EHTC 2008 in Strasbourg
Duct Design OptimizationDuct Design Optimizationof an of an HVACHVAC* System* System
Using Fast CFD SolverUsing Fast CFD SolverSC/TetraSC/Tetra
Figures Courtesy of Denso Co.
**HHeating, eating, VVentilating and entilating and AAirir--CConditioningonditioning
EHTC 2008 in Strasbourg
Cartesian MeshCartesian Mesh
Data Cleaner and TranslatorData Cleaner and Translator
Unstructured MeshUnstructured MeshMultiMulti--purpose CFDpurpose CFD
Cartesian MeshCartesian Mesh
MultiMulti--purpose CFDpurpose CFD
Construction / Architectures Civil engineering / ChemicalElectrical Appliances / Electronics
Automotive / Machinery Electrical AppliancesConstruction / Bio-chemicals
•Data Translation•Data Cleaning•Data Modification
MAIN INDUSTRIES:MAIN INDUSTRIES:MAIN INDUSTRIES:
MAIN INDUSTRIES:MAIN INDUSTRIES:MAIN INDUSTRIES:
Company Information
EHTC 2008 in Strasbourg
SC/Tetra37%
STREAM36%
Others21%
Chemistry6%
Engineering16%
Construction4%
Electricity16%
Machinery12%
Automobile25%
Company Information
EHTC 2008 in Strasbourg
CFD as a ShopCFD as a Shop--floor Technologyfloor Technology
Introduction to SC/Tetra
Design engineers can use SC/Tetra as a tool to establish a hypothesis on how things happen
With:
- User-friendly Interface and Usabilityfor those who don’t have time to read through manuals
- Powerful Automatic Mesherfor those who don’t use CFD very often
- State-of-the-art Postprocessorfor understanding thermo-fluid behavior intuitively
And more,
- Outstanding computational speed and accuracy
- Low memory consumption
EHTC 2008 in Strasbourg
User Friendly Interface and Usability -- NavigationNavigation
Introduction to SC/Tetra
EHTC 2008 in Strasbourg
Powerful Automatic Mesher– Powerful Powerful MesherMesher
Better Accuracy!
Introduction to SC/Tetra
EHTC 2008 in Strasbourg
解適合解析
The 3rd and 4th results are almost the same.
This shows that about 420,000 mesh elements with resolution like the 3rd
one is required forcertain accuracy.
Powerful Automatic Mesher–– Automatic Mesh RefinementAutomatic Mesh Refinement
Introduction to SC/Tetra
# of mesh elements: # of mesh elements: 600K600K
# of calc.: 4 times# of calc.: 4 times
EHTC 2008 in Strasbourg
Condition settingCondition setting
SCT preSCT pre
Powerful Automatic Mesher–– Automatic Mesh RefinementAutomatic Mesh Refinement
ImportImport
Targeted # of mesh elements
# of calculation
Targeted # of mesh elements
# of calculation
Mesh generationMesh generation CalculationCalculation
Post processingPost processing
New Octreefrom previous
computation result
New Octreefrom previous
computation resultSCT solverSCT solver
SCT postSCT post11
22
33 444
Introduction to SC/Tetra
EHTC 2008 in Strasbourg
3. Analysis of an HVAC* System
**HHeating, eating, VVentilating and entilating and AAirir--CConditioningonditioning
EHTC 2008 in Strasbourg
Cross-flow fan
Evaporator
1. Analysis prior to optimization
Analysis of an HVAC System
EHTC 2008 in Strasbourg
Existing Existing model model (staged)(staged)
Primitive Primitive model model (straight)(straight)
Analysis casesAnalysis casesAnalysis cases
Reason for staging:To obtain uniform
velocity distribution in the evaporator
Analysis of an HVAC System
EHTC 2008 in Strasbourg
Primitive Primitive model model (straight)(straight)
Analysis of an HVAC System
Analysis meshesAnalysis meshesAnalysis meshes
1.7 million mesh elements
EHTC 2008 in Strasbourg
Cross-flow fan
Evaporator
Moving mesh @2235rpmMoving mesh @2235rpm
Pressure lossPressure loss
Flow rateFlow rate
Pressure: 0 PaPressure: 0 Pa
Transient analysis: Time step 1 ms up to 0.5s
Analysis of an HVAC System
Analysis conditionsAnalysis conditionsAnalysis conditions
EHTC 2008 in Strasbourg
Results: Velocity distribution in EvaporatorResults: Velocity distribution in Evaporator
Non-uniformUniform
Analysis of an HVAC System
Existing Existing model (staged)model (staged) Primitive Primitive model (straight)model (straight)
EHTC 2008 in Strasbourg
Quantification of “Uniformity”Standard deviation of exit velocity
Quantification of “Loss”Difference in area-averaged pressurebetween inlet and outlet
Uniform velocity distribution in
Evaporator
MultiMulti--purposepurpose
Reduction of loss
2. Setting the 2. Setting the objectivesobjectives of optimizationof optimization
Analysis of an HVAC System
EHTC 2008 in Strasbourg
Bernoulli's theoryContracted
Small cross-section
Large velocity
Small pressure
Small velocity in Evaporator
3. Setting of optimization design parameters
Basic idea for Uniform velocity distribution
Basic idea for Basic idea for Uniform velocity distributionUniform velocity distribution
Magnitude of pressure
ContractContract
Analysis of an HVAC System
EHTC 2008 in Strasbourg
速度小
Confirmation of basic idea using CFD (1)Confirmation of basic idea using CFDConfirmation of basic idea using CFD (1)(1)
Analysis of an HVAC System
EHTC 2008 in Strasbourg
ContractSmall
velocity
Confirmation of basic idea using CFD (2)Confirmation of basic idea using CFD (2)Confirmation of basic idea using CFD (2)
Analysis of an HVAC System
EHTC 2008 in Strasbourg
Optimization methodsWorkflow structuring
Fast, robust calculation High quality mesh morphing
Optimization Engine
CFD SolverGeometry modification
tool
Optimization
EHTC 2008 in Strasbourg
Optimization Engine
Optimization Engine
Geometry modification
tool
Geometry modification
tool SC/TetraSC/TetraDEF
BAT
SL
PREMDFMDFMDFMDFMDF
BAT
parameters Results
Morphing parameters
Base meshMorphing definitions
Analysis conditions
Morphed mesh
EvaluationResult
Legend:System
FileControlI/O
Optimization
EHTC 2008 in Strasbourg
Control points forMorphing
Control points forControl points forMorphingMorphing
Optimization
MorphingMorphing
EHTC 2008 in Strasbourg
Design parameter:Range of morphing parameters
Sampling no.: 40
Design parameter:Range of morphing parameters
Sampling no.: 40
Information for sampling
Time limit for optimization
5 min.
Table of parametersTable of
parametersCase/ZoneCase/Zone G1G1 G2G2 G3G3 G4G4
1 0.017692 0.025385 0.015000 0.001846: : : : :: : : : :
40 : : : :
4.1 Sampling by 4.1 Sampling by Design of Experiments methodDesign of Experiments method
n Sampling no.2 63 104 155 21
Design of Experiments methodOptimized Latin Hypercube Sampling
Design of Experiments methodOptimized Latin Hypercube Sampling
Optimization
EHTC 2008 in Strasbourg
CaseCase G1G1 G2G2 G3G3 G4G4 S. D.S. D. Ave. PAve. P1 0.017692 0.025385 0.015000 0.001846 0.239995 -222.449: : : : : : :: : : : : : :
40 : : : : : :
Mesh for primitive model
MorphingMorphing
CFDCFD
4.2 4.2 CFD calculationCFD calculation of samplesof samples
Optimization
EHTC 2008 in Strasbourg
Design parameters Response
4.3 Approximation model4.3 Approximation model
Approximation model by RBF Neural NetworkApproximation model by RBF Neural Network
Optimization
CaseCase G1G1 G2G2 G3G3 G4G4 S. D.S. D. Ave. PAve. P1 0.017692 0.025385 0.015000 0.001846 0.239995 -222.449: : : : : : :: : : : : : :
40 : : : : : :
EHTC 2008 in Strasbourg
Determination coefficient R2
(Multiple correlation coefficient)
Standard deviation: 0.95633
Area-averaged pressure: 0.97671
Determination coefficient R2
(Multiple correlation coefficient)
Standard deviation: 0.95633
Area-averaged pressure: 0.97671
Desirable results with R2 > 0.95
Optimization
EHTC 2008 in Strasbourg
Primitive model (straight)
Std. Deviation of exit velocity: 0.267268
Area-averaged P of inlet: -254.155
Existing model (staged)
Std. Deviation of exit velocity: 0.
Area-averaged P of inlet : -2
Better
Better
Multi-purpose GA
Multi-purpose GA A0
B0
C0
Std.
Dev
.
Ave. P
5. Multi5. Multi--purpose optimizationpurpose optimization
Pareto solutionPareto solution
No. of Samples (1st gen.): 20No. of generations : 50
No. of Samples (1st gen.): 20No. of generations : 50
RBF Neural NetworkRBF Neural Network
Optimization
EHTC 2008 in Strasbourg
Change of design parameters on Pareto solution Blue: Pareto solutionGray: Other
G2: Can be fixed at minimum (no morphing)
G4: Can be fixed at maximum (max morphing)
Engineering Data Mining
G1: Close to maximum (max morphing)
G3: Can be utilized for design variations
Optimization
EHTC 2008 in Strasbourg
Better
Advances Pareto frontAdvances Pareto front
G1 & G4 atnear maximum
G1 & G4 atnear maximum
G1 & G4Increased amount of
morphing
G1 & G4Increased amount of
morphing
Re-optimization in approximation modelRe-optimization in
approximation model
G1 0.030 -> 0.040G4 0.006 -> 0.008
A1
B1
C1
OptimizationPrimitive model (straight)
Std. Deviation of exit velocity: 0.267268
Area-averaged P of inlet: -254.155
Existing model (staged)
Std. Deviation of exit velocity: 0.23035
Area-averaged P of inlet : -241.58
EHTC 2008 in Strasbourg
Example of Pareto SolutionS.D. (approximate) Ave. P (approximate)
A0 0.228692 -252.273B0 0.209740 -247.556C0 0.196647 -241.784A1 0.230152 -254.240B1 0.202036 -248.779C1 0.175985 -242.050
Existing model 0.230355 -241.587
Optimization
EHTC 2008 in Strasbourg
S.D.(approximate)
S.D.(CFD)
Ave. P(approximate)
Ave. P(CFD)
A0 0.228692 0.224107 -252.273 -250.307
B0 0.209740 0.204853 -247.556 -247.096
C0 0.196647 0.211522 -241.784 -241.448
A1 0.230152 0.223554 -254.240 -249.539
B1 0.202036 0.201163 -248.779 -245.823
C1 0.175985 0.195815 -242.050 -239.319
Existing model 0.230355 -241.587
Some discrepancySome discrepancy
6. Confirmation of optimized model6. Confirmation of optimized model
Optimization
EHTC 2008 in Strasbourg
Existing model (staged)Existing model (staged)Existing model (staged)
Primitive model (straight)Primitive Primitive model (straight)model (straight)
Optimized modelOptimized Optimized modelmodel
Velocity distribution for EvaporatorVelocity distribution for Evaporator
C1
Optimization
EHTC 2008 in Strasbourg
Item Application Hours CPU
Meshing (existing) SC/Tetra 3h
Developing UDF 5h
CFD Calculation (existing) SC/Tetra 0.5h 6h
Meshing (primitive) SC/Tetra 0.5h
CFD Calculation (primitive) SC/Tetra 7.5h
Determining morphing control points
Geometry modification tool 1h
Defining morphing, developing batch Geo. mod. tool 3h
Structuring work-flow Opt. Engine 1h
Design of experiments (40 cases)
Opt. EngineGeo. mod. tool
SC/Tetra1h 300h*
Optimization Opt. Engine 1h
Confirmation SC/Tetra 1h 7.5h
Total 17h 321h
Wor
kW
ork --
flow
flow
* * 7.5h X 40 cases (up to 16 jobs simultaneous7.5h X 40 cases (up to 16 jobs simultaneous for for 22.5 hour22.5 hour))
Optimization