Aspects of fundamental reaction kinetics and legacy combustion properties in data-assimilated combustion reaction model development

A recently developed Foundation Fuel Chemistry Model (FFCM-2) assimilated comprehensively over 1,000 legacy combustion property targets. This paper discusses two key attributes of the FFCM-2 model development. We show that machine learning approaches alone are incapable of producing predictive reaction models for fuel combustion. Although these approaches can be useful for uncovering systematic error in a reaction model, fundamental chemical kinetic knowledge would be required when the model cannot be reconciled with experiment through parametric optimization. The impact of the recent exponential growth of combustion property measurements is assessed in terms of their role in improving the FFCM-2 model accuracy and precision. It is concluded that although roughly one half of the data used in the data assimilation of FFCM-2 comes from the last decade (2010-2020), the quantitative impact of these data is small and diminishing as compared to data published prior to 2010. Implications of this finding are discussed, along with suggestions for improvement.