Dimensionality reduction of chemical kinetic mechanisms using data-driven clustering techniques for MILD combustion
(2024)
Files
PhD_Thesis_Pietro_Pagani_Public.pdf
Open Access - Adobe PDF
- 15.67 MB
Details
- Authors
- Supervisors
- Abstract
- The current global energy landscape, marked by volatility and complexity, highlights the vulnerabilities of a fossil fuel-dependent world. Considering also that the energy derived from fossil fuels supplies over two-thirds of the world energy needs, the development of fuel-flexible and environmentally friendly combustion technologies has become urgent. Among the proposed techniques, Moderate or Intense Low-oxygen Dilution (MILD) combustion, stands out for its high efficiency and reduced pollutant emissions, particularly of nitrogen oxides (NOx) and soot. MILD combustion is also highly versatile, capable of utilizing a broad range of fuels, making it a promising solution for micro gas turbines (MGTs) for decentralized energy production. This phD thesis explores the application of MILD combustion in MGTs, enhancing the accuracy of computational fluid dynamics (CFD) simulations. The key focus is given on data-driven modeling, which has emerged as a powerful tool in combustion research for developing locally reduced models from high fidelity data. The present work advances the state-of-the-art by enhancing the Sample-Partitioning Adaptive Reduced Chemistry (SPARC) workflow, resulting in the optimized eSPARC framework. Such method couples adaptive chemistry and machine learning to build, in the preprocessing phase, a library of skeletal mechanisms, associated to clusters of similar thermo-chemical states and identified in a training dataset. Moreover, the novel approach integrates advanced reduction techniques, such as Computational Singular Perturbation (CSP), physics-informed clustering, and a novel error estimation step, optimizing the balance between accuracy and computational efficiency. This enhanced eSPARC workflow demonstrates improvements in computational speed of CFD in MILD conditions: the application to RANS and LES of the Adelaide Jet-in-Hot-Coflow (AJHC) burner showed speedup between 2 and 4 on the CPU, depending on the size of the oxidation mechanism.
- Affiliations
Citations
Pagani, P. (2024). Dimensionality reduction of chemical kinetic mechanisms using data-driven clustering techniques for MILD combustion. https://hdl.handle.net/2078.5/235904