Large Eddy Simulation as a design toolfor aero …...Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués

Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Large Eddy Simulation as a design tool for

aero-engine combustors

3rd MUSAF Conference, 28th October 2016

S. Richard J. Lamouroux G. Taliercio T. Lederlin

Safran Turbomeca became Safran Helicopter Engines in may 2016


Not a scientific - but maybe useful - presentation (!), more a high fidelity

CFD end-user vision

• Scope of CFD for combustor design

• Relevancy and feasibility of LES

• State of the art of LES (for combustion) at Safran HE

• Challenges for LES of real combustors

• Outlooks for Hifi CFD methods in the aero-propulsion industry

Objectives of this talk

3rd MUSAF Conference 20162


Combustor design, a complex trade-off

• Thermodynamic efficiency (pressure drop, combustion

efficiency)

• HP core lifetime (temperature profiles@combustor exit,

combustor liner temperature/gradient maps)

• Pollutant emissions (focus on NOx and soot)

• Operability (Lean-Blow Off, Ignition/Relight, combustion

stability)

• Non-CFD related constraints : engine integration,

manufacturing, costs, maintenance

CFD in the combustor design process



3 main types of CFD exploitation

• Physics analysis and understanding (after or in parallel to

experimental campaigns / engine failure or malfunction) -

Nsimulation~O(1-10)

• Iterative design – selective design (prior to experimental campaigns

– reduction of the design space) - Nsimulation~O(10-100)

• Optimization – robust design (in replacement /drastic reduction of

experiments – best configuration for given targets) -

Nsimulation~O(100-1000)

CFD in the combustor design process



Matrix representation of the possible CFD scope


CFD type of use Physics analysis Iterative /

selective designOptimization/RD

Pressure drop

Combustion

efficiency

OTDF/RTDF

profiles

Liner

temperature

NOx

Soot

Lean-blow-off

Combustion

stability

Ignition

Thermodynamic

efficiency

HP core lifetime

Pollutant

emissions

Operability

Design targe vs CFD use

Design targets

Some questions to be addressed for each cell

of this matrix

- Relevancy of HiFi CFD (precision vs needs) ?

- Feasibility of HiFi CFD in an industrial

context (cost & return time) ?

- Availability of Hifi CFD tools?


Relevancy of LES for combustor design


Combustion in an aero-engine

• Is first a complex (and anisotropic) problem of mixing!

Progressive introduction of air in the burner (~60-70% of

air do not directly participate to combustion)

• Is a “moderate” turbulent Reynolds number problem

(Ret~5000-10 000) : ratio of the largest to the smallest

scales of the flow ~ 500 – 1000 in the vein � narrow

turbulence spectrum

• Has highly unsteady and intermittent characters

(turbulence intensity~10-20%) � non-linear coupling

between turbulence and chemistry difficult to model

Combustor physics features all the ingredientsto efficiently apply Large Eddy Simulation

Primary zone view (2000 frame/s)

liner cooling

primary holesinjectordilution hole

80%

12%

8%

20%

20%40%

1/r

Turb

ule

nce

Sp

ect

rum

lt ηk

resolved

modelled


Industrial feasibility of LES


Equivalent CPU hours on Intel

Ivy Bridge architecture

• LES can be predictive (depends

on the numerics, models, user)

but is ALWAYS expensive

• Optimization/Robust Design for

operability through the use of

“brute force” LES is today not a

reachable target

Pushing LES for design is a strategic choice implying de dicated means. There’s still an important need for lower-order modelli ng (or multifidelity) in which LES can play a role

Physics analysis Iterative /


Pressure drop

Combustion

efficiency

OTDF/RTDF

profiles

Liner

temperature

NOx

Soot

Lean-blow-off

Combustion

stability

Ignition

Thermodynamic

efficiency

HP core lifetime

Pollutant

emissions

Operability


CPUhReturn time (base 2000 cores)


Fidelity vs Quantity

• Natural trend to refine the grid for reducing CAD

simplifications and sensitivity to modelling

• Maturity of LES yields a rapid growth of the

number of configurations to compute (incremental

design improvement, multiple design points

simulation…)

• CPU needs for productive LES has already started to

exceed supercomputer power evolution �� CPU

consumption for combustion has been multiplied by

100 in the last five years at SAFRAN HE

Safran nom de l’activité / Confidentiel / Date / Direction (menu "Insertion / En-tête et pied de page")8

Link with meshing capabilities

0,01

0,1

1

10

100

1000

10000

100000

1995 2000 2005 2010 2015 2020 2025

Standard

Challenge

Number of elements per burner sector

(10-20 sectors in the full burner)

X1 sim.

X5 sim.

X50 sim.

Number of elements per burner sector

(10-20 sectors in the full burner)

1

10

100

1000

10000

100000

1990 1995 2000 2005 2010 2015 2020 2025 2030

CPU need for

productive LES

CPU power

increase (moore)

No

rma

lise

dC

PU

co

nsu

mp

tio

n

Both fidelity and quantity will continue growing in the coming years but maybe not together (multifidelity LES?)

(in M elements)


LES code availability for industrial use


A recurrent debate : Engineering vs First principles LES

First Principles LES Engineering LES

Expert user Broad range of users

High quality grid Poor quality grid

Non-dissipative numericsRobust (and dissipative)

numerics

Massively parallel and

scalableMinor parallelism

Comprehensive physics Main phenomenology

Basic rules

• Robustness / user dependency : mesh generation

should obey strict rules (grid skewness, refinement)

– LES requirements drive grid generation, not the

contrary

• Numerics : A LES code should have proven good

turbulence transport and dissipation capabilities on

simple configurations (vortex convection, HIT…)

• HPC capabilities are mandatory

• Modelling : Physical models implementation should

be validated on basic experiments AND confronted

to complex applications within the same software

(strong coupling with numerics)Is this still LES?

LES is not a quick-win business but necessitates long-term investments and

partnerships with experts from academia



AVBP YALES2

Equations Compressible Low-Mach var. density

Numerics2nd/3rd/4th order in space,

2nd/3rd order in time

2nd/4th order in space,

4th order in time

Parallelism Massive Massive

Grid Hybrid unstructured grids Hybrid unstructured grids

Turbulence

modellingSmagorinsky/Wale/Sigma Smagorinsky/Wale/Sigma

Walllaw-of-the wall/specific

effusion cooling model

law-of-the wall/specific

effusion cooling model

Turbulent

combustion

TFLES/PCM-

FPI/CFM/FTACLESTFLES/PCM-FPI/FTACLES

NOx : tabulated NOMANI

model

NOx : tabulated NOMANI

model

Soot : Hybrid Tabulated /

Leung model

Two-phase flowEulerian and Lagrangian

formalism with fuel filmingLagrangian

Pollutant

State of the art of LES at Safran HE

10

Codes… … and workflow

• More than 1000 degrees of freedom to set-up

for a combustor sector calculation (initial &

boundary conditions, models parameters) � Hifi

production needs Hifi set-up

• Complete engineering work-flow from

combustor sketch to post-processing developed

with CERFACS and CORIA


LES field of use in combustion


Computational Readiness Level

• Use of both compressible (AVBP)

and low-Mach (YALES2) LES

solvers

• CPU-limited deployment of

optimization and Robust Design

• Return time challenged by in-

house experimental capabilities

for iterative design applications

Operability is still an issue from bothmodelling and CPU cost points of view

HiFi CFD maturity

Compatible

with Design

Activities

Currently

unfeasible

Physics analysis Iterative /


Pressure drop

Combustion

efficiency

OTDF/RTDF

profiles

Liner

temperature

NOx

Soot

Lean-blow-off

Combustion

stability

Ignition


Thermodynamic

efficiency

HP core lifetime

Pollutant

emissions

Operability

Current zone of possible

applicability


Some achievements of LES applied to combustion


Lean blow-off limit prediction

• Simulation of a fuel-air ratio transient from stable

combustion to extinction (like in experiments)

• Extinction detected by LBO precursors like

unburnt fuel at the combustor exit

• Detailed analysis of the blow-off mechanisms

Pre

curs

or

Case 1

Case 2

LBO limit

Time

�

�

� �

Normalized LBO limits for

several combustors



Low NOx / Low-Smoke combustor technology analysis

• NOMANI tabulated model for NOx (prompt, thermal)

Side viewT

NOx formation

very sensitive to :

• temperature

• FAR

• residence timeLP

injectors

combustor 1

combustor 2

combustor 3

combustor 4

combustor 5

FAR

Pilot

injector

Trends and order of magnitude can be predicted



Soot mass fraction

Soot number density

Injection wheel

cumulated mean

instantaneous

• Intermittent formation/oxydation of soots

• No negligible source term : balance makes soot level prediction very difficult

Low NOx / Low-Smoke combustor technology analysis

• Hybrid tabulated (gaseous species) / modified Leung (soot) model



Soot concentration converted into smoke number for comparison to experimental data

obtained at the engine test (ICAO correlation)

• Impact of injection technology and

operating condition are well described

• Exact smoke number values are still

difficult to predict (lack of detailed

physics)

• Fuel composition impact is correctly

accounted thanks to detailed chemistry

if

if

GTL (paraffinic

fuel)

Jet A (low

aromatic content)

Jet A (high

aromatic content)

smok

enu

mbe

r




RTDF profile set-up

��(�) =�� (�)− �4

�4 −�3

combustor inlet combustor exit

RTDF[%]0Hub

Tip

Radius

(normalised temperature)


Challenges for LES of complete combustors


Many efforts in the past for turbulence / turbulent combustion modelling

• High level of fidelity reached on swirler aerodynamics

• Temperature and main species concentrations fields also predicted

• Grid convergence demonstrated

axial velocity


Challenges for LES of complete combustors


… but restricted to isolated injectors and often “low pressure/pre-heated”

conditions hiding major bottlenecks to be found as soon as LES is the subject

of an industrial use

• How to account for cooling devices (effusion cooling, rings…) or properly validate the

associated models? � represents up to 60% of the combustor air-flow rate!

• Excessive filtering/thickening of the flames at relevant pressure/temperature

conditions � validity of the sgs closure? / imply huge residence time modifications

• Fuel air ratio at the combustor exit far from flammability limits � chemistry

validation under high dilution (especially for tabulated approaches – slow

oxidation/post-oxydation chemical paths not included in look-up tables)?

• Pre/post-processing : mesh generation, local refinement / data storage and transfer

for analysis (0.5 Tb/simulation)


Cooling devices modelling


Hole resolution generally limited to primary /

dilution holes

• Correctly resolving the pressure loss (and

the air-mass flow rate) within holes

requires around 25-30 points following the

literature

~O(5mm) ~O(0,5mm)



1

10

How far can we solve cooling systems?

Big hole

Small hole

Today

… do we have to wait for 2039

and keep relying on

homogeneous “porous wall”

approaches

• Only few and recent attempts

at CERFACS and UNIFI to

propose heterogeneous

models based on local suction

and injection

Zone of interest

thickened holeapproach


Filtering/thickening of flames at relevant pressure and temperature


The flame (heat release, main species

gradients) should be resolved on a few

grid points (~5-10)

• The real flame thickness δflame is by far

lower than the grid size and needs to be

filtered or thickened at a scale ∆comb~5-

10∆x

• The higher the combustor pressure (and

therefore temperature), the greater the

flame is thickened

• Some concerns about the validity of sgs

models at high pressure levels / strong

modification of the residence time in

reaction zones (impact on polluants having

long chemical time scales)

Operating pressure (bar)

0

50

100

150

200

250

0 10 20 30 40 50

Série1

Série2Primary zone

Injector vicinity

∆co

mb

/δfl

am

e

∆x~200µm

∆x~600µm


Conclusion


Physics (unsteadiness, Turbulent Re) makes LES particularly relevant for being intensively applied to

aero-engine combustors design

A strong interaction between numerics development, model implementation and validation is

needed to assess the codes validity before applying them to design activities

The LES industrial maturity perimeter extends rapidly, more than CPU capabilities offered by the

Moore law � dedicated resources will be needed in near the future : LES is not a simple evolution

of CFD but a strategic orientation

Efficient workflows dedicated to HiFi methods are mandatory : LES has been a tool for CFD experts

in the last 20 years but bridges should now be established for end-users


Outlooks for Hifi-methods in the aero-propulsion industry


The extension of LES to the fields of Robust Design and Uncertainty Quantification will need for a

drastic reduction of grid generation efforts (CAD and meshing software, automatic/dynamic &

robust mesh refinement strategies)

• This is particularly true with the arising of 3D printing in the aeronautic market

• The time needed for setting up a simulation is still close to the one for running it

LES is today mature for simulating nominal engine operating conditions, but important modelling

improvements are still required for addressing operability issues and accounting for the fuel

composition influence : dedicated two-phase flow combustion models, multicomponent approaches for both

physical and chemical behaviors, ignition, …

Data management (including storage…) and post-processing is becoming an issue : • Methodologies have to be developed for getting much than averaged (and RMS) fields from unsteady

simulations, with objective criteria to help analyzing/comparing cases

• Multi-fidelity approaches for data analysis (?) : resolution requirements for numerical accuracy are not

necessarily the same as for physics analysis


Acknowledgments

Many thanks to our partners CERFACS and CORIA daily supporting SAFRAN in the

development and use of LES for design activities

Acknowlegments to our colleagues from SAFRAN Aircraft Engines for providing part

of the material of this talk

Special thanks to

A. DauptainCERFACS

G. StaffelbachCERFACS

V. MoureauCORIA

J. DombardCERFACS

Y. MerySAFRAN

Aircraft Engines

A. FiguerSAFRAN

Aircraft Engines

Q. MaléSAFRAN

Helicopter Engines

R. MercierSAFRAN Tech


Safran nom de l’activité / Confidentiel / Date / Direction (menu "Insertion / En-tête et pied de page")25

Documents

Large Eddy Simulation as a design toolfor aero …...Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués