Share This
//Fluid Dynamics
Ph.D. Research

Large-scale computations
of turbulent flows:

numerical experiments & HPC

Complex turbulent flow patterns are common in many engineering applications including aerospace, automotive, chemical, biological, and other flow systems. To understand the physics that govern the dynamics of unsteady motions in a turbulent flow its essential to conduct spatially and temporally-resolved numerical investigations. The focus of this work is bifold: physics and numerical development. On the physics front, this work leverages exisiting high-performance computers (HPC) to conduct large-scale numerical experiments of turbulent flow separation and modulation using passive-flow techniques. The numerical development work exlores a modern framework that can extract benefits from the many-core, extreme-scale concurrency of next-generation HPC platforms.

//Fluid Dynamics
Flow Separation.
Flow separation
Passive Flow-Control

Modulation of flow
over a backward-facing ramp with a
wall-mounted cube

Flow separation is a common phenomenon in many engineering applications. Often flow separation from the surface yields undesirable effects such as increase in drag forces, decrease of lift, reduced mixing, etc. Flow control devices help in mitigating such undesirable effects by delaying or mitigating flow separation. Passive vortex generators (VGs) are one such form of flow control devices that are immersed in the turbulent flow and entrain the momentum from outer mean inflow.

In this study, we use a submerged, wall-mounted cube of height smaller than the turbulent boundary layer thickness as a passive VG. Previous knowledge of flow patterns that form around a cube along with it's simple design makes this study canonical. Flow separation and modulation on a backward-facing ramp is studied with the aim to investigate the interactions of flow features in the separated region with those around the vortex-generators. The dependence of these interactions on the cube configuration is studied by varying the size and its upstream position from the leading ramp edge. The features of the study are:

Spatially-Evolving Turbulent boundary Layer

The synthetic inflow prescribed at the inlet mimics real turbulence and saves computational cost. For details check paper by Shinde et at. (2017).

Large-Eddy Simulations

Wall-resolved large-eddy simulations are conducted to resolve the small-scales in the separated region and to investigate their interactions that enable flow modulation.

//Numerical Modeling
Enabling simulations of tubulent flows at exascale and beyond

Parallel & distributed framework to
facilitate extreme-scale computations
of compressible turbulent flows

The next generation computational models will enable advances in the field of climate science, ocean flow behavior, space sciences, biology, complex materials and other areas of scientific research. To meet the growing needs in science, the computational capability of the fastest supercomputers must continue to grow. Driven by limitations in power consumption, memory bandwidth and latency, and high demand in accuracy, the next generation HPC sites for exascale computing are expected to feature heterogeneous architectures. The architectural developments warrant the need for new model development and redesign of existing algorithms. These challenges limit the use of traditional numerical methods used in computational studies of fluid mechanics at extreme-scales largely due to poor scalability and low performance portability. High-order methods such as the discontinuous Galerkin method (DG) provide benefits that mitigate the above issues. Apart from being arbitrary high-order schemes, these methods offer the advantage of high-scalability, in part due to the compact dependence on neighbors. The features of the study are:

Recovery-Assisted Discontinous Galerkin Method (RaDG)

The recovery operation can be configured to get a higher-order approximation of the solution. For details check paper by Johnson & Johnsen (2017).

Power-Performance Balance

Mixed-precision framwork enables power-performance tunability. Preliminary results indicate savings of 1-2 W per node which is expected to yield major cost benefits at extreme-scales.

Parallel & distributed framework

Multi-CPU & GPU compute capability to enable extreme-scale simulations on hetrogenous architectures. Scaled upto 2,000 nodes on Summit Supercmoputer (ORNL) with 4 GPUs on each node.


My Featured Work

Near-wake flow modulation by a cube used as a passive vortex generator (VG) on a backward facing ramp. Presented at the AIAA 2018, Flow Control Conference held between June 25-29, 2018, Atlanta, GA.

Get this paper

Want to talk about a Project ?

Get in Touch