BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.4.12//EN TZID:Asia/Kolkata X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT UID:231@cds.iisc.ac.in DTSTART;TZID=Asia/Kolkata:20261106T110000 DTEND;TZID=Asia/Kolkata:20261106T120000 DTSTAMP:20261028T070745Z URL:https://cds.iisc.ac.in/events/ph-d-thesis-defense-toward-machine-learn ing-accelerated-simulations-of-cavity-stabilized-flames/ SUMMARY:Ph.D Thesis Defense: Toward machine learning-accelerated simulation s of cavity-stabilized flames DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Def ense\n\n\n\nSpeaker: Mr. Priyabrat Dash\nS.R. Number: 06-18-01-10-12-20-2- 19095\nTitle: Toward machine learning-accelerated simulations of cavity-st abilized flames\nResearch Supervisor: Prof. Konduri Aditya\nThesis examine r : Dr. Shubhangi S. Bansude\, Indian Institute of Technology\, Gandhinaga r\, Gujarat\nDate &\; Time : November 06\, 2026 (Friday)\, 11:00 AM\nVe nue : #102\, CDS Seminar Hall\n\n\n\nABSTRACT\nDirect numerical simulation s (DNS) provide a rigorous framework for resolving the multiscale interact ions governing turbulent reacting flows in practical combustors. By captur ing all dynamically relevant spatio-temporal scales\, DNS complements expe rimental approaches by providing access to flow and thermochemical quantit ies that are difficult to measure at the required spatio-temporal resoluti on. Cavity flameholders\, essential to gas turbines and ram/scramjet engin es\, represent one such configuration where this capability is especially valuable. These devices generate low-speed recirculation zones that contin uously entrain heat and radicals\, facilitating flame stabilization. Despi te its unparalleled detail\, the prohibitive computational cost of DNS lim its its adoption for complex configurations. This thesis leverages the dat a-rich nature of DNS to develop machine learning frameworks that can accel erate simulation workflows\, augment experimental diagnostics\, and yield deeper physical insight into the dynamics of cavity-stabilized combustion. \n\nTo address the computational cost of DNS\, this work first develops a physics-guided self-supervised super-resolution (SR) methodology to recons truct fine-scale structures in isotropic turbulence from coarse-grained fl ow fields. By treating SR as a state estimation problem\, this approach do es not require paired high-resolution training data\, making it well-suite d for flow configurations where fully resolved simulations are prohibitive . Separately\, a graph neural network (GNN) based SR framework is develope d for turbulent reacting flows on complex meshes\, including structured no n-uniform and unstructured configurations. Leveraging message passing laye rs\, this approach enables accurate reconstruction of unresolved flow feat ures across geometrically complex domains\, extending the applicability of data-driven SR to practical combustion configurations.\n\nThe mesh-native nature of GNNs established in the SR framework naturally extends to other modeling challenges in reacting flow simulations. A GNN-based closure has been developed to predict filtered species production rates from filtered thermochemical scalars on non-uniform meshes\, with a priori evaluations demonstrating strong generalization across varying hydrogen blend composit ions and filter widths. To further reduce/alleviate the complexity of deta iled chemical kinetics\, two GNN-based mechanism reduction formulations ar e introduced\, GNN-SM and GNN-AE\, leveraging surrogate-assisted and autoe ncoder-based strategies respectively to identify kinetically dominant path ways\, achieving up to 95% reduction in species and reactions while mainta ining predictive accuracy.\n\nThe mesh-agnostic nature of GNNs established in the SR framework naturally extends to other modeling challenges in rea cting flow simulations. A GNN-based closure has been developed to predict filtered species production rates from filtered thermochemical scalars on non-uniform meshes. A priori evaluation has demonstrated strong generaliza tion across varying fuel blend compositions and filter widths. To reduce t he complexity of detailed chemical kinetics\, two GNN-based mechanism redu ction formulations are introduced\, GNN-SM and GNN-AE\, leveraging surroga te-assisted and autoencoder-based strategies respectively to identify kine tically dominant pathways. Up to 95% reduction in species and reactions ha s been observed\, while maintaining predictive accuracy.\n\nDNS data is fu rther used to augment experimental diagnostics in trapped vortex combustor s. Vision transformer (ViT) models are trained on simulation data with syn thetic noise to infer velocity fields from planar laser-induced fluorescen ce (PLIF) measurements\, avoiding the intrusiveness of particle image velo cimetry. These models demonstrate accurate reconstruction of large-scale v ortical structures\, enabling data fusion and digital twin development whe re intrusive diagnostics are impractical.\n\nFinally\, DNS is employed to advance our physical understanding of cavity-stabilized combustion. Massiv ely parallel compressible flow simulations of lean premixed ethylene-air f lames in a backward-facing step reveal how the primary recirculation zones transport radicals and alter dominant chemical pathways to support flame stabilization. These investigations are extended to the HIFiRE 2 supersoni c combustor\, where dual-mode and scram-mode operations are analyzed to ch aracterize flame stabilization regimes and the evolution of chemical pathw ays in non-premixed supersonic flames. Collectively\, this work integrates high-fidelity DNS with geometric-aware machine learning frameworks\, esta blishing a foundation for accelerated simulations\, improved diagnostics\, and reduced-order modeling in cavity-stabilized reacting flows.\n\n\n\nAL L ARE WELCOME CATEGORIES:Events,Thesis Defense END:VEVENT BEGIN:VTIMEZONE TZID:Asia/Kolkata X-LIC-LOCATION:Asia/Kolkata BEGIN:STANDARD DTSTART:20251106T110000 TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE END:VCALENDAR