What happens when you control an extruder with a data-driven model?

Significantly less waste and lower specific product costs!

CHALLENGES

• The important quality parameter MFI (melt flow index, a type of temperature-dependent viscosity) can only be determined accurately in the laboratory.
• Online rheometers for MFI measurement are unstable and unreliable, expensive to maintain, permanent calibrations are necessary, systems get clogged, etc.
• Both methods for determining the quality of the output take time. During this time, production continues. In the worst case, rejects are produced. These can be exponential the longer it takes to determine the quality.
• However, the extruder is in principle also a kind of rheometer, as it kneads the material and must expend energy to do so, which depends on the ductility.
• The aim of the project was therefore to use the online measurement data of the extruder to calculate the MFI values with a data-driven model, as precise as a laboratory analysis would be, and to use this model to control the catalyst.

OUR SOLUTION APPROACH

• Recording of laboratory values with timestamps during normal production.
• Identification of relevant process parameters (entropy analysis).
• Fully automatic model building with NeuroModel®.
• Validation of the model with unknown data.
• Connection of the model to the control system.
• Use of the model prediction as “measured value” of catalyst control loop.

OPERATIONAL EXCELLENCE 4.0

• 100% quality check of the entire output in real-time.
• Prediction of the quality before the product is discharged from the extruder in real-time.
• Significant reduction of waste during product changeover.
• We were able to reduce waste by 42%.
• This resulted in annual savings per extruder of 180.000 Euro.