dasmahapatra: orcid.org/0000-0002-9757-5315

Catalogue record

Export record