Online estimation of material parameters using neural networks and uncertainty quantification

  • chair:Online-Schätzung von Materialparametern mittels neuronaler Netze und Unsicherheitsquantifizierung
  • type:Masterarbeit
  • tutor:

    M. Sc. Johannes Mitsch,  E-Mail: johannes.mitsch@kit.edu


  •

Master thesis

Online estimation of material parameters using neural networks and
uncertainty quantification





Motivation

The precise estimation of material parameters is crucial for the accuracy of manufacturing process simulations. Traditional methods for parameter determination, such as characterization experiments and least squares estimation, often provide only an initial estimate of material parameters, which is uncertain due to model and measurement deviations. In this work, the focus is on improving the initial estimate through online learning. Here, classical neural networks are to be used to update the material parameters in real time. In order to quantify the uncertainties inherent in real processes, further estimation methods such as the Extended Kalman Filter (EKF) will be used. These innovative approaches enable recursive parameter estimation and should bring the simulated results more in line with the experimental data. Overall, this can reveal insufficiently modeled material behavior and provide valuable feedback for model refinement.

Work content:
  • Research on the state of the art and research of online parameter estimation
  • Development and implementation of methods for parameter estimation for numerical simulation methods
    (Supervision by an institute from mechanical engineering and an institute from computer science)
  • Benchmark of the developed method against conventional characterization methods
  • Written elaboration and documentation of the results
Requirements
  • Initiative and independent way of working
  • Strong analytical skills
  • Interest in simulation, numerics & estimation methods
  • Programming experience in Python
  • Previous knowledge of estimation theory is helpful

Subject area: Machine learning with application relevance in mechanical engineering

Supervision: FAST - Department of Lightweight Construction, IAR - Chair of Intelligent Sensor-Actuator Systems (ISAS)

Type of work: simulative/numerical

Start: immediately

Contact: M. Sc. Johannes Mitsch, E-Mail: johannes.mitsch∂kit.edu

Tobias Würth, M.Sc.
Phone: +49 721 608-45410
E-Mail: Tobias.Wuerth∂kit.edu
Markus Walker, M.Sc.
Phone: +49 721 608-44354
E-Mail: Markus.Walker∂kit.edu