In an emerging energy landscape that increasingly discourages the use of traditional fossil fuels, there remain applications for which the continued use of high energy density liquid fuels is required, such as aviation and other uses where space and weight are critical design factors, or long term energy storage where cost and long term availability are required. To achieve this while transitioning to green sources of energy requires the design of next-generation combustion engines that can burn alternative fuels such as bio-derived or synthetic fuels; this process will be heavily dependent on design tools such as computational fluid dynamics packages, underpinned by accurate chemical kinetic models for the fuels in question. These kinetic models often contain thermodynamic information about hundreds of unique chemical species and thousands of chemical reactions forming an interconnected network between species governing their rates of production and destruction. Historically, generation of such high-fidelity kinetic models has required decades of research---too long for the engines that will require advanced fuels.
Development of a kinetic model that is predictive of certain quantities of interest (ignition delay times, flame speeds, etc) can broadly be broken into four distinct stages: 1) initial ``crude'' model generation, 2) experimental design, 3) experiments and ab-initio theory calculations, and 4) kinetic model optimization. Advances in data-enabled science and ever-increasing computing power have offered pathways towards eventually automating this process. This work aims to introduce a collection of tools and building blocks that will assist in the overall aim of automatic kinetic model development, and in doing so fill important gaps in the current capabilities available in the literature. In particular, the work here touches on aspects of all four of the stages in the model development process described above.
With regard to 1), while there are tools available in the literature for automatic generation of kinetic models for an increasingly large library of fuels, these models remain subject to the constraints imposed by current chemical kinetic model structures and combustion codes. Here, automatic screening procedures are introduced that investigate the impact on kinetic model prediction errors due to two distinct issues related to pressure-dependent chemistry: the lack of a new class of chemical reaction type in current chemical kinetic models, and effects due to how species-specific energy transfer parameters are represented in pressure-dependent stabilization reactions within kinetic models.
With regard to 2) and 3), a Bayesian optimal experimental design algorithm is paired with computer-controllable perfectly-stirred reactor experiments with unique capability to both explore a combinatorically complex experiment parameter space (including flowing up to ten unique gas mixtures simultaneously) and measure dozens of chemical species using rapid, on-line diagnostics. This setup allows for key reaction pathways to be carefully "sensitized'' with the addition of trace quantities of key chemical species, a capability that has not been used elsewhere in literature. Generally speaking, other experimental design algorithms in literature have not explored experimental design spaces that are radically different from those used by experienced researchers in their manual experimental design processes, and the complexity of the mixtures explored by most traditional combustion experiments is limited to two or three different chemical species at most. The sensitization of key reaction pathways unlocks the ability to perform truly transformational parameter inferences with minimal amounts of experimental data.
With regard to joining step 3) to 4) in the above process, semi-automated post-processing codes allow for rapid optimizations to be performed for a prior kinetic model on the basis of experiments chosen by our experimental design algorithm. Critically, a combination of the experimental design algorithm developed here and the jet-stirred reactor experiments described was tested on the kinetic model for N₂O decomposition, which has uncertainties for key reaction rates that have persisted for decades (indeed, researchers suggest kinetic rate constants for N₂O+O=N₂+O₂ that differ by at least four orders of magnitude!). Optimizations using the Multi-Scale Informatics (MSI) tool developed by our research group were run on the basis of experimental data obtained in the aforementioned experiments, and used to gain insights about the rate constant for a key reaction in N₂O decomposition chemistry, N₂O+O=N₂+O₂ , serving as a proof-of-concept for key portions of what will form the backbone of an automatic kinetic model development pipeline.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/19e7-3731 |
| Date | January 2023 |
| Creators | Barbet, Mark |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0763 seconds