Process Analysis by Machine Learning

Data Analysis Ensures Product Quality

The understanding the dependency between the quality of a product and the different significant process variables is one of the major challenges in modern production. The combination of physical, technical, or chemical processes with economic conditions results in a number of relevant process parameters and dependencies between them.

To optimize the production it is necessary to quantitatively describe the behavior of quality and performance values by changing individual production process parameters. The possibility to predict quality values from process parameters is essential.

For this purpose we generate process models using:

  • measurement data from the real production process
  • expert knowledge about the production process
  • theoretical knowledge

Lacking theoretical knowledge can and must be compensated through quantitatively and qualitatively valuable measurement data. Especially in complex processes, modeling on theoretical insights of the process structure rapidly reaches its respective limits. As far as measurement data, like from systematical test series, is available, a partially or fully data-based model generation is an efficient alternative.

Prognosis Modeling

A good prognosis model can be used for the following purposes:

  • simulation of the process behavior by variation of influential factors
  • estimation of improved/optimal process parameters
  • systematic product design
  • online support on running production plant

Finally, depending on its structure and whether the model equations are explicitly accessible, the specialist gains a better theoretical understanding of the manufacturing process.

Machine Learning Algorithms

  • Deep Learning
    • Deep Belief Networks
    • Deep Neural Networks
  • Bayesian Nets, Chow-Liu Nets, Markov Random Fields
  • Random Forests, Lasso
  • Subspace Clustering
  • Neural Networks