This article addresses the problem of input saturation in iterative learning control (ILC), a problem area that has seen relatively little research effort. A new design is developed based on a branch of 2-D systems theory and, in particular, Lyapunov-based stability theory. The resulting design is compared in simulation with an alternative. In the final part of this article, the results of an in-depth experimental investigation of the new designs properties are reported.
The paper develops a new iterative learning control (ILC) law for a system in which the reference trajectory and system parameters change between passes in a known manner, the control plant is subjected to random disturbances, and measurements are carried out with noise. The design is based on the recently developed stochastic stability theory of repetitive processes, that are a class of 2D systems. This theory uses vector Lyapunov functions and their divergence to obtain stabilization conditions. In this paper the proposed method of vector Lyapunov functions is used in combination with Kalman filtering. For reducing the transient error caused by the reference trajectory change, the ILC law switches, depending on this change. A supporting example is given, demonstrating that this law yields a significant reduction of the transient error, that generated by the reference trajectory change.
11 August 2025 in IEEE Transactions on Automatic Control
This paper develops iterative learning control designs to handle actuator nonlinearities that can occur during implementation. Unlike existing designs, the new results can be adapted to several commonly encountered nonlinearities. The application of the new design is highlighted through a detailed case study based on a model for the dynamics of a robot system constructed from measured frequency response data. In particular, this case study shows that the control law developed can accelerate error convergence and compensate for the effects of nonlinear actuator dynamics.
Iterative learning control (ILC) applies to systems that repeat the same finite-duration task repeatedly, where each repetition is termed a trial and the finite duration is termed the trial length. The control method involves specifying a reference trajectory, calculating the error at each trial, and designing a control law to drive the error sequence toward convergence as the number of trials increases. This paper considers the ILC design problem for stochastic discrete linear systems with multiple reference trajectory switching and parameter variations, using a combination of optimization and vector Lyapunov functions. The design also includes compensation for the temporary increase in learning error caused by switching the reference trajectory and parameter variations. A numerical example demonstrates the properties of the new design and includes a brief comparison with an alternative design.
Repetitive processes complete the same operation defined over a finite interval of time over and over again. Once an execution, termed a pass, is complete the processes resets to the starting location. Once complete, the next pass can begin, either immediately after the end of resetting or after a further period of time. Each pass output is known as the pass profile and on any pass this profile contributes to the dynamics of the next pass profile. This can result in oscillations that increase in amplitude from one pass to the next and so on. These processes are a particular case of 2D systems and the appearance of oscillations in the sequence of pass profiles cannot be prevented by application of standard control action. In this paper new results on control law design for linear dynamics are developed using a passivity based stability theory setting developed for nonlinear dynamics that requires the use of vector Lyapunov functions. As one possible application, the new results are applied, with a supporting example based on model of Quanserr flexible link , to iterative learning control design.
1 January 2026 in IET Control Theory & Applications
Iterative learning control (ILC) applies to systems where the same finite‐duration task is completed repeatedly or where there is a stoppage time between finite‐duration tasks. Each completion is commonly termed a trial, and the control law can use all data generated on previous trials. Given a supplied reference trajectory, the error on any trial is the difference between this trajectory and the trial output. The design problem is to find a series of control signals, one for each trial, that will reduce the error from trial to trial. The trial‐to‐trial error convergence rate is a critical part of an ILC law design. A desirable feature is obtaining the fastest possible convergence rate consistent with an acceptable dynamic response along each trial. This paper considers an ILC design problem for discrete systems with changing the reference trajectory and parameters under input saturation. A new approach is proposed to accelerate convergence, based on a combination of the gradient optimization method and the authors' method of vector Lyapunov functions for repetitive processes. The ILC law switches depending on the reference trajectory change for reducing the transient error caused by this change. A supporting example is given, demonstrating the effectiveness of the design proposed.