In this paper, a neural network-based adaptive iterative learning control scheme is developed to solve the trajectory tracking problem for rigid robot manipulators with arbitrary initial errors. Time-varying boundary layers are used to relax the zero initial error condition which must be observed in traditional iterative learning control design, and adaptive learning neural networks are constructed to approximate uncertainties in robotic systems, whose optimal weights are estimated by using partial saturation difference learning method. For arbitrary bounded initial state errors, the tracking error of robot manipulators will asymptotically converge to a tunable residual set as the iteration number increases. An illustrative example and the comparisons are provided to demonstrate the effectiveness of the proposed neural network-based adaptive iterative learning control scheme.
20 January 2023 in IEEE Transactions on Circuits and Systems II: Express Briefs
In this paper, a novel definition describing a switching steady-state motion, such that the tracking error always remains within a steady-state error domain, is introduced. Both the dead-beat reaching law and discrete terminal sliding mode function are proposed and applied to the design of the dead-beat terminal sliding mode controller which drives the tracking error to a steady-state error domain. In order to characterize the attractiveness and invariance properties of the controlled system, the upper/lower bounds of the steady-state error domain and the bound on absolute attractive layer of the tracking error are systematically derived. Finally, both numerical simulation and experimental results on the dc-dc buck converters in three different cases, i.e., the load resistance variation, the input voltage variation, and the desired output voltage variation, confirm the usefulness of the proposed method.
25 April 2022 in IEEE Transactions on Circuits and Systems II: Express Briefs
This brief presents the development of a dead-beat terminal sliding mode control strategy, based on the smooth terminal sliding function and the dead-beat reaching law (RL). By means of the dead-beat RL, the sliding mode occurrence is always guaranteed and the desired tracking error dynamics can be achieved by the chosen smooth terminal sliding function. A finite-order backward-difference uncertainty compensator is incorporated into the controller design to improve the uncertainty rejection capability. The steady-state error band is analyzed in detail, for characterizing the attractivity and steady-state tracking performance with respect to control parameters. Effectiveness of the proposed control strategy is illustrated by simulation results on a DC-DC buck converter.
In the open-end winding permanent magnet synchronous motor (OEW-PMSM) with common DC link, the common-mode voltage (CMV) will cause leakage current and zero-sequence current, which will lead to the decrease in the system life and efficiency. To solve this problem, the loop characteristics of leakage current and zero-sequence current were analyzed, and the condition for eliminating the leakage current and the zero-sequence current was deduced. Then, the CMV of the voltage vectors for the OEW-PMSM system was calculated, and the appropriate voltage vectors satisfying the conditions were selected to form the control vector sets. Combined with the model predictive torque control (MPTC), a cost function without the weight factor was proposed. The voltage vector sets were predicted by the cost function. The optimal voltage vectors were selected to control the OEW-PMSM, which can eliminate the leakage current and the zero-sequence current caused by the CMV. The effectiveness of the proposed method was verified by the simulation results.
In this paper, the tracking problem for a class of nonlinearly parametric time-delay systems with arbitrary initial state errors and iteration-varying reference trajectories is studied. An adaptive iterative learning control scheme is developed by using Lyapunov approach. A time-varying boundary layer is constructed for removing the zero initial error condition, which is necessary in normal iterative learning control algorithms. The time-varying nonlinearly parametric uncertainties and time-delay uncertainties are well compensated by using parameter separation technique, signal replacement mechanism and robust strategy. As the iteration number increases, the state tracking error asymptotically converges to a tunable residual set. In the end, a simulation example further verifies the theoretical results.