Skip to Content

Prediction in and control of complex systems

In our department, we basically deal with two different types of complex industrial systems:

  • Measurement systems
  • Machine vision systems

Our focus in dealing with measurement systems is set on the prediction of current or future states of certain system variables or quality indicators in order to be able to perform supervision of the process and to early recognize upcoming problems (faults, failures, ...) in the system for saving waste in production parts or expenses for measurement sensors and machine (components) wearings. The complexity of the system is usually due to the high number of variables we have to deal with, typically ranging from a few hundreds up to thousand, due to significant non-linearity of the modeling problem, and due to intrinsic dynamics of industrial processes. Applied
 data-driven methodologies are:

  • High-dimensional modeling for system identification and prediction
  • Incremental and evolving modeling from data streams
  • Uni- and multivariate statistical methods
  • Methods for anomaly and fault detection, isolation and diagnosis
  • Methods for early problem recognition and (heuristics-based) optimization

Our focus in machine vision systems relates to classification statements with high accuracy (see also 'Optical fault detection and quality control’), as well as to decision support systems for customers, users and operators. The methodologies we apply in this context are:

  • Classifiers
  • On-line learning methods
  • Distance measures
  • Image segmentation algorithms
  • Feature extraction (low-level and high-level features)