LES of turbines: Effect of upstream disturbances

High fidelity 'quasi DNS' type simulations are carried out to explore the effect of upstream disturbances on the losses produced in low pressure turbines. We considered the individual and coupled interaction of upstream disturbances typically observed in gas-turbine engines like wakes, free-stream turbulence and the state of the incoming boundary layer. Additional effects of surface roughness due to in-service degradation of turbine blades are also considered.

Real & distributed roughness effects on transition

Fundamental studies are carried out to explore the effect of surface roughness on the transitional and turbulent boundary layers. Highly resolved DNS are carried out using a high-order in-house solver COMP-SQUARE. The study considers different configurations which include zero/adverse pressure gradients with real/distributed roughness. We extract dominant modes in the flow using low order strategies like Proper Orthogonal Decomposition (POD) and Dynamic Mode Decomposition (DMD).

Mixed-Fidelity frameworks for coupled interactions

Next-generation aircraft engines operate at a high bypass and low fan pressure ratios. The intakes of such aero engines have larger diameters and thus require a shorter length and slimmer lips to compensate for an increase in drag. When compared to conventional intakes, shorter intakes suffer from reduced diffusion capability which causes the flow to separate more easily on the nacelle. There is also an increased aerodynamic interaction between the intake and downstream fan. The overall performance is sensitive to this interaction of the separated flow with the fan. Various approaches like eddy-resolving simulations / mixed-fidelity approach are used to study the intake-fan interaction under off-design conditions of High-incidence and Crosswinds. We are also exploring different flow control strategies like plasma flow control / vortex generating jets to control flow separation.

High-fidelity Solver (COMP-SQUARE) Development

For most of the eddy-resolving simulations, we use the in-house solver, COMP-SQUARE which I have developed during my research fellowship in Cambridge. It is a structured compressible flow solver written in the generalized (curvilinear) coordinate system. This solver has been developed largely based on the numerics discussed in Visbal & Gaitonde (JCP, 2002). The solver uses upto sixth-order compact schemes for spatial discretization and fourth order explicit Runge Kutta for temporal integration. Multi-block communication is achieved using Message passing Interface (MPI). Local artificial diffusivity is used to capture shock waves in the supersonic flows.

The solver is capable of running on both CPUs using MPI parallelism and multi-node multi GPUs, thanks to the support of NVIDIA and CDAC. Interested candidates with good coding skills (Fortran/C/C++/CUDA) can contact me on nrv@iitm.ac.in

AS6041 Advanced CFD Course

As a part of this course, students develop a high order CFD solver entirely from scratch. We test the efficacy of the basic solver on canonical test cases. Subsequently, each student is assigned a project to enhance the basic code. Some of the interesting enhancements from this course are shown here. (wall-distance of a burning propellant (Hemanth Chandravamsi), COVO & Double periodic shear layer on Dynamically deforming grids (Achu Shankar), Shock He bubble interaction (S Navneeth), etc)

Low Order Methods and Data Driven Modelling

Our research group is actively utilizing machine learning techniques to develop low order methods. Projects involve exploiting the high-fidelity data from LES/DNS to train CNN/DNN/Tensor based NNs to improve low order methods. Most of the training is done on our in-house GPU server which currently is equipped with V100 GPUs (funded by DST/SERB and ICSR) and Quadro P6000 GPU (generously donated by NVIDIA)