0803 Quality of model prognosis - from lab data to structural performance

  • Jörg F. Unger, BAM
  • Steffen Freitag
  • Daniel Straub
  • Bruno Sudret
  • Francisco Chinesta
  • Michael Beer
  • Phaedon-Stelios Koutsourelakis

Models are developed to explain physical system behavior and transfer this knowledge to predict the systems performance under new scenarios when the assessment is not possible on a purely experimental basis. The quality of a model prediction is of utmost importance, but it is challenging to assess the quality for predictions of systems under real-world conditions. Setting up a model, in particular for complex systems, often requires expert knowledge and subjective modelling choices are difficult to avoid. This is also a major obstacle when newly developed models are published and intended to be used in industrial application. Furthermore, it is often difficult to reproduce and validate implementations of scientific papers since relevant information is missing and the effort of a full reimplementation is often significant. In addition, there is the challenge of generalizing sophisticated models or model parameters calibrated based on lab experiments or in situ measurements to real application and in particular use them in a predictive manner, e.g. under exposure to previously unobserved environmental conditions.

The aim of this MS is to discuss procedures to evaluate the quality of a model prognosis from individual lab data up to structural simulations with the perspective to automate the complete process. This includes but is not limited to
• Digital twins of materials and structures
• Uncertainty quantification and reliability analysis
• Bayesian model calibration, model updating and model comparison
• Identification of model parameters, model bias and databased model extensions
• Propagation of uncertainties and model errors
• Machine Learning for merging lab data with structural simulation models
• Data augmentation via fusion of lab data and simulation-based synthetic data
• Structuring data using ontologies and sharing this linked data using the FAIR principals
• Reproducible workflows in scientific computing

© WCCM-APCOM 2022. All Rights Reserved.