Robust Data-Driven Iterative Learning Control for Linear-Time-Invariant and Hammerstein-Wiener Systems.

Area of Science:

• Control Systems Engineering
• Machine Learning
• System Identification

Background:

• Iterative learning control (ILC) uses output predictors for trajectory determination.
• Robust ILC aims to manage predictor uncertainties and ensure learning convergence.
• Existing methods lack data-driven approaches for stochastic predictor parameter uncertainties.

Purpose of the Study:

• To develop a data-driven ILC method for linear time-invariant (LTI) systems with stochastic predictor errors.
• To extend the data-driven ILC and robust monotonic convergence (RMC) analysis to nonlinear Hammerstein-Wiener (H-W) systems.
• To establish a framework for robustifying ILC using system I/O data.

Main Methods:

• Developed a data-driven ILC for LTI systems, linking predictor matrix errors to stochastic disturbances.
• Analyzed robust monotonic convergence (RMC) in a mean-square sense (MS-RMC) using closed-form expectations.
• Extended the data-driven ILC and MS-RMC analysis to nonlinear H-W systems.

Main Results:

• Established a relationship between predictor errors and stochastic disturbances in LTI systems.
• Demonstrated mean-square sense robust monotonic convergence (MS-RMC) for the closed-loop learning gain matrix.
• Validated the proposed methods through simulations, showing improved convergence and tracking performance.

Conclusions:

• The proposed data-driven ILC effectively handles stochastic parametric uncertainties in both LTI and H-W systems.
• The MS-RMC analysis provides a robust framework for ILC design with identified predictor uncertainties.
• The methods offer enhanced tracking performance uncorrelated with system uncertainties.