Foundational Fuel Chemistry Model 2 — Can data assimilation yield useful insights in reaction rate constants?

The foundational fuel chemistry model version 2 (FFCM-2), consisting of 96 species and 1054 reactions, is a recently developed reaction model for C 0-4 hydrocarbon combustion. In the development of FFCM-2, reaction rate constants, primarily sourced from reaction rate theory and experiments, were evaluated for their uncertainties. Optimization and uncertainty minimization were conducted using a neural-network-based method of uncertainty minimization employing polynomial chaos expansions (NN-MUM-PCE), which assimilated over 1000 sets of legacy combustion property data while ensuring that all rate constants adhered to their associated physical constraints and uncertainty bounds. The current study focuses on an analysis of the data-assimilated and uncertainty-minimized reaction rate constants. By comparing the trial and data-assimilated rate constants with literature combustion experiments, we demonstrate that large-scale data assimilation, when combined with appropriate physical constraints and a large, well-evaluated combustion property dataset, can effectively minimize rate uncertainties, provide valuable insights into missing reactions and individual rate constants. Novelty and significance statement FFCM-2, a detailed reaction model for foundational fuel combustion, was developed by assimilating a comprehensive combustion experiment database for C 0 − 4 fuels. The work described herein illustrates two important outcomes of the FFCM-2 effort: when a sufficiently large set of well-evaluated legacy combustion property data is utilized, the data assimilation approach can (a) offer insights into missing reactions, and (b) constrain rate parameter uncertainties, thus leading to a combustion reaction model with greater predictability and fundamental kinetic fidelity. Overall, we illustrate that data assimilation, which underpins many approaches now broadly referred to as physics-informed artificial intelligence (AI), can yield physical results, including accurate rate constants.