Session: 03-20-02: Manufacturing: General

Paper Number: 144463

144463 - Advanced Registration Control in Roll-to-Roll Printing: Spatial-Terminal Iterative Learning Strategies 

Roll-to-Roll (R2R) systems, featuring a series of motorized or idle rollers, are pivotal for high-volume, continuous production of flexible substrates across various industrial applications, including the manufacturing of electronics for wearable devices and coating for high-performance batteries. One of the significant challenges in R2R printing processes lies in maintaining tight alignment tolerances, as specified by the device resolution for multi-layer printed electronics. The alignment of the printed patterns in different layers is known as registration. The precision of registration is critical to ensure product quality for R2R processes. However, the inherent deformability of flexible substrates, coupled with the complicated interaction dynamics between substrates and rollers result in significant difficulties for high-precision registration control in R2R systems. Registration errors are essentially positional or dimensional deviations caused by undesired variations in substrate tensions and substrate speeds. Despite the employment of real-time feedback controllers such as Proportional-Integral-Derivative (PID) or Model Predictive Control (MPC) for R2R tension control, they are not responsive enough to transient disturbances and they fail to continually enhance performance in the presence of angle-periodic disturbances, a common scenario in R2R systems due to roller rotations. In this study, we introduce a Spatial-Terminal Iterative Learning Control (STILC) method, incorporating an iteratively updated basis function, aimed at eradicating registration error in an R2R gravure printing process with unknown system parameters. By minimizing the L_∞ norm of the predicted tension fluctuation profile for the subsequent iteration, we iteratively refine an adaptive basis function. This function, alongside the registration error, is utilized in the update of the STILC compensation profile via a P-type Iterative Learning Control (P-ILC) law. Our numerical experiments show that the STILC with an adaptive basis function effectively eliminates the registration error caused by slippages and tension fluctuations due to roller-motor axis mismatches, roller roundness errors, and cogging torques of motors. Additionally, we also verify by simulations that our method provides a faster convergence rate than the STILC with invariant basis functions and a Norm Optimal SILC (NOSILC) which minimizes the L₂ norm of the tension fluctuation profile iteratively. Furthermore, we compare the performances when different initial basis functions are applied. The simulation results show that a properly-selected initial basis function can significantly improve the convergence rate. This STILC approach with an adaptive basis function is promising for various industrial applications involving spatially periodic disturbances, especially for those with unknown dynamics and without given state trajectories to track rigorously, which is common in engineering practice.

Presenting Author: Zifeng Wang Northeastern University

Presenting Author Biography: Zifeng Wang received his B.S degree in mechanical engineering from Shanghai Jiao Tong University in China in 2018. He was a mechanical engineer in Beko China R\&D Center from 2018 to 2020. He is currently a Ph.D candidate in Industrial Engineering at Northeastern University, Boston, MA, USA. His research interests include online learning control and data-driven methods in advanced manufacturing processes, aimed at leveraging the repetitiveness and abundant data in manufacturing scenarios to improve the automation and intelligence of industrial production systems.

Authors:

Zifeng Wang Northeastern University
Xiaoning Jin Northeastern University