BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.2.3.1//EN TZID:Asia/Kolkata X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT UID:190@cds.iisc.ac.in DTSTART;TZID=Asia/Kolkata:20260420T110000 DTEND;TZID=Asia/Kolkata:20260420T120000 DTSTAMP:20260409T120614Z URL:https://cds.iisc.ac.in/events/ph-d-thesis-colloquium-102-cds-20-april- 2026-toward-machine-learning-accelerated-simulations-of-cavity-stabilized- flames/ SUMMARY:Ph.D: Thesis Colloquium: 102 : CDS: 20\, April 2026 "Toward machine learning-accelerated simulations of cavity-stabilized flames" DESCRIPTION:DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES\nPh.D. Thesis Col loquium\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 -stabilized flames\nResearch Supervisor: Dr. Konduri Aditya\nDate &\; T ime : April 20\, 2026 (Monday)\, 11:00 AM\nVenue : #102\, CDS Seminar Hall \n\nABSTRACT\nDirect numerical simulations (DNS) provide a rigorous framew ork for resolving the multiscale interactions governing turbulent reacting flows in practical combustors. By capturing all dynamically relevant spat io-temporal scales\, DNS complements experimental approaches by providing access to flow and thermochemical quantities that are difficult to measure at the required spatio-temporal resolution. Cavity flameholders\, essenti al to gas turbines and ram/scramjet engines\, represent one such configura tion where this capability is especially valuable. These devices generate low-speed recirculation zones that continuously entrain heat and radicals\ , facilitating flame stabilization. Despite its unparalleled detail\, the prohibitive computational cost of DNS limits its adoption for complex conf igurations. This thesis leverages the data-rich nature of DNS to develop m achine learning frameworks that can accelerate simulation workflows\, augm ent 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 su per-resolution (SR) methodology to reconstruct fine-scale structures in is otropic turbulence from coarse-grained flow fields. By treating SR as a st ate estimation problem\, this approach does not require paired high-resolu tion training data\, making it well-suited for flow configurations where f ully resolved simulations are prohibitive. Separately\, a graph neural net work (GNN) based SR framework is developed for turbulent reacting flows on complex meshes\, including structured non-uniform and unstructured config urations. Leveraging message passing layers\, this approach enables accura te reconstruction of unresolved flow features across geometrically complex domains\, extending the applicability of data-driven SR to practical comb ustion configurations.\n\nThe mesh-native nature of GNNs established in th e SR framework naturally extends to other modeling challenges in reacting flow simulations. A GNN-based closure has been developed to predict filter ed species production rates from filtered thermochemical scalars on non-un iform meshes\, with a priori evaluations demonstrating strong generalizati on across varying hydrogen blend compositions and filter widths. To furthe r reduce/alleviate the complexity of detailed chemical kinetics\, two GNN- based mechanism reduction formulations are introduced\, GNN-SM and GNN-AE\ , leveraging surrogate-assisted and autoencoder-based strategies respectiv ely to identify kinetically dominant pathways\, achieving up to 95% reduct ion in species and reactions while maintaining predictive accuracy.\n\nThe mesh-agnostic nature of GNNs established in the SR framework naturally ex tends to other modeling challenges in reacting flow simulations. A GNN-bas ed closure has been developed to predict filtered species production rates from filtered thermochemical scalars on non-uniform meshes. A priori eval uation has demonstrated strong generalization across varying fuel blend co mpositions and filter widths. To reduce the complexity of detailed chemica l kinetics\, two GNN-based mechanism reduction formulations are introduced \, GNN-SM and GNN-AE\, leveraging surrogate-assisted and autoencoder-based strategies respectively to identify kinetically dominant pathways. Up to 95% reduction in species and reactions has been observed\, while maintaini ng predictive accuracy.\n\nDNS data is further used to augment experimenta l diagnostics in trapped vortex combustors. Vision transformer (ViT) model s are trained on simulation data with synthetic noise to infer velocity fi elds from planar laser-induced fluorescence (PLIF) measurements\, avoiding the intrusiveness of particle image velocimetry. These models demonstrate accurate reconstruction of large-scale vortical structures\, enabling dat a fusion and digital twin development where intrusive diagnostics are impr actical.\n\nFinally\, DNS is employed to advance our physical understandin g of cavity-stabilized combustion. Massively parallel compressible flow si mulations of lean premixed ethylene-air flames in a backward-facing step r eveal how the primary recirculation zones transport radicals and alter dom inant chemical pathways to support flame stabilization. These investigatio ns are extended to the HIFiRE 2 supersonic combustor\, where dual-mode and scram-mode operations are analyzed to characterize flame stabilization re gimes and the evolution of chemical pathways in non-premixed supersonic fl ames. Collectively\, this work integrates high-fidelity DNS with geometric -aware machine learning frameworks\, establishing a foundation for acceler ated simulations\, improved diagnostics\, and reduced-order modeling in ca vity-stabilized reacting flows.\n\n\n\nALL ARE WELCOME CATEGORIES:Events,Ph.D. Thesis Colloquium END:VEVENT BEGIN:VTIMEZONE TZID:Asia/Kolkata X-LIC-LOCATION:Asia/Kolkata BEGIN:STANDARD DTSTART:20250420T110000 TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE END:VCALENDAR