As part of a research project, SKZ, a solution developer for the plastics industry, is using machine learning to monitor and control critical process parameters. The data infrastructure required for the machine learning model
is provided by Cybus Connectware.
Challenge
Lower production quality and more waste due to lack of data insight
Key processes in plastics manufacturing are the extrusion and compounding of thermoplastic materials. The raw materials, colors and additives are processed in a specialized system known as a Twin Screw Extruder (TSE) to produce high quality pellets or granules. A TSE functions as a combination of a mixer and a conveyor,
where heat is applied by shearing and heating to melt and mix various components.
However, the process is challenged by a narrow range for optimum material temperature, requiring a precise and efficient production process. Temperature deviations reduce production efficiency and lead to waste. At the end of the process, the melt flows through a 10 cm diameter output die. Temperature measurements via sensors are limited to the wall of the die. However, while outer temperatures are recorded, the central temperature of the melt can be up to 50 Kelvin higher, making direct central temperature measurements unfeasible.
In addition, the heterogeneity of the shop floor environment and unclear naming schemes for data fields increases the challenge of accessing relevant data.
Challenges
- Narrow temperature window for optimum
- Gathering comprehensive data of the melt
- Inaccurate measurement due to lack of sensors in the centre of the melt
Solution
Machine learning model provides data insights
To address these challenges, SKZ developed a machine learning model running on Cybus Connectware. This model centralizes data from multiple sources, such as feeder and ambient temperatures, and predicts necessary adjustments to the Twin Screw Extruder machine data. Cybus Connectware as the data infrastructure for the machine learning model, replaces the need for a temperature sensor inside the melt. With the external sensors, the internal melt temperature can be approximated to optimize production conditions.
In addition to the machine learning model, Connectware also collects, processes and distributes the data from the heterogeneous assets and sensors to stakeholders and visual dashboards for operators. The dashboard data includes temperature profiles and controls for feed temperature adjustments.
Solution
- Machine learning model
- Recommendations for operators via visual dashboards
- Integrating data from heterogeneous assets via Cybus Connectware
Benefits
Product quality significantly improved
The machine learning model and availability of shop floor data is a significant leap in optimizing plastic production processes. It offers a solution for obtaining and processing crucial machine data that was previously inaccessible, leading to more efficient and higher quality production with less rejects. This advancement also opens possibilities for further optimizing products and solving market challenges, advancing intelligent manufacturing in the plastics industry.
Future plans of SKZ include developing a new GUI for browsing the data layer and effectively transferring data to the machine learning model. The data extracted from the machines, such as extruder temperature and screw rotation speed, will be key inputs to the model. Additionally, the challenge of unclearly named data fields is being addressed through suggestions for pattern recognition to identify the necessary sensors.
Benefits
- Improved Product Quality
- Less rejects
- Availability of shop floor data
