April 2nd, 2015I 1User meeting – Chatou, France
April 2nd, 2015
User meeting – Chatou, France
Automotive fan system simulation with Code_Saturne
April 2nd, 2015I 2User meeting – Chatou, France
OutlineIntroduction
Virtual development / optmization processes
ANR project Pepito
Motivation / objectives
Demonstrator
Simplified case description
Methodology and workflow
Study on CPU time and comparison with commercial software
Industrial case
Description of the test rig simulation / Simulation set/up
Results and comparisons with experimental data
Conclusion and perspective
April 2nd, 2015I 3User meeting – Chatou, France
April 2nd, 2015
User meeting – Chatou, France
Introduction
April 2nd, 2015I 4User meeting – Chatou, France
Virtual development
Simulation is a strong asset in the design process:
Faster development cycle
Deep analyze of the physics
Reduced costs...
Cooling module sizing and development
Introduction of simulation
Automated simulations
Optimization
2014: CAD in 1 hours
FEA & Rheology
Acoustics Design
Performance
prediction
System
integration and
validation
April 2nd, 2015I 5User meeting – Chatou, France
Optimization process
Optimization (Isight)
NOLH sampling for DoE (statiscaldistribution for all factors)
Response surface model (RSM) with radial base functions
Genetic algorithm NSGA2 for research of optima (ranking along generation)
Numerical DoE Response surface Genetic algorithm
Standard Optimized Non conventional design
Fan design
11 parameter DOE (10 geometrical, 1 physical parameter)
Optimized designs are often unconventional, and take into account the real physics (3D effects, tip recirculation, hub,…)
April 2nd, 2015I 6User meeting – Chatou, France
ANR project Pepito
Plan d’Expérience Pour l’Industrie du Transport et l’Optimisation
Fan system case (turbomachine simulation)
Optimization in large dimension
Domain extension up to 60 factors
Wide range of variation for each factor
High Power Computing
Several thousands of simulation
Open-source code for parallel computing
(unlimited number of case running together and
each on hundreds of core )
Code_Saturne selected for accuracy and quality
assurance
April 2nd, 2015I 7User meeting – Chatou, France
Motivation / objectives
Handle usage of Code_Saturne
Test and implement simulation methodology for fan system
Assess code performance
Size case and DoE according to CPU ressources
Estimate accuracy of results used for surface response models
Demonstrator Full test-rig simulation
April 2nd, 2015I 8User meeting – Chatou, France
April 2nd, 2015
User meeting – Chatou, France
Demonstrator
April 2nd, 2015I 9User meeting – Chatou, France
Simplified case description
Simple and realistic geometry
Small inlet and outlet domains
3 blade fan with tip clearance
Axi-symmetrical case
Relevant test case for turbomachinery
Plane and circular interfaces
Multiple Reference Frame for steady simulation
Sliding mesh for unsteady simulation
Easy use for comparisons
Small simulation case, i.e. ~ 1,2 million tetrahedral cells
April 2nd, 2015I 10User meeting – Chatou, France
Simulation processes
Geometry design (Catpart)
Mesh generation
Solver
Post-processing
Geometry design(.nas)
Mesh generation
Pointwise / Hypermesh
Solver
Code Saturne
Post-processing
Paraview
Actual process
(commercial software)
New process
(with Code_Saturne)
Tested and developped during Pepito project
To be further used in an industrial context
April 2nd, 2015I 11User meeting – Chatou, France
Results : global performances (steady)
Result comparisons
Good agreement between codes on global performances
To be completed with a deeper analysis of the flow, even if the interest for this fan is
limited
-50
0
50
100
150
200
250
0 0,2 0,4 0,6
Δp
(P
a)
Mass flow rate (kg/s)
ΔP steady simulation
Saturne
Commercial code
0
0,05
0,1
0,15
0,2
0,25
0 0,1 0,2 0,3 0,4 0,5
To
rqu
e (
N.m
)
Mass flow rate (kg/s)
Torque steady simulation
Saturne
Commercial code
April 2nd, 2015I 12User meeting – Chatou, France
Results : global performances (unsteady)
Result comparisons
Same performance with Code_Saturne between steady and unsteady simulations
Lower pressure levels with unsteady simulation for the commercial code
On-going investigation to explain the difference:
Separation effect on the profile not correctly predicted (squared thick leading edge)?
-50
0
50
100
150
200
250
0 0,1 0,2 0,3 0,4 0,5
Δp
(P
a)
Mass flow rate (kg/s)
ΔP averaged on one blade passage
Saturne
Commercial code
0
0,05
0,1
0,15
0,2
0,25
0 0,1 0,2 0,3 0,4 0,5
To
rqu
e (
N.m
)
Mass flow rate (kg/s)
Torque averaged on one blade passage
Saturne
Commercial code
April 2nd, 2015I 13User meeting – Chatou, France
Influence of the solver precision on accuracy
0
50
100
150
200
250
300
350
1 401 801 1201 1601
Δp
(P
a)
Iterations
Effect of solver precisionPrecision 10^-8
Precision 10^-3
Precision 10^-5
Commercial code
0,00E+00
1,00E-01
2,00E-01
3,00E-01
4,00E-01
1 401 801 1201 1601
To
rqu
e (
N.m
)
Iterations
Effect of solver precisionPrecision 10^-8
Precision 10^-3
Precision 10^-5
Commercial code
solver
precision
Δp (Pa)
ave. last 400 ite.
Torque (Nm)
ave. last 400 ite.
10-8
(default
Value)
199 0,194
10-5 194 0,196
10-3 197 0,197
Identical convergence curves for both codes over
2000 iterations
Discrepancy of about 5Pa (~2,5%)
Small oscillations with lower precision
April 2nd, 2015I 14User meeting – Chatou, France
Solver Precision N_cycle ( Pressure) N_cycle ( Velocity) Iterations/Hour
10-8 (default Value) ~400 ~30 719
10-5 ~95 ~15 1290
10-3 ~20 ~5 1714
Influence of the solver precision on calculation time
2,4
Reduced number of cycle per iteration with lower precision target
Potential gain on simulation time (divided by 2) if ~2,5% discrepancy is acceptable
Unknown risk on the robustness of the solution, 10-5 might be a good compromise
1,8
April 2nd, 2015I 15User meeting – Chatou, France
Results: comparison of CPU time
167 45 310
50
100
150
200
Tim
e e
lap
sed
(m
in)
Wall clock time for 2000 iterations (steady)
Saturne defaultparameters
Saturne optimizedparameters
Commercial code
22 570
10
20
30
40
50
60
Tim
e e
lap
sed
(m
in)
Wall clock time for one revolution in 100 steps (unsteady) Saturne
Commercial code
Default parameter with Code_Saturne can not compete with commercial code in term of
speed for steady simulation
Adapted parameters are in favor of CPU time over robustness and accuracy. It is
necessary to validate such a set up (some discrepancies observed).
Code_Saturne is very competitive for unsteady simulation (even with default parameters)
adapted
April 2nd, 2015I 16User meeting – Chatou, France
April 2nd, 2015
User meeting – Chatou, France
Industrial case
April 2nd, 2015I 17User meeting – Chatou, France
Full test-rig simulationDetailed description of both fan and labs (masking effect of the torquemeter, ground and wall description, etc…)
Experiment Simulation
Geometry and domain of simulation
April 2nd, 2015I 18User meeting – Chatou, France
Mesh and simulation set-up
Meshing process (Pointwise)
2D meshs generated with expert model
Unstructured mesh (tetrahedral)
Refinement and smooth transition in critical area (tip clearance, leading and trailing edges, etc…)
Automation of the mesh process with scripts
RANS simulation
RANS and URANS simulation
K-ω turbulence model, two-layer model for boundary conditions
Monitoring of global performances and residuals to check convergence
Fan surface mesh
Tetrahedral mesh on a profil
April 2nd, 2015I 19User meeting – Chatou, France
Very good agreement between experiment, reference simulation and Saturne predictions for pressure
rise.
Slight offset between experiment and CFD for torque (on-going investigation on experimental set-up).
Numerical / experimental result comparisons
-50
50
150
250
350
450
550
1000 2000 3000 4000 5000
Δ(P
a)
Volumic flow rate (m^3/h)
ΔP industrial case
Saturne
Reference CFD : currentprocess with commercial code
Experimental results
0
0,5
1
1,5
2
2,5
1000 2000 3000 4000 5000
To
rqu
e (
N.m
)
Volumic flow rate (m^3/h)
Torque industrial case
Saturne
Reference CFD : currentprocess with commercial code
Experimental results
April 2nd, 2015I 20User meeting – Chatou, France
April 2nd, 2015
User meeting – Chatou, France
Conclusion and perspectives
April 2nd, 2015I 21User meeting – Chatou, France
Overall conclusion
Code_Saturne tests at Valeo
Full simulation process for Code_Saturne successfully experimented
Accuracy validated for our industrial cases
“CPU” performance assessed for steady and unsteady simulation
Applications for optimization
Workflow for full system simulation under development (cooling module with fan system)
Design of Experiment to be conducted with large number of parameter
Other applications identified, to be deployed in R&D department
April 2nd, 2015I 22User meeting – Chatou, France