Automatic reaction mechanism generation :

[摘要] Growing awareness of climate change and the risks associated with our society;;s dependence on fossil fuels has motivated global initiatives to develop economically viable, renewable energy sources. However, the transportation sector remains a major hurdle. Although electric vehicles are becoming more mainstream, the transportation sector is expected to continue relying heavily on combustion engines, particularly in the freight and airline industries. Therefore, research efforts to develop cleaner combustion must continue. This includes the development of more efficient combustion engines, identification of compatible alternative fuels, and the streamlining of existing petroleum resources. These dynamic systems have complex chemistry and are often difficult and expensive to probe experimentally, making detailed chemical kinetic modeling an attractive option for simulating and predicting macroscopic observables such as ignition delay or CO₂ concentrations. This thesis presents several methods and applications towards high fidelity predictive modeling using Reaction Mechanism Generator (RMG), an open source software package which automatically constructs kinetic mechanisms. Several sources contribute to model error during automatic mechanism generation, including incomplete or incorrect handling of chemistry, poor estimation of thermodynamic and kinetics parameters, and uncertainty propagation. First, an overview of RMG is presented along with algorithmic changes for handling incomplete or incorrect chemistry. Completeness of chemistry is often limited by CPU speed and memory in the combinational problem of generating reactions for large molecules. A method for filtering reactions is presented for efficiently and accurately building models for larger systems. An extensible species representation was also implemented based on chemical graph theory, allowing chemistry to be extended to lone pairs, charges, and variable valencies. Several chemistries are explored in this thesis through modeling three combustion related processes. Ketone and cyclic ether chemistry are explored in the study of diisoproyl ketone and cineole, biofuel candidates produced by fungi in the decomposition of cellulosic biomass. Detailed kinetic modeling in conjunction with engine experiments and metabolic engineering form a collaborative feedback loop that efficiently screens biofuel candidates for use in novel engine technologies. Next, the challenge of modeling constrained cyclic geometries is tackled in generating a combustion model of JP-10, a synthetic jet fuel used in propulsion technologies. The model is validated against experimental and literature data and succeeds in capturing key product distributions, including aromatic compounds, which are precursors to polyaromatic hydrocarbons (PAHs) and soot. Finally, oil-to-gas cracking processes under geological conditions are studied through modeling the low temperature pyrolysis of the heavy oil analog phenyldodecane in the presence of diethyldisulfide. This system is used to gather mechanistic insight on the observation that sulfur-rich kerogens have accelerated oil-to-gas decomposition, a topic relevant to petroleum reservoir modeling. The model shows that free radical timescales matter in low temperature systems where alkylaromatics are relatively stable. Local and global uncertainty propagation methods are used to analyze error in automatically generated kinetic models. A framework for local uncertainty analysis was implemented using Cantera as a backend. Global uncertainty analysis was implemented using adaptive Smolyak pscudospcctral approximations to efficiently compute and construct polynomial chaos expansions (PCE) to approximate the dependence of outputs on a subset of uncertain inputs. Both local and global methods provide similar qualitative insights towards identifying the most influential input parameters in a model. The analysis shows that correlated uncertainties based on kinetics rate rules and group additivity estimates of thermochemistry drastically reduce a model;;s degrees of freedom and can have a large impact on model outputs. These results highlight the necessity of uncertainty analysis in the mechanism generation workflow. This thesis demonstrates that predictive chemical kinetics can aid in the mechanistic understanding of complex chemical processes and contributes new methods for refining and building high fidelity models in the automatic mechanism generation workflow. These contributions are available to the kinetics community through the RMG software package.

[发布机构] Massachusetts Institute of Technology