| Summary: | Optical fibers are ubiquitous in the 21st century as they form the backbone of the internet and electronic communication and enable a global village to exist. Optical fibers play a pivotal role in modern technology and communication for several reasons. They enable high speed data transmission over large distances, while minimizing the data interception. In addition, they are also used in fields like medicine (fiber-optic imaging and endoscopy), sensing technologies (used in temperature, pressure, and strain sensors), and industrial settings (for data transmission and control systems). Therefore, it is of utmost importance that the manufacturing process of optical fibers is better controlled by developing advanced control algorithms that enhance the state-of-the-art PID (Proportional–Integral–Derivative) controllers. This thesis attempts to showcase the work done to establish a framework and a “digital twin” for deploying advanced learned control algorithms on industrial platforms such as Programmable Logic Controllers (PLC) based on machine learning models such as DDPG (Deep Deterministic Policy Gradient). To develop and train such control algorithms, a desktop version of a fiber draw tower was designed, manufactured, and controlled via a PLC. System dynamics data was collected using a readily available preform substitute and the manufactured desktop Fiber Extrusion Device (FrED) was used to train the DDPG-based control algorithms/model. The model was then subsequently tested and compared to state-of-the-art PID algorithms. To that effect, this thesis establishes a framework and enables the path to further develop advanced control algorithms to better control the manufacturing process of fiber optics. This pivotal step promises to significantly enhance the precision and efficacy of optical fiber manufacturing processes, amplifying their impact across industries and technological frontiers.
|
|---|