Nithin, Priyabrat win runner-up Best Oral Presentation at the AeSI CFD Conference, IIT Madras

Manuru Nithin Padiyar from our lab has received the runner-up Best Oral Presentation award at the AeSI CFD Conference, IIT Madras for his work on Graph Neural Networks for Chemical Mechanism Reduction, co-authored by Priyabrat Dash and Konduri Aditya.

Manuru Nithin Padiyar from our lab secured runner-up in the Best Oral Presentation award in the 27th AeSI CFD Symposium organized by the Aeronautical Society of India, at IIT Madras from 11th August to 12th August, 2026. Their talk was titled “Graph Neural Networks for Chemical Mechanism Reduction for Applications in Combustion” and the work was co-authored by Priyabrat Dash (PhD, CDS) and Professor Konduri Aditya (CDS). The work proposes a novel, graph neural network (GNN) based framework to reduce chemical mechanisms. Typical high-fidelity reacting flow simulations involve billions upon billions of grid-points with several DoFs at each point, which are to be calculated based on various governing equations. With the increase in the number of species and reactions in a mechanism, the number of DoFs increase, thereby increasing the computational cost. Thus, mechanism reduction serves as an effective way to reduce computational cost without compromising simulation fidelity. The work utilises a GNN auto-encoder (GAE) setup to compute relative importance scores of each species and reaction, which are then iteratively pruned based on the score order. Reduced mechanisms are evaluated against the original based on autoignition, premixed flame, and counterflow flame results to validate their accuracy. The framework was demonstrated on methane, ethylene and iso-octane mechanisms with reductions on the basis of both autoignition and premixed flame results.