Invitation Only
Speaker: Michael Frenklach, Ph.D (University of California, Berkeley)
Title: From Data to Knowledge through Models in Combustion
I will present a methodology for transforming experimental data into predictive models, illustrated through a concrete example from combustion chemistry. Disparate data are analyzed jointly using a numerical framework based on response surface techniques and constrained optimization, enabled by the identification of active variables and active subspaces. A rigorous measure of consistency among heterogeneous data sets is developed, with Lagrange multipliers identifying factors that govern consistency; once consistency is established, the resulting model–data system propagates prior uncertainties into prediction. By integrating the full collection of available data into a single joint analysis, substantially more information is extracted than is possible from isolated studies. This methodology, termed Bound‑to‑Bound Data Collaboration (B2B‑DC), provides a systematic scientific approach for assessing and assuring mutual consistency of data, rigorously quantifying uncertainty, evaluating model predictability, identifying deficiencies in existing data, guiding high‑impact experimental and theoretical efforts, fostering community consensus, and transforming assembled data into new predictive knowledge.