Arjun Sharma
Sandia National Labs
Applied and Computational Mathematics
Biography
Arjun Sharma is a Postdoctoral Appointee at Sandia National Laboratories (Computer Science Research Institute). He works at the intersection of fluid mechanics, computing, and machine learning, building fast, trustworthy models for design and discovery. Recent projects include improving classic wing aerodynamics with learned corrections, non local transport of nutrients in bacterial suspension, air-sea flux exchange in the DOE’s climate model, and machine learning physics parametrization for climate models. He codes in Python (JAX/PyTorch), Matlab and Fortran, uses GPUs, and runs larger studies on HPC clusters, with an emphasis on clear, reproducible workflows and understanding physical mechanisms. Arjun earned a Ph.D. in Mechanical Engineering (Applied Mathematics minor) from Cornell University. Before graduate school he worked in high-performance engineering at Rolls-Royce (aeroacoustics) and in Formula One (aerodynamics). He has taught and TA’d core courses in thermodynamics and fluid dynamics and enjoys mentoring students at different levels. For SRP, Arjun offers flexible projects where students can lean into coding, physics, or visualization and leave with a portfolio-ready artifact (well-documented repo, figures, and a short write-up). His mentoring style is supportive and structured, weekly one-on-ones, code reviews, and clear milestones, with opportunities for co-authorship when results warrant
SRP Project Title
AI + Physics for Better Propellers/ turbines
Topical Areas
Applied Computer Science; Applied Mathematics; Artificial Intelligence and Intelligent Systems; Environmental Engineering; Fluid and Plasma Physics; High Performance Computing; Mechanical Engineering; Open Source Software; Performance Evaluation and Benchmarking; Software Engineering
Abstract
Propellers and turbines, spinning blades on drones, aircraft, ships, and wind turbines, have been modeled since before powerful computers. Engineers built quick “low-order” tools such as blade-element models (slice each blade into small segments), actuator-line models (represent the blade as a line of force), and vortex methods (capture swirling wakes). These tools are fast and great for early design exploration but can miss important flow features. By contrast, high-fidelity computational fluid dynamics (CFD) resolves those details but is often too slow for everyday use. This project blends the best of both worlds. We keep the fast physics model and train a neural network to learn the missing piece, the difference between low-order predictions and high-fidelity results. This “grey-box” approach respects physics while data improves it. We focus on practical performance measures, how much push (thrust) and how hard to spin (torque), rather than every pressure or velocity detail, keeping the method simple for early design. Summer work may include building a benchmark dataset, designing and training neural network, visualizing results, and packaging a reusable tool. Students can focus on coding, physics, or visualization based on interest. The outcome: better early designs for aircraft, ships, and clean-energy systems.
Desired Skills
• Curiosity about physics and data • Some programming (Python or MATLAB—or any language) • Basic data handling/plotting (matplotlib, or similar) • Comfort with algebra/calculus; willingness to learn new tools or mathematics • Willingness to learn, ask questions, and work as a team Nice to have: basic ML; exposure to aerodynamics/CFD; familiarity with HPC clusters; interest in aircraft, drones, or clean energy.
Lightning Talk Title
Faster, Trustworthy Fluids: Al + physics for Propellers, Wings, Aerosols.
Keywords
propellers, turbines, drones, aircraft, airfoils, aerosols, droplets, ice, turbulence, aerodynamics, thrust, torque, machine-learning, neural-networks, design optimization, climate, air-quality, SINDy