Neural network assisted combustion chemistry reaction model optimization and uncertainty minimization
Abstract/Contents
- Abstract
- Combustion chemistry is hierarchical in nature. A detailed combustion chemistry reaction model that describes the kinetics of a basis set of compounds that include H2, CO, C1-C4 hydrocarbons and selected aromatics (benzene and toluene) is the foundation for modeling the reaction chemistry of all hydrocarbon compounds. Recently, a physics-based Hybrid Chemistry (HyChem) modeling approach was proposed for modeling heavier hydrocarbons and real distillate fuels. It has been demonstrated that an accurate and uncertainty-minimized foundational fuel chemistry model is critical to HyChem modeling. In an earlier work, Foundational Fuel Chemistry Model Version 1.0 (FFCM-1) was developed for the combustion of H2, CO and CH4. In the current work, we develop Foundational Fuel Chemistry Model Version 2.0 (FFCM-2), extending the earlier FFCM-1 to all relevant C0-C4 species. FFCM-2 is a reaction model that consists of 96 species and 1054 reactions, describing the reaction kinetics of C0-C4 foundational fuels. The trial model was compiled by evaluating the existing kinetic studies: each rate coefficient was sourced from either direct experimental measurement, ab initio theoretical calculations, or estimation from similar kinetic systems. The uncertainty factor for each rate coefficient was also evaluated. Legacy combustion property data, including laminar flame speed, ignition delay time, species measurements in shock tube and selected speciation data in flow reactor and burner stabilized flame, were collected by extensive literature review. 1192 targets were selected from the database, each with data uncertainty quantified. They span a wide range of thermodynamic conditions. FFCM-2 was optimized and uncertainty-minimized against these 1192 targets. To address the high-dimensionality of FFCM-2, we advanced the approach of neural network-based methods of uncertainty minimization using polynomial chaos expansions (NN-MUM-PCE), extending the earlier MUM-PCE framework employed in FFCM-1 to new capabilities. We demonstrate that neural networks as response surfaces are superior to the earlier polynomial response surfaces, in their accuracy, efficiency, robustness, and scalability for large-scale reaction model optimization and uncertainty minimization. The resulting FFCM-2 is shown to accurately predict combustion of C0-C4 hydrocarbons, with appreciably reduced prediction uncertainties. The results for optimization and uncertainty minimization are discussed in detail, including the choice of weighting factor for reaction kinetics vs. target data, inconsistent targets, freezing of unimportant parameters, etc. The optimized rate parameters that deviate significantly from the trial assignments were justified and interpreted by comparing against an extensive set of literature studies. The optimized FFCM-2 is applied to HyChem modeling of two real fuels, the conventional Jet A fuel and the alcohol-to-jet (ATJ) sustainable aviation fuel. A systematic optimization framework was established by updating the HyChem parameters with changes or improvements to the foundational fuel chemistry, using as targets the species data obtained from the thermal and oxidative pyrolysis in shock tube and flow reactor to solve the inverse problem. The FFCM-2 based HyChem models is shown to accurately predict the global combustion properties with significantly reduced model uncertainties. The two FFCM-2 based HyChem models are also extended to predict ignition delay times in the negative-temperature coefficient region.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Zhang, Yue, (active 2023) |
|---|---|
| Degree supervisor | Wang, Hai, 1962- |
| Thesis advisor | Wang, Hai, 1962- |
| Thesis advisor | Edwards, Matthew R. (Matthew Reid) |
| Thesis advisor | Hanson, Ronald |
| Degree committee member | Edwards, Matthew R. (Matthew Reid) |
| Degree committee member | Hanson, Ronald |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Mechanical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Yue Zhang. |
|---|---|
| Note | Submitted to the Department of Mechanical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/rq641fc4021 |
Access conditions
- Copyright
- © 2023 by Yue Zhang
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...