Flows in several turbomachinery components (low- and high-pressure turbines/compressors) are extremely complex due to the presence of separation, transition and relaminarization. Predictions of each of these three phenomena are beyond the current abilities of turbulence models. We employ high-fidelity approaches including DNS, LES and hybrid LES/RANS for investigation of losses and heat-transfer components in these flows. A high-end native code ANUROOP, which can work with complex configurations and exploit GPUs, has been developed exclusively for this purpose.

Understanding the external aerodynamics of aircraft, missiles, rockets, spacecraft, UAVs is crucial for their design and development. We perform unsteady high-fidelity simulations to analyze aerodynamics of these aerospace vehicles encompassing the full flight envelope. Current focus areas are- (1) high-lift analysis with the deployment of flaps and slats, (2) optimal wing designs for turboprops with consideration of wing-propeller interactions, and (3) unsteady flow analysis in cargo aircraft with upswept afterbody configurations.

Development and improvement of turbulence and transitional models for practical applications are still one of the most important problems in relevant industries. To address this, we augment modern theoretical knowledge and high-end datasets with state-of-the-art numerical approaches to improve existing models or propose new models. Currently, the efforts are directed towards- (1) improve transition models based on DNS/LES data, (2) characterize reverse transition (relaminarization) and propose models in the framework of RANS, (3) include curvature and displacement effects in boundary layer simulations.

Spatio-temporal datasets obtained from a flowfield, either from an experimental or a numerical approach, are usually high-dimensional and analyzing them require optimal tools. We employ state-of-the-art statistical, spectral and low-order decomposition (POD/DMD) techniques to gain useful insights from these datasets. Further, we also build reduced-order models based on decomposed data. Our group has fully parallel POD/DMD codes that can work on several million grid points.