Data-based Optimization of Plastics Processing

Energy Efficiency and Flexibility through Digital Twins

The number of settings in extrusion processes is often quite large.

But which settings lead to the desired product with the lowest possible energy consumption, acceptable quality and low costs? Or perhaps you would like to exploit flexibilities, like varying the throughput, in order to link production to the electricity price in the sense of demand-side management.

Knowledge based on experience is often used to answer the question of appropriate settings. Although this approach has its justification, more precise analyses and mathematically sound methods–that also allow the inclusion of expert knowledge–can often optimize processes better.

 

To this end, we work with methods from the fields of:

  • Machine Learning
  • Artificial Intelligence 
  • statistical data analysis

The goal: to create a digital twin of the process and use it to optimize the process.

For example, we use neural networks to calculate an inverse mapping of the digital twin, which uses the desired process result (energy consumption, quality, costs) to calculate the required settings (recipe, temperatures, rotation speed, ...).

Do you also want to optimize your processes? To do so, let us take the following steps together:

  1. Data collection: In order to develop data-based models, data is of course needed. If data is already available, we support you in examining it. We also evaluate the information content using methods from information theory. If it turns out that the existing data cannot cover or explain all important phenomena, or if no data are available yet, we develop a design of experiments together. With minimal effort and cost we augment the database in an optimal way and thus obtain the required information for further modeling. 
  2. Extension of the data set: If necessary, we adapt a simulation tool individually to your machine in order to generate further data with it.
  3. Modeling: Depending on the complexity and amount of available data, a suitable model class is selected, and a model is adapted for the problem at hand. This includes the creation of the digital twin as well as the development and implementation of the optimization strategy.
  4. Assistance system: The results are cast into an assistance system so that the optimization routines can be used directly on-site at the machine.  

Did this pique your interest or do you have further questions? Then simply contact us.

Data-based Optimization of Plastics Processing
© Fraunhofer ITWM
Data-based Optimization of Plastics Processing