(180d) Accelerating the Process of Combustion Mechanism Discovery through Correlated Uncertainty and Sensitivity Analysis

Determining the mechanism of complex reaction networks is a critical step towards designing cleaner combustion fuels. To model combustion, hundreds of species and thousands of reactions must be enumerated and parameterized. Networks of this scale can be generated automatically with tools like Reaction Mechanism Generator (RMG). However, many of the generated model parameters are rough estimates based on rate rules or Benson group additivity. These estimates can be improved with density functional theory (DFT) and transition state theory calculations, but the computational cost is prohibitively expensive to apply to every reaction in the mechanism. Model parameters can also be improved by running new experiments, but this incurs even greater costs in time and equipment.

This study attempts to accelerate the process of modeling new fuels by identifying the most impactful parameters and improving their accuracies with automated DFT calculations and/or carefully targeted experiments. In this study, the prioritized list of parameters to improve benefits from consideration of correlated uncertainties and sensitivities of the model parameters with respect to the training data in the RMG database. This approach accounts for the fact that multiple reactions may be estimated using the same training data, and many data can contribute to a single reaction estimate. Therefore, consideration of correlated uncertainty allows us to identify training reactions, and not just reactions in the model, whose improvement will have an outsized effect on overall model accuracy. The end result is an efficient automated workflow in which a mechanism, generated in RMG, is automatically and iteratively improved using DFT calculations selected from a ranked list of key parameters.

This process is demonstrated and the accuracy of the model is tested against experimental ignition delays, flame speeds, and species concentrations. This workflow is offered as a means to dramatically accelerate the timeframe for modeling new fuels.