Data Mining

Extrusion and calendering machines can provide up to 2,000 parameters from control systems. The amount of data poses a challenge for production and quality employees and often leads to process errors not being found or correlations not being recognized due to the large number of parameters.

With the help of machine learning processes (mathematical calculation for, among other things, artificial intelligence), data on dependencies can be analyzed in the shortest possible time. The methods used (support vector machines, classification trees, neural networks) are able to calculate models on the basis of different amounts of data for finding the cause and for forecasting.

To get to this point, some preliminary work has to be done. Here is a suitable sequence that should help as a guideline. Put together a team first, then combine all the necessary control, order and quality data using an ETL process, visualize the first data, set limits using SPC and find the causes of production problems automatically, e.g. using a feature Selection. You then play the knowledge you have gained back into the machine and define self-controlling routines. (See following pictures)

Production forecasting process to reduce scrap. This process was tested and implemented on extrusion, printing, paper and board machines.

Selection of a suitable AI for the prediction of production scrap.

Model process

©2020 Kohlert Consulting

  • LinkedIn