Optimize your flow using FLOW-3D and CAESES®
Ceyhan Erdem (FRIENDSHIP SYSTEMS),
Melissa Carter (Flow Science),
Adwaith Gupta (Flow Science) and
Stefan Harries (FRIENDSHIP SYSTEMS)
FLOW-3D Users Conference, Barcelona, June 5-7, 2017
FRIENDSHIP SYSTEMS © 2017
Optimize your flow using FLOW-3D and CAESES®
Illustrated for
a microfluidic mixer and
a die-casting runner system
FLOW-3D Users Conference, Barcelona, June 5-7, 2017
FRIENDSHIP SYSTEMS © 2017
CAE System Empowering Simulation
Upfront CAD Upfront Optimization
• Simulation-ready • Design explorations
• Highly automated • Formal optimization
Variable Pre- Software Post- Optimization &
Geometry processing Connection processing Assessment
Upfront CFD
• Robust auto meshing
• Batch processing
Mesh Flow
Generation Solver
3 FRIENDSHIP SYSTEMS © 2017
Simulation-driven design (SDD)
4 FRIENDSHIP SYSTEMS © 2017
Selected references
5 FRIENDSHIP SYSTEMS © 2017
The next 22 minutes
▪ Introduction
– Focus
– Levels of application
– Work flow
▪ Microfluidics
– Level 1
– Level 2
▪ Die-casting
– (Level 1)
– Level 2
▪ Conclusions
▪ Outlook
6 FRIENDSHIP SYSTEMS © 2017
Introduction
▪ Focus
– Use of FLOW-3D in optimization by
coupling to CAESES®
▪ Applications are for illustration
– Actual simulation setups are not discussed
in any depth
– Setups follow best practice by specialists at
Flow Science
7 FRIENDSHIP SYSTEMS © 2017
Levels of application and work flow
▪ Level 1:
Easily vary flow and simulation set-up
(e.g., speeds, temperature…)
▪ Level 2:
Readily include shape variations, too
▪ Work flow for level 1
1. Run FLOW-3D once for your baseline
2. Use CAESES and FLOW-3D in a coupled
mode
3. Let CAESES change the free variables
4. Execute FLOW-3D repeatedly in batch
mode
5. Collect and analyze results within CAESES
▪ Addition for level 2
1. Build and utilize a parametric model
8 FRIENDSHIP SYSTEMS © 2017
Microfluidic mixer
9 FRIENDSHIP SYSTEMS © 2017
Microfluidic mixer
26 μm
Source: J. Homann, Wikipedia
Human hair
ranges from
50 to 100m in
diameter
31 μm
10 μm
10 FRIENDSHIP SYSTEMS © 2017
Microfluidic mixer
▪ What does the mixer do? Source: Protein folding, Kjaergaard, Wikipedia
– Initiate the folding process of proteins, i.e.,
changes of their molecular structure from
random coils to 3d-folded shapes
▪ Why do we care about 3d-folded proteins?
– Needed in a wide range of biological
functions
– Used in drug discovery, DNA sequencing
and amplification, in molecular diagnostics
and food engineering
▪ How can you produce something that is
smaller than a hair’s diameter?
– Lithography
– Wet etching
– Bonding of substrates
11 FRIENDSHIP SYSTEMS © 2017
Optimizations at different levels
▪ Level 1
– Start with a good mixer geometry
– Vary inlet speeds
▪ Level 2
– Parameterize geometry
– Vary geometry, too
Reference
Ivorra, B.; Redondo, J.L.; Santiago, J.G.; Ortigosa, P.M.;
Ramos, A.M. “Two- and three-dimensional modeling
and optimization applied to the design of a fast
hydrodynamic focusing microfluidic mixer for protein
folding,” Physics of Fluids 25, 032001 (2013)
12 FRIENDSHIP SYSTEMS © 2017
Objective and constraints
▪ Objective Source: Ivorra et al., 2013
– Time it takes for the scalar concentration
of injected proteins to drop from a 90% to
30% concentration at a representative
probe position (y = 10 μm in the c.p.)
▪ Constraints
– Constant mixer depth = 10 μm
– Xmax = 12 μm and Ymax = 30 μm
– θ Є [0,π/3]
– s = 1.5 μm, c = 2.0 μm, e = 5.0 μm
– amin = 0.5 μm and bmin = 1 μm
– lc Є [2.5 μm, 5.0 μm]
Simplification
– ls Є [1.0 μm, 9.0 μm]
Compute only one quarter utilizing symmetries
– le Є [0.1 μm, 20.0 μm]
– Re = ρνL/η ≤ 15
13 FRIENDSHIP SYSTEMS © 2017
Mixer simulation within FLOW-3D
▪ Mixer
– Center inlet channel:
Fill in a mixture of unfolded proteins and a
chemical denaturant (6 M GdCl solution)
– Two side inlet channels:
Inject a background buffer
▪ Base fluid
– Density = 1013 kg/m 3
– Dynamic viscosity = 8·10-4 Pa·s
▪ Scalar properties
– Advection: second order in fluid 1
– Molecular diffusion coef. = 2.026·10-6
14 FRIENDSHIP SYSTEMS © 2017
Mixer simulation within FLOW-3D
▪ Models applied for
– Fluid sources
– Scalars
– Viscosity and turbulence
– Automatic volume of fluid advection
▪ Mesh
– Non-conforming
– Total number of mesh blocks: 1
– Total number of real cells appr. 40 000
– Cell size = 0.1 m in x- and y-direction Clock time
(unit depth in z-direction) About 7 minutes per variant on a 4 core Intel
i7-5820K @ 3.30GHz x 12
▪ Stopping criteria
– Time = 25 s
15 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Parametric Journal and batch
geometry files controlling the
created within simulation within
CAESES FLOW-3D
Files obtained Result files
from FLOW-3D obtained from
and FlowSight FLOW-3D
for assessments and used by
optimization
strategy
16 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Input file prepin
From material properties to number of mesh
elements and density, everything can be addressed
and controlled through CAESES
17 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
18 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Output data
Data obtained from FLOW-3D are transferred to
CAESES and used for objectives and/or constraints
19 FRIENDSHIP SYSTEMS © 2017
Optimize your flow at level 1
uc
us us
Lower Baseline Upper
bound bound
Center uc m/s 0.02 0.038 0.05
Side us m/s 5.0 5.2 7.0 *
* constraint Re = ρνL/η ≤ 15 requires that Vside ≤ 7 m/s
20 FRIENDSHIP SYSTEMS © 2017
Level 1 results from a DoE
Worst level 1 Baseline Best level 1
Lower bound Worst level 1 Baseline Best level 1 Upper bound
Vcenter m/s 0.02 0.048 0.038 0.02 0.05
Vside m/s 5.0 5.125 5.2 7.0 7.0
Objective s 1.999 1.899 1.306
Change 105% – 69%
21 FRIENDSHIP SYSTEMS © 2017
Level 1 results from a DoE
Worst level 1 Baseline Best level 1
Contours obtained at time when the best design’s scalar concentration dropped to 30%
22 FRIENDSHIP SYSTEMS © 2017
Optimize your flow at level 2
▪ Build a parametric model within CAESES
▪ Change geometric parameters, too
23 FRIENDSHIP SYSTEMS © 2017
Parameters for level 2 optimization
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
24 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
25 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
26 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
27 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
28 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
29 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
30 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
31 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
32 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
33 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
34 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
35 FRIENDSHIP SYSTEMS © 2017
Parameters
• a1_min • R1
• a2_min • R2
• b_min • s_length
• c_length • side_loc
• l_c • theta
• l_e • w2_angle
• l_s
36 FRIENDSHIP SYSTEMS © 2017
Evolution over various DoE and deterministic searches
37 FRIENDSHIP SYSTEMS © 2017
Level 2 results from DoE and deterministic optimization
Initial design Baseline Best level 2
Initial design Worst level 1 Baseline Best level 1 Best level 2
Objective s 11.92 1.999 1.899 1.306 0.0935
Change 628% 105% – 69% 5%
38 FRIENDSHIP SYSTEMS © 2017
Level 2 results
Initial design Baseline Best level 2
Contours obtained at time when the best design’s scalar concentration dropped to 30%
39 FRIENDSHIP SYSTEMS © 2017
Die-casting
Cast part
From an engine maker
40 FRIENDSHIP SYSTEMS © 2017
Cast part and runner system
Ø= 0.075 m
0.224 m
0.098 m
0.121 m 0.112 m 0.150 m
0.051 m 0.020 m
0.011 m
41 FRIENDSHIP SYSTEMS © 2017
Objectives and constraints
▪ Objectives
– Overall entrained air volume and
– Free surface defect concentration
(i.e., impurities arising from oxide fast shot
formation on the free surface) phase
▪ Constraints
– Velocity in gates need to be between 20
and 60 m/s
slow shot
– Gates should be fed from runner system at
phase
the same time
– All defects should go to overflows
– Fast shot starts when metal hits the gates
– Flow should be from thin to thick sections
and on shortest route
42 FRIENDSHIP SYSTEMS © 2017
Adjustments within CAESES
fast shot
phase
calculator within
CAESES to
determine the
fast shot, slow
shot speeds for
slow shot
each parametric
phase
model
43 FRIENDSHIP SYSTEMS © 2017
Die-casting within FLOW-3D Cast
▪ Plunger
– CuCoBe (Copper die)
– Thermal conductivity = 300 W/(m·K)
– Density · specific heat = 3.52 · 106 J/(m3·K)
▪ Remaining space mold
– Steel H13
– Thermal conductivity = 28.6 W/(m·K)
– Density · specific heat = 3.56 · 104 J/(m3·K)
– Max. thermal penetration depth = 0.014
44 FRIENDSHIP SYSTEMS © 2017
Die-casting within FLOW-3D Cast
Dim. Liquid Solid
Cast material Aluminum Alloy A380
(LM24, ADC10/AC4B)
Density kg/m3 2460 2710
Dyn. viscosity Pa·s 0.0012 -
Specific heat J/(kg·K) 1245.3 963
Thermal conductivity W/(m·K) 104 96.2
Initial temp. °C 649
Temperature °C 574.4 497.3
Latent heat of fusion J/kg 3.89 · 105
45 FRIENDSHIP SYSTEMS © 2017
Die-casting within FLOW-3D Cast
46 FRIENDSHIP SYSTEMS © 2017
Die-casting within FLOW-3D Cast
▪ Mesh
– Mesh type: non-conforming
– Total number of mesh blocks: 2
– Total number of real cells 1 400 000
– Cell size = 2 mm
47 FRIENDSHIP SYSTEMS © 2017
Die-casting within FLOW-3D Cast
▪ Simulation properties
– One fluid with free surface model
▪ Models applied
– Air entrainment
– Cavitation
– Defect tracking
– Density evaluation
– Gravity and non-inertial reference frame
– Heat transfer
– Moving and simple deforming objects Clock time
About 2 hours per variant on a 8 core Intel
– Solidification i7-5820K @ 3.30GHz x 12
– Viscosity and turbulence
▪ Stopping criterion
– Filling fraction reached 0.999
48 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Parametric Journal and batch
geometry files controlling the
created within simulation within
CAESES FLOW-3D
Files obtained Result files
from FLOW-3D obtained from
and FlowSight FLOW-3D
for assessments and used by
optimization
strategy
49 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Input file prepin
From material properties to number of mesh
elements and density, everything can be addressed
and controlled through CAESES
50 FRIENDSHIP SYSTEMS © 2017
Connection between FLOW-3D and CAESES
Output data
Data obtained from FLOW-3D are transferred to
CAESES and used for objectives and/or constraints
51 FRIENDSHIP SYSTEMS © 2017
Optimize your flow at level 1
Lower Baseline Upper
bound bound fast shot
phase
Speed during m/s 1.5 1.6 2.5
fast shot
Casting fluid °C 620 649 680
temperature
slow shot
phase
52 FRIENDSHIP SYSTEMS © 2017
Parameters for level 2 optimization
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
53 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
54 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
55 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
56 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
57 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
58 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
59 FRIENDSHIP SYSTEMS © 2017
Parameters
• gate_loc_on_mold
• R_tonque
• tonque_tangency
• gate_length
• gate_angle
• left/right runner end parameter
• mid runner end parameter
• runner_conn
60 FRIENDSHIP SYSTEMS © 2017
Evolution over various DoE and deterministic searches
61 FRIENDSHIP SYSTEMS © 2017
Level 2 results from DoE and deterministic optimization
Worst design Best level 2
62 FRIENDSHIP SYSTEMS © 2017
Level 2 results wrt entrained air volume
Worst design Baseline Best level 2
Worst design * Baseline Best level 1 * Best level 2 *
Entrained air volume · 10-5 9.927 (109,8%) 9.045 8.896 (98,4%) 7.718 (85.3%)
Free surface defects · 10-2 2.977 (100,7%) 2.956 2.956 (100%) 2.933 (99.2%)
* evaluation criteria wrt entrained air volume 63 FRIENDSHIP SYSTEMS © 2017
Level 2 results wrt entrained air volume
Worst design Baseline Best level 2
Worst design * Baseline Best level 1 * Best level 2 *
Entrained air volume · 10-5 9.927 (109,8%) 9.045 8.896 (98,4%) 7.718 (85.3%)
Free surface defects · 10-2 2.977 (100,7%) 2.956 2.956 (100%) 2.933 (99.2%)
* evaluation criteria wrt entrained air volume 64 FRIENDSHIP SYSTEMS © 2017
Level 2 results wrt free surface defect concentration
Worst design Baseline Best level 2
Worst design * Baseline Best level 1 * Best level 2 *
Entrained air volume · 10-5 9.927 (109,8%) 9.045 8.896 (98,4%) 7.718 (85.3%)
Free surface defects · 10-2 2.977 (100,7%) 2.956 2.956 (100%) 2.933 (99.2%)
* evaluation criteria wrt entrained air volume 65 FRIENDSHIP SYSTEMS © 2017
Level 2 results wrt free surface defect concentration
Worst design Baseline Best level 2
Worst design * Baseline Best level 1 * Best level 2 *
Entrained air volume · 10-5 9.927 (109,8%) 9.045 8.896 (98,4%) 7.718 (85.3%)
Free surface defects · 10-2 2.977 (100,7%) 2.956 2.956 (100%) 2.933 (99.2%)
* evaluation criteria wrt entrained air volume 66 FRIENDSHIP SYSTEMS © 2017
Conclusions
▪ Optimize at different levels of complexity
– Level 1:
Easily vary flow and simulation set-up
(e.g. ,speeds, temperature…)
– Level 2:
Readily include shape variations, too
▪ Two examples given for illustration
▪ Benefit from accurate CFD (FLOW-3D and
FLOW-3D Cast) by coupling to robust variable
CAD and optimization (CAESES®)
▪ Benefit from tight cooperation between
Flow Science and FRIENDSHIP SYSTEMS
67 FRIENDSHIP SYSTEMS © 2017
“To boldly go where no man has gone before”
68 FRIENDSHIP SYSTEMS © 2017
www.friendship-systems.com
Dr.-Ing. Stefan Harries MSE
[email protected] FRIENDSHIP SYSTEMS © 2017
Outlook
70 FRIENDSHIP SYSTEMS © 2017
Outlook
71 FRIENDSHIP SYSTEMS © 2017
Simulation-driven design from m to km
Source: J. Homann, Wikipedia Source: DSME
Human hair
ranges from
50 to 100m in
diameter
SDD over 7 orders of magnitude
From microfluidics with 10 m of beam to
maritime systems with 400m in length
72 FRIENDSHIP SYSTEMS © 2017
