EXPERIMENTAL VALIDATION AND UNCERTAINTY QUANTIFICATION OF PARTITIONED MODELS
Event details
| Date | 05.05.2015 |
| Hour | 10:30 › 11:30 |
| Speaker | Garrison Stevens, Glenn Department of Civil Engineering,Clemson University, Clemson, South Carolina, USA |
| Location | |
| Category | Conferences - Seminars |
Multi-scale and multi-physics modeling involves the simultaneous use of multiple constituent models that resolve different scales or physics by iteratively communicating outputs of these models with each other, referred to as partitioned analysis. Such coupled models aim to achieve a compromise between accuracy and efficiency when a single scale or physics is not a sufficiently accurate representation of the system. In strongly coupled models, the iterative nature of coupling operations causes uncertainties and errors in constituent models to be passed back and forth through the interface, which may lead to the accumulation of these uncertainties and errors in the coupled model. Thus, the predictive ability of a coupled model depends upon the accuracy and precision of its constituent models. A distinct advantage of partitioned analysis, that has not been recognized previously, is the transparency it offers for the quantification of uncertainties and errors in model predictions through the decomposition of a physical system into multiple scales. The predictive ability of constituent models developed independently can be improved in an isolated manner by rigorously calibrating model predictions against experimental measurements conducted in the corresponding domains. Furthermore, understanding and quantifying the sources of uncertainties and errors in the coupled models can allow model developers to strategically improve the model’s predictive capability. Herein, a partitioned approach for experiment-based validation and uncertainty quantification is presented. Bias-corrected partitioned analysis utilizes separate-effect experiments conducted within each constituent’s domain to test the validity of the independent constituents and integral-effect experiments executed within the coupled domain to validate the entire coupled system. This approach for calibration and validation utilizing both separate- and integral-effect experiments is demonstrated on a multi-scale plasticity model coupling a finite element code at the macro-scale and visco-plastic self-consistent code at the meso-scale. Results demonstrate that improvements in predictive capability can be achieved by utilizing separate-effect experiments as a means of appropriately bias-correcting constituent model predictions during coupling iterations.
Practical information
- General public
- Free
Organizer
- IMAC
Contact
- Gaudenz Moser