You're invited to attend
"Explicit Filtering Large Eddy Simulation and RANS Training Methods for Turbulent Reacting and Non-reacting Flows"
by
Santosh Hemchandra
Professor, Department of Aerospace Engineering
Indian Institute of Science, Bangalore
Thursday, June 22
2:00 p.m.
Food Processing Tech Auditorium 102 in the NARA complex
Refreshments will be served
Abstract
Large eddy simulation (LES) for turbulent flows aims to provide physically realistic and time
accurate predictions of the dynamics of a range of large length and time scale flow motions.
The expectation from this approach is that of achieving improved prediction accuracy of the
statistics of flow field quantities over traditional Reynolds/Favre averaged methods while
realistically matching available computational resources. Additionally, LES is expected to be a
source for training data needed data driven flow prediction and analysis methods. For both
these reasons, realizing accurate LES for technologically relevant turbulent flows in realistic
geometries is an important research problem.
The explicit filtering LES (EFLES) method formulated by Joseph Mathew at IISc along
with Rainer Friedrichs and other colleagues at TU Munich, is one potential approach to
achieving reliable LES. EFLES is formally derived from the approximate deconvolution
modeling (ADM) approach to LES. Formally, the method involves evolving the flow state using
a stable time accurate numerical method and applying a low pass filtering step on the
computed fields at every time step. The latter ensures that turbulent kinetic energy generated
by large scale resolved motions does not “pile up” at the smallest scale and contaminate the
dynamics of large scales. In addition, with mesh refinement, EFLES smoothly reduces to a DNS
with no additional modification of the method.
The talk will present a recent extension of this method to reacting turbulent flows
developed by my research group in IISc and show examples of computations from published
and ongoing work. The simulation studies that will be discussed are as follows. First, a
premixed turbulent round methane-air jet flame at an equivalence ratio of 0.8 and unburnt
gas temperature of 800 K. The nominal 1D premixed flame speed and thermal thickness are,
𝑠! = 2.0 𝑚𝑠"# and 𝛿! = 300 𝜇𝑚 respectively. The nominal turbulent Reynolds number,
𝑅𝑒$ = 34 and Karlovitz number, 𝐾𝑎 = 25, places the flame in the thin reaction zones regime
of turbulent premixed combustion. LES and DNS results for flow statistics will be compared
for this flame. The second case is a model swirl stabilized gas turbine combustor, PRECCINSTA
at DLR Stuttgart with a nozzle flow 𝑅𝑒 = 24,000. The final case will be results from ongoing
simulations of the IISc RQL rig and lifted non-premixed flame.
RANS simulation has the potential for being a reduced order model for complex
reacting flows that can be used for design optimization of combustors. However, RANS model
parameter uncertainties make solutions from RANS models unreliable. I will present recent
results from our turbo expo 2023 paper on learning RANS parameters from LES data using the
Bayesian inference approach for the premixed turbulent jet flame case. The method I will
present has been implemented using the “off the shelf” version of the reactingFoam solver
available with openFoam 2206. However, the method is general and can be adapted to work
with any RANS solver that allows for specifying RANS model parameters.
About the speaker
I’m an associate professor in the Department of Aerospace Engineering, Indian Institute of
Science (IISc), Bangalore, India. My research group’s focus is broadly on uncovering new physical insights into reacting and non-reacting flows that are of relevance to space and aircraft propulsion industry funded mainly by key industry OEMs in these areas. We work actively to transition these insights into design tools and practice to help mitigate fuel burn,emissions (soot, NOX) and other operability concerns in GT and rocket systems such as
thermoacoustic instability and jet noise.
We use methods such as stability analysis, large eddy simulations, experimental
measurements and data driven analysis that are aimed broadly at understanding coherent
unsteadiness mechanisms in combustor flows and in other related areas such as sources of
jet noise and wall-bounded flows. More recently, we’ve been working actively to incorporate
machine learning methods that can assimilate data from LES or experiments into simpler, less
computationally expensive models that are easily used for engineering design practice.
I earned my Ph. D. in 2009 from the Lieuwen group in the department of aerospace
engineering at Georgia Tech. and have been on the faculty at IISc since 2012.