Research Projects

The proposal aims to develop a physics‑informed machine learning (PIML) framework that can reconstruct detailed flow fields velocity, density, and pressure from high‑speed aerodynamics experiments, even when only sparse measurements are available. To achieve this, it combines Momentum Potential Theory (MPT), which uses Schlieren images to identify the flow into its irrotational (acoustic and thermal) via Helmholtz decomposition, with Physics‑Informed Neural Networks (PINNs), which embed the Navier–Stokes equations directly into the learning process. In practice, high‑speed Schlieren data from single and interacting sonic jets will first be processed through the MPT-based solver. Momentum density fluctuations (density times velocity) serve as source terms in a Poisson equation to extract the irrotational (acoustic) component. Next, Physics informed neural networks will be trained on space‑time coordinates and these MPT outputs, with automatic differentiation enforcing both the governing equations and MPT‑derived constraints. This ensures that the PINNs can infer complete pressure, density, and velocity fields reliably, even when experimental data is limited. The project is organized into three phases. In Phase 1, the MPT solver will be developed and validated using well controlled sonic jet experiments as a baseline, ensuring that the Helmholtz-based decomposition accurately isolates the irrotational components. Phase 2 will focus on designing and training the PINN: here, sparse pressure measurements or low fidelity CFD data will be integrated as needed, and the loss function will combine MPT‑derived constraints with the Navier–Stokes residuals enforcing conservation of mass, momentum and energy via automatic differentiation to achieve high accuracy and physical consistency. In Phase 3, the combined MPT–PINNs framework will be applied to more complex flow problem, such as hypersonic boundary layer transition over flat plate (sharp and blunt leading edge). Throughout all phases, experimental observations and auxiliary data such as pressure sensor readings will guide model refinement, while high‑fidelity numerical simulations will serve as a validation benchmark to verify that the MPT and Navier–Stokes–based losses produce reliable, quantitative reconstructions of the flow fields. By merging MPT’s physically grounded decomposition with PINNs’ data‑driven enforcement of flow physics, this research approach will provide more quantitative insight into high‑speed flow experiments without the full computational cost of DNS or LES. The resulting tools will accelerate research in aeroacoustics, shock–boundary layer interaction (SBLI), and hypersonic boundary layer transition prediction and modeling. Moreover, it will equip Indian research institutions with advanced PIML capabilities to foster industry collaboration over critical aerodynamic technologies and support hypersonic technology development.