Application of Machine Learning in Process Control in Optical Fiber Manufacturing

Abstract

The current era of big data and IoT has propelled the manufacturing industry to the era of “Industry 4.0”. This thesis presents an approach to manufacturing process control through the use of Machine Learning models in the optical fiber manufacturing industry. Utilizing measured production data from the fiber drawing tower, a long short-term memory (LSTM) neural network structure is used to find the correlation between the inputs and outputs of the process. Different experiments were conducted on the physical draw tower and the simulation to gauge the accuracy of the model and how well it mimics the plant’s performance. This thesis, then, presents an in-depth investigation to the deployment of the digital twin model on an industrial PLC in order to control the diameter of the produced optic fiber at a given setpoint. The model would be able to predict and anticipate changes in the diameter and adjust the gains on the PLC to keep the process under control. This could potentially replace the iterative and laborious process of controller tuning and serve as a tool to be widely utilized in manufacturing settings.