Supplementary materials for Ph.D. thesis "AI Digital Twin of Roll-to-Roll Printing System for Autonomous Manufacturing"
Description
The supplementary materials demonstrate the autonomous Bayesian optimization of a proportional-integral (PI) web tension controller through a Digital Twin (DT) of a roll-to-roll (R2R) manufacturing system. The operation is conducted on the DT computer, which communicates with the Physical Twin (PT), i.e., the R2R manufacturing system, via the OPC UA protocol and the web tension step response is displayed using DT operation module. Additionally the step response file is collected via SFTP communication. In each iteration, the Bayesian optimization algorithm suggests new values for the proportional (Kp) and integral (Ki) controller gains based on the results of previous experiments. The step response data is then processed to extract key dynamic control characteristics, including time constant, overshoot, and settling time. These characteristics are combined into a quality score as a weighted sum, which is used in Gaussian process modeling. The updated model is subsequently used to query new Kp and Ki gains through the Expected Improvement acquisition function. This real-time communication and model updating optimize the control of the R2R system. The video playback speed of the Supplementary video S1 is increased by a factor of 10.