Gradient descent optimisation for ILC-based stochastic distribution control

A novel iterative learning control (ILC) algorithm was developed and applied to an active flow control problem. The technique uses pulsed air jets to delay flow separation on a two-element high-lift wing. The ILC algorithm uses position-based pressure measurements to update the actuation. The method was experimentally tested on a wing model in a 0.9 m × 0.6 m low-speed wind tunnel at the University of Southampton. Compressed air and fast switching solenoid valves were used as actuators to excite the flow, and the pressure distribution around the chord of the wing was measured as a feedback control signal for the ILC controller. Experimental results showed that the actuation was able to delay the separation and increase the lift by approximately 10%–15%. By using the ILC algorithm, the controller was able to find the optimum control input and maintain the improvement despite sudden changes of the separation position.

In order to improve the pulp quality and to reduce the energy consumption, the fiber length distribution (FLD) is generally employed as one of the important technological indexes in the refining process. Considering that the traditional mean and variance of fiber length are unable to adequately characterize the non-Gaussian distribution properties, this paper proposes a novel geometric analysis based double closed-loop iterative learning control (ILC) method for probability density function (PDF) shaping of output FLD in the refining process. Primarily, a radial basis function (RBF) neural network with Gaussian-type is utilized to approximate the square root PDF in the inner loop, where the RBF basis function parameters (center and width) are tuned between any two adjacent batches by using an ILC law, and the subspace identification method can be applied to establish the state-space model of weight vector. Then, for the sake of accelerating the convergence rate of the closed-loop system, a geometric analysis based ILC method is adopted in the outer loop. Finally, both simulation and experiments demonstrate the effectiveness and practicability of the proposed approach.

This paper presents model reference adaptive control (MRAC) approach to control the shape of output distribution in non-Gaussian stochastic systems. The method is based on Iterative Learning Control (ILC) and employs a neural network framework for controller design. The output probability density function (PDF) tracking problem is first reduced to dynamic neural network (NN) weight control. It is assumed that the dynamic behaviour of such weights is nonlinear and unknown. To apply the ILC-based tuning, the control horizon is split up to certain number of intervals hereinafter called batches. The proposed ILC method is comprised of two main modes, namely within each batch and between any two adjacent batches, and includes three stages; (a) NN-based nonlinear dynamic system identification ( b) MRAC of the weight control loop within each batch, and (c) Tuning the RBF centers, widths, and controller neural network parameters between any two adjacent batches. Simulation results confirm the effectiveness of the method.

An estimation-based iterative learning control (ILC) algorithm is applied to a realistic industrial manipulator model. By measuring the acceleration of the end-effector, the arm angular position accuracy is improved when the measurements are fused with motor angle observations. The estimation problem is formulated in a Bayesian estimation framework where three solutions are proposed: one using the extended Kalman filter (EKF), one using the unscented Kalman filter (UKF), and one using the particle filter (PF). The estimates are used in an ILC method to improve the accuracy for following a given reference trajectory. Since the ILC algorithm is repetitive no computational restrictions on the methods apply explicitly. In an extensive Monte Carlo simulation study it is shown that the PF method outperforms the other methods and that the ILC control law is substantially improved using the PF estimate.

Iterative learning control (ILC) algorithms appeared in connection with the problems of increasing the accuracy of performing repetitive operations by robots. They use information from previous repetitions to adjust the control signal on the current repetition. Most often, information from the previous repetition only is used. ILC algorithms that use information from several previous iterations are called higher-order algorithms. Recently, interest in these algorithms has increased in the literature in connection with robotic additive manufacturing problems. However, in addition to the fact that these algorithms have been little studied, there are conflicting estimates regarding their properties. This paper proposes new higher-order ILC algorithms for linear discrete and differential systems. The idea of these algorithms is based on an analogy with multi-step methods in optimization theory, in particular, with the heavy ball method. An example is given that confirms the possibility to accelerate convergence of the learning error when using such algorithms.

On reference governor in iterative learning control for dynamic systems with input saturation

A robotic Ping-Pong server system could play an important role in skill training for players. Current available systems in the market provides different modes with rough control of the ball landing point. In this paper, we present the design and implementation of a robotic Ping-Pong server system to achieve precise ball landing point. A cost-effective multimodal landing point detection system is proposed to detect the landing point by fusing video and audio signals. A lookup table is used to relate the server launching parameters with the landing points. To keep the lookup table dynamic, an iterative learning control (ILC) algorithm is adopted to automatically update the lookup table in case of environmental changes. The ILC algorithm is tested and validated in simulation. Results showed the system was able to precisely serve the ball to the desired point automatically.

In this chapter, it is first observed that, when a typical iterative learning control(ILC) algorithm is applied to a class of linear dynamic systems with time-delay, erratic estimation of delay time may cause the control input to diverge. In order to resolve such a difficulty due to uncertainty of the delay time, a new ILC algorithm is proposed, in which the holding mechanism is adopted to hold the control input at a constant value for the duration of the delay time uncertainty. As a consequence, the output of the system tracks a given desired trajectory at discrete time-points which are spaced by the size of the uncertainty of delay time. In addition to the consideration of the delay time, the effect of the initial state error is also studied with a modification of the proposed ILC algorithm. Numerical examples are given to show the effectiveness of the proposed algorithms.

Iterative learning control (ILC) algorithms are typically used to improve the performance of repetitive processes. Numerous successful applications of ILC, such as computer numerical control (CNC) machining processes, robot manipulation, and lithography processes, have been reported. However, ILC often exhibits less than satisfactory performance since the control unit operates at a limited sampling rate in consideration of cost. In this study, the multiloop, multirate structure of servo motor control systems is taken advantage of and a fast-update multirate ILC (FILC) scheme is proposed for high-accuracy trajectory tracking, where compensation is updated at a rate faster than that of the original position loop without the need for redesigning the feedback controller. The key difficulty related to the design of the FILC lies in the extremely complex and time-varying dynamics inherent in multirate systems. To address this problem, a novel equivalent single-rate parametric model description of the multirate system is derived, which enables the use of the efficient norm optimal ILC algorithm. Consequently, a computationally efficient FILC is obtained to improve the performance. Subsequently, the proposed FILC is applied to the position control of the CNC motion stage. Simulation and experimental results are used to verify the effectiveness of the proposed method.

Lots of iterative learning control (ILC) algorithms have been developed to improve tracking performance of systems that track same trajectory repetitively. In this paper, we use an open-closed-loop PD-type ILC scheme for the control of a class of nonlinear systems. Sufficient condition for guaranteeing the convergence of the learning scheme is given under a few reasonable assumptions, and then proved with linear operator theory. It is shown that the proportional coefficients of the ILC scheme is not a factor for guaranteeing the convergence, the condition is weaker than known results for similar ILC schemes.

In this paper, it is proposed a discrete-time multivariable iterative learning control (ILC) algorithm which guarantees the monotonic convergence of the tracking error norms along with the number of the trials. It is pointed out that an interactor matrix plays an important role in the iterative learning control. Since ILC is a design method not requiring plant parameters explicitly, the information on the interactor matrix should be treated same manner as in an adaptive control systems. In this paper, it is presented a design of ILC using G-interactor which is derived not requiring the plant parameters. It is also presented a design of ILC using a diagonal precompensator which will make the interactor of the compensated plant be diagonal. Numerical simulations are presented to confirm the validity of the proposed design.

An iterative learning control algorithm within prescribed input-output subspace

Newton-method based iterative learning control method for nonlinear systems

There are a vast number of manufacturing applications that are repetitive in nature and therefore can benefit from Iterative Learning Control (ILC) algorithms. However, some of these applications are unfit for continuous open loop signal updates from ILC either because the complete manufacturing cycle includes abrupt transitions in system dynamics or is prohibitively long for efficient implementation. This paper explores a method to control one such system, Micro Robotic Deposition (μRD), using ILC as an open loop control signal identification technique. Instead of continuously updating the ILC control signal for the complete operation, we exploit the characteristic that all μRD cycles are a sequence of a few basis tasks and only these basis tasks are learned. Control signals for these basis tasks build a library of basis signals, which can then be appropriately sequenced as the control signal for the complete manufacturing cycle. This paper introduces a method to build this basis signal library and extract and coordinate the signals depending on predefined μRD operations and material used as specified by numerically controlled machine language. The methods applied to μRD display the ability to drastically improve end product quality with a significantly shortened signal identification process.

In this paper, a novel iterative learning control (ILC) algorithm is presented for a class of joint position constrained robot manipulator systems. Unlike the traditional ILC probelms, where the reference trajectory is iteration invariant, the reference trajectory in this work can be non-repetitive over the iteration domain. A tan-type time-varying Barrier Lyapunov Function (BLF) is proposed to deal with the constraint requirements which can be both time and iteration varying. We show that under the proposed ILC scheme, uniform convergence of the full state tracking error beyond a small time interval in each iteration can be guaranteed over the iteration domain, while the constraints on the joint position vector will not be violated during operation. An illustrative example is presented in the end to demonstrate the effectiveness of the proposed control scheme.