Utilizing machine learning for parameter optimization
On a quest for operational excellence
Glanzstoff was working with Resultance, a mid-sized consultancy firm specialized in supporting organizational and digital transformation projects, operational performance improvement and change management. Glanzstoff was looking to decrease errors in their production and consequently increase yield for the manufacturing of high-tenacity polyester filament yarns. Speed and flexibility are at the core of their operations, and they were also committed to apply more sustainable solutions in their manufacturing process.
Polymerization: a long, complex process subject to high demands
The polymerization process is subject to high demands and great complexity, and so, understanding the whole production chain, a 7-hour long process, was crucial to engineer improvements in the system.
Polyester filament yarns are spun using a process called melt spinning or melt extrusion. This process starts by melting the polymer pallets at high temperatures, typically between 260 and 300 degrees Celsius. The melted polymer is then forced through a spinneret, which is a device that has several small holes. The spinneret helps to shape and cool the polymer into fibers. After the fibers are formed, they are rapidly cooled and solidified by passing through a water bath or blowing air on them. The solidified fibers are then wound onto a bobbin or spool to complete the yarn.
Glanzstoff experienced a costly challenge in the final step of their production process. More specifically, the solidified fibers would sometimes break when they were spun onto the bobbin. They wanted to reduce these fiber fractures and determine the root cause, so they can avoid this event from happening in the future, resulting in less material waste and an increased production yield. Additionally, they wanted to optimize the timing of their flushing operations, so that the frequency at which the reaction tanks must be cleaned between their batches could be reduced.
Interruption of the production process
When these fiber fractures occur, the operators at Glanzstoff would have to interrupt the production processto continue the spinning of the plastic fiber. Fixing this issue is a manual task, the downtime would depend on the seriousness of the error. Additionally, these fractures result in scrap and consequently a higher manufacturing cost for Glanzstoff. Between every batch, they clean their reaction tanks, which is a manual and time-intensive task.
Minimizing fractures, maximizing yield
The Glanzstoff engineers and production operators needed a system that would enable them to predict these fiber fractures based on their historical machine data, as well as gain more insight in their root-causes to avoid fractures in the future. Additionally, they wanted to optimize the timing of their flushing operations, so that the frequency at which the reaction tanks are cleaned between their batches could be reduced.
Gaining insight and control over the polymerization process
By analyzing the historical data they had captured from this polymerization process, mainly machine and sensor data, we built a model that was able to successfully predict the fiber fractures based on key parameters and adjust them to increase material yield from the production line. Further parameter optimization enabled the model to optimize the production process of these polyester filament yarns. We also optimized the timing of their flushing operations between batches, which contributed to further operational excellence.
Two optimization techniques
Initially, two different optimization options were considered in the polymerization process. In the first one, we looked at the case of optimizing the timing of flushing operations so that the frequency at which the reaction tanks must be cleaned could be reduced. The second option was to reduce the number of fractures in the plastic fiber at the end of the production line with a model that would predict these fiber fractures based on their historical machine data.
A significant reduction in fiber fracture and waste
The predictions of the final model, a result of solid teamwork, resulted in 33% less fractures in the winder– a significant optimization of the final yield, leading to a clear ROI. We also optimized the timing of their flushing between their batches, which also contributed to their operational efficiency. All in all, great results and a job very well done.