Digital Twin for Enhanced Decision-Making by Optimization of Fiber Spinning Process

Traditionally, the fiber spinning process has been operated in the manner of a trial-and-error. Since the final properties of a product can’t be verified until all processes are complete, it often requires inefficient, and repetitive efforts, and causes significant costs to achieve the target characteristics of a product. Furthermore, most decision-making is based on tacit knowledge and individual experiences, therefore quality instability and high defect rate may be possible. Also, the complexity of the fiber spinning process, which consists of multiple dispersed stages, requires comprehensive monitoring and management. To address these challenges, this paper proposes to apply digital twin to the fiber spinning process, by integration of process equipment. The digital twin suggested in this paper provides real-time processing of shop floor data for better decision-making. It also derives optimal parameters of processes for achieving the desired characteristics of products by reinforcement learning. In addition, a machine learning-based predictive model supports pro-active simulations for predicting product quality before operations. As a conclusion, the digital twin supports operators more efficient work environment, reduces decisions based on personal experiences, and shorten time, cost and efforts for many trial-and-errors.