DEPARTMENT OF COMPUTATIONAL AND DATA SCIENCES
M.Tech Research Thesis Colloquium
Speaker : Mr. Surya Datta Sudhakar
S.R. Number : 06-18-01-10-22-24-1-24309
Title : “Data-driven and low-precision methods for improving and accelerating computational fluid dynamics solvers.”
Research Supervisor : Dr. Konduri Aditya
Date & Time : May 25, 2026, 03.30 PM
Venue : # 102 CDS Seminar Hall
ABSTRACT
Advancing simulation of turbulent and reacting flows poses two distinct challenges, the accurate representation of unresolved physical processes and the efficient utilization of the modern computing hardware. This thesis addresses both directions through data-driven subgrid-scale modeling and low-precision solver development.
The first part of this thesis investigates the transferability of data-driven subgrid-scale closures for passive scalar transport in non-equilibrium turbulence, motivated by the central importance of passive scalar mixing in a wide range of engineering and environmental flows, including combustion, atmospheric transport, pollutant dispersion, and heat and mass transfer. Accurate modeling of unresolved scalar transport remains a major challenge in large-eddy simulation, while training separate neural-network closures for every flow condition is computationally expensive and impractical. This study considers decaying two-dimensional turbulence with passive scalar transport as a stringent testbed due to its non-stationary dynamics, evolving spectra, intermittency, and strong nonlocal interactions across scales.
Neural-network-based closure models are trained to predict unresolved scalar transport from filtered high-fidelity simulation data, with the Schmidt number governing scalar mixing behavior. Transfer learning across Schmidt number regimes is examined through selective fine-tuning of pretrained models and compared against conventional closure models, demonstrating the superior adaptability of learned closures. The results reveal a strong directional asymmetry, with models trained at higher Schmidt numbers transferring more effectively to lower Schmidt regimes than the reverse. Layer-wise analysis shows that predictive improvements arise primarily from adapting shallow network layers, while modifications to deeper layers often reduce performance. Loss landscape analysis further provides insight into the optimization behavior underlying successful transfer. Spectral analysis shows that the fine-scale behavior of the learned scalar closures exhibits greater universality across Schmidt numbers, whereas large-scale closure behavior remains more regime-dependent. These findings establish a computationally efficient and physically interpretable framework for transferable machine-learned scalar closures in large-eddy simulation.
The second part of this thesis investigates low-precision (FP16) computing for accelerating chemically reacting flow simulations. Although modern CPU and GPU architectures offer substantially higher throughput and lower memory and communication costs for reduced-precision arithmetic, direct application to reacting-flow solvers is challenging due to the extreme dynamic range of species concentrations, reaction rates, and thermodynamic variables, which introduce significant risks of numerical underflow and overflow. To address this, a dynamically scaled solver framework for lean hydrogen–air autoignition with detailed chemical kinetics is developed, incorporating adaptive scaling, exponent bookkeeping, and selective higher-precision treatment of numerically sensitive operations. A risk-band analysis framework is further introduced to systematically identify variables and computational regions susceptible to precision-induced numerical failure. The computational performance of the developed implementations, spanning reference formulations, templated C++, and GPU-capable kernels, is characterized through roofline analysis using Intel VTune and NVIDIA NSight Compute. The results establish the feasibility of reduced-precision reacting-flow solvers while revealing the interplay between numerical stiffness and dynamic range constraints.
Overall, this thesis investigates two complementary strategies for improving computational fluid dynamics solvers by developing transferable machine-learned closures for improved large-eddy simulation modeling and by reducing communication cost through low-precision numerical computation.
ALL ARE WELCOME
