A Machine Learning Approach to Improve Diameter Control in Desktop Fiber Extrusion Processes

Abstract

A machine learning approach to controlling the diameter of a desktop fiber extrusion process with a PLC is developed and evaluated against the performance of PID control. The deep reinforcement learning model can learn how to control the output diameter of the process based on a given target without any knowledge of the system dynamics. The model learns how to control the output diameter after being trained on hours of data recorded from an open loop control process. After training the model can receive sensory information from a PLC, calculate an action based on the desired target and send the action to the PLC to execute.