Multiscale Modeling with Dynamic Discrepancy

The dynamic discrepancy methodology is an approach to multi-scale modeling based on quantification of uncertainty in scale-bridging.  It is a reduced order approach using stochastic functions to efficiently replace model variability removed when reducing problem complexity.  It can be utilized in conjunction with rigorous reduced-order strategies such as the proper orthogonal decomposition.  The stochastic character of the method leads to the opportunity for machine learning in reduced model training.  A demonstration of the method on problems in modeling of carbon capture systems will be presented, using both hypothetical and real data sets.