t

Automatic Simulation-
Simulation-Driven Medical

ec
oj
Device Optimization Design and

Pr
Advanced Multiphase Modeling

r ia
Jeongho Kim
ng
Sa

James F. Antaki, PhD

July 2, 2010

CENTER FOR ADVANCED RESEARCH DESIGN INNOVATION AND OPTIMIZATION OF MEDICAL DEVICES
 Outline
Out e

t
ec
• Background

oj
• Hypothesis

Pr
• C
Case St
Study
d 11: PVAD

ia
• Case Study 2: Magnetic Cell Separator
r
ng
• Case Study 3: Mixture Theory
Sa

2
 Manual Design Evaluation Process

t
ec
=

oj
Pr
r ia
ng
Manual Design Process Results
Sa
• Time consuming • Produces limited number of designs
• Error prone tasks • Produces questionable design quality
• Engineers spend more time preparing

3
 Automated Design Optimization

Optimizer

t
ec
=

oj
Pr
r ia
ng
Sa
Optimizer as a Software Robot Results
• Automates and iterates design process • Increases evaluations
• Engineer defines simulation process • Improves quality
• Engineer defines goals and constraints • Engineers spend more time analyzing
• Saves valuable engineering time

4
 t
ec
oj
Case Study 1:

Pr
PediaFlow Ventricular Assist Device

ia
r
ng
Sa

5
Motivation
ot at o

t
ec
 Birth to 2 yrs (3 to 15 kg),

oj
 A mixed-type turbodynamic pump

Pr
 Magnetically levitated rotor

r ia
ng
Sa

6
 Magnetically Levitated Miniature Mixed Flow Blood Pump

PERMANENT
MAGNET

t
BEARINGS

ec
ROTOR

oj
SENSOR

Pr
DRIVE COILS
LINE

r ia
FLOW
STRAIGHTENER

MOTOR
ng IMPELLER
Sa
FIELD COILS BLADES
AXIAL
SUSPENSION MAGNETIC
COILS (CLEARANCE)  PF3: World’s smallest maglev VAD
GAP

7
Blood Flow Path
IMPELLER STATOR

t
ec
oj
Pr
r ia
ng
Sa

SECOND MIDDLE STATOR

8
 Pump
p Design
g Parameters
BLADE CAMBER BLADE
- incidence HEIGHT
- deflection PROFILE

t
HUB PROFILE - shape TAIL PROFILE

ec
NUMBER OF

oj
BLADES

Pr
ROTOR

ia
SPEED BLADE PROFILE
- section

r
ng - taper
- sweep
- height
Sa
ROTOR STATOR - twist
LENGTH LENGTH

JFA 9.13.94

9
“Optimization”
p by
y Trial and Error

t
ec
oj
Pr
ia
r
ng
Sa

10
 CFD
C Analysis
a ys s

• ANSYS CFX 12 12.0

0
• Steady-state incompressible flow

t
ec
• SST turbulent model

oj
• Newtonian fluid

Pr
r ia
ng
Sa

11
 CFD
C Analysis
a ys s
Rotating
g machine design
g using
g ANSYS Turbosystem
y

t
ec
oj
Pr
ia
r
ng
Sa

CFX-Pre: pre-processor
CFX-Solve: solver
CFX-Post: post-processor

12
Elements of (Shape) Optimization

INITIAL
GEOMETRY

Shape Parameters

t
ec
oj
CFD
Simulation

Pr
Flow Field:

ia
- velocity
- pressure

r
- shear stress
ng
Design Objective Function Search Procedure:
Sa
Objectives & Gradient Evaluation Update Parameters

Acceptance
Criteria reject
OPTIMAL
GEOMETRY

14
Parametric Models of
T b
Turbomachinery
hi Blades
Bl d

t
ec
oj
Pr
ia
r
ng
Sa

Middle Stator Blades

15
 Automated
uto ated Optimization
Opt at o
• Optimizer: Isight (SIMULIA)
• Algorithm: NLPQL
• Objective: Maximize Efficiency

t
ec
• Constraint: Static Head >= 80 mmHg

oj
Pr
• Optimization automatically changes design variables to find the
"best" design satisfying specified criteria.

r ia
ng
Sa

16
Results:
esu ts Hexahedral
e a ed a Mesh
es

t
ec
oj
Pr
ia
Control total number of mesh
Maintain good mesh quality

r
ng
Sa

17
Results:
esu ts Shape
S ape Optimization
Opt at o

t
Initial Blade

ec
oj
Pr
ia
r
ng
Sa

Optimal
p Blade

18
Results:
esu ts St
Streamlines
ea es

Initial Blade

t
ec
oj
Pr
Optimal Blade

r ia
ng
Sa

Initial Design Optimal Design

P
Pressure Ri
Rise 108 157
Efficiency 12.9 % 18.7 %

45.3% increase in Pressure

19
 FDA Nozzle
o e

t
ec
oj
Pr
r ia
ng
Sa

Pure hexahedral meshes

Working with Sam

20
 Time
e Table
ab e

t
2nd 3rd 4th 1st 2nd 3rd 4th

ec
Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,,
2010 2010 2010 2011 2011 2011 2011

oj
Blood Damage

Pr
(FDA Nozzle and PVAD)

ia
Gradient-Based Opt.

r
Response Surface Opt. ng
Sa
Multi-Point Opt.

Multi Objective Opt.

Multi-Objective Opt

21
 Accomplishments
cco p s e ts
Journal papers
• Ki J.,
Kim, J Antaki,
A t ki J.,
J Simulation-Based
Si l ti B dAAutomatic
t ti O
Optimization
ti i ti off th
the PediaFlowTM
P di Fl TM VAD,
VAD ini
preparation.

t
• JF Antaki1, MR Ricci, JE Verkaik, ST Snyder, TM Maul, J Kim, D Paden, BE Paden, HS

ec
Borovetz, PediaFlow™ Maglev Ventricular Assist Device: A Prescriptive Design Approach,
Cardiovascular Engineering Technology 1(1), 2010

oj
Pr
Conference abstracts and proceedings
• Kim J
Kim, J., Antaki,
Antaki J.,
J Simulation-Based
Simulation Based Automatic Optimization of the PediaFlowTM VAD. VAD
NIH-FDA-NSF Workshop 2010.

ia
• Kim, J., Antaki, J., Simulation-based Design and Optimization of the PediaFlow VAD. NIH-

r
FDA-NSF Workshop 2009. ng
• Shu, F., Verkaik, J., Snyder, S., Paden, D., Kim, J., Antaki, J., Ventricular Assist Device for
Toddlers with Hybrid Magnetic-Mechanical Bearings. ASAIO Abstracts, 2009. 55(2):p.147.
Sa

• Kim, J., Hund, S., Daly, A., Kameneva, M, Antaki, J., Eulerian Method for Numerical
y in PediaFlow VAD. 5th IFAO Proceedings,
Prediction of Hemolysis g , 2009. 5:p.59.
p
• Kim, J., Antaki, J., Computational Fluid Dynamic Shape Optimization of the PediaFlowTM
VAD. ASAIO Abstracts, 2009. 55(2):p.155.

P
Pumpkin
ki PVAD quarterly
t l reports
t (April
(A il 2010,
2010 July
J l 2010)

22
 t
ec
oj
Case Study 2:

Pr
Magnetic Cell Separator

ia
r
ng
Sa

23
 mPhoresisTM
M
Magnetic
ti CCell
ll S
Separator
t
• A dialysis
dialysis-like
like device that separate malaria infected
RBCs from the blood using magnetic field.

t
ec
Healthy Red Blood Cell

oj
Pr
r ia
ng
Sa

Infected Red Cell

• The malaria parasite converts hemoglobin into hemozoin.
hemozoin
• Hemozoin becomes paramagnetic.
24
 Methods:
B
Boundary
d V
Value
l Problem
P bl
MAGNETIC POLE
GRID PLATE
BLOOD FROM INFECTED PURIFIED
PATIENT PERMANENT CELLS BLOOD

t
ec
MAGNET

oj
Pr
ria
ng
Sa

COMPUTATIONAL DOMAIN

Infected cells are drawn towards the pole plate

by the magnetic field
field, and are skimmed away
away.

25
 Methods:
et ods Automated
uto ated Design
es g

• Design Variable: Height

Height, Flow Rate,
Rate Wire Pitch
• Objective: Minimize Length

t
ec
• Constraint: 99% Beads Captured
p

oj
• Optimizer: Isight (SIMULIA)

Pr
• Algorithm: NLPQL

r ia
ng
Sa

26
 Results:
esu ts Bead
ead Trajectories
ajecto es

t
ec
1 cc/min

oj
Pr
ia
Magnetic Particles (Spherotech, Inc.)

r
- Diameter of 0.3 micron ng
- Mass
M off 1
1.8378e-017
8378 017 kkg
- Susceptibility of 0.26
5 cc/min
Sa

27
 Results: Automatic Optimization
p

• NLPQL provided a local optimal point, depending on

t
a starting point.

ec
• Need different initial points to find a best optimal

oj
point.

Pr
ia
Minimize Length Side Bounds Initial Optimal Initial Optimal

r
flow rate, cc/min ng
1 to 10 5 1 10 9
height, micron 50 to 300 200 50 300 50
Sa
pitch, micron 100 to 400 200 100 400 100
length, mm 65.4 0.7 927.6 6.9
Iteration 29 19

Needed to capture 99% beads

28
 Time
e Table
ab e

t
2nd 3rd 4th 1st 2nd 3rd 4th

ec
Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,,
2010 2010 2010 2011 2011 2011 2011

oj
Magnetic Particle

Pr
Dynamics

ia
Lagrangian-Eulerian

r
Mixture Theory ng
Sa
Design Optimization

29
 Acomplishments
co p s e ts
Journal papers
• Kim, J., et al., Magnetic particle Dynamics, in

t
ec
manuscript.
manuscript

oj
Pr
Conference abstracts and proceedings

ria
• Kim, J., Gandini, A., Antaki, J., Numerical
ng
Study of Magnetic Field Separator to Remove
Sa

Malaria-Infected Red Blood Cells from the

Whole Blood. NIH-FDA-NSF Workshop 2010.

30
 t
ec
oj
Case Study 3:

Pr
Developing an Advanced Multiphase Model:

ia
r
Mixture Theory
ng y
Sa

31
 Blood
ood Properties
ope t es

• Plasma
– occupies 55-60% of total blood volume

t
ec
• RBCs RBC

oj
DEPLETION
– 40-45% of blood volume ZONE

Pr
– 8-10
8 10 μm diameter

ia
– biconcave discoids

r
ng
– aggregation and deformability
Sa
– RBC tumbling
• White
Whit bl
blood
d cells
ll andd Pl
Platelets
t l t
– only contain 5% by volume Zhao et al.

– important
i t t roles
l in
i immunity
i it and
d hematostasis
h t t i
32
 Microscopic Characteristics in
Bl
Blood
d Fl
Flow
• Microhemorheology
– Shear thinning, Fahraeus effect, Fahraeus-Lindqvist

t
ec
effect, Plasma skimming, g Platelet margination,
g etc.
– Device-related micro-hemorheological study is

oj
lacking.

Pr
• Single Phase Model

ia
– cannot predict the concentration profile of blood cells

r
ng
(phase separation)
Sa
– is invalid at micro scale.
• There is no reliable hemorheological model that can
predict hemodynamics for blood flow in blood contacting
devices.

33
 Multiphase Modeling:
Mi t
Mixture Theory
Th
• the Theory of Interacting Continua
• based on the ideas of diffusion proposed by Fick

t
ec
• a homogenization approach

oj
– each component is regarded as a single continuum

Pr
and at each instant of time, every point in space is
considered to be occupied by a particle belonging to

ia
each component of the mixture.

r
ng
• The foundation of the theory is given in books by
Truesdell (1984); Dobran (1991) and Rajagopal and Tao
Sa

(1995).
• Johnson et al (1991) and Massoudi et al. (1999) have
formulated a two-phase flow theory based on this mixture
theory
theory.

34
 Benefits
e e ts oof tthe
e Mixture
tu e Theory
eo y

• To predict the phase separation of

plasma-RBC mixture

t
ec
– velocity field of RBCs and plasma

oj
Pr
– co
concentration
ce a o field edo
of RBCs
Cs aand
dpplasma
as a

ia
• Can be applied to physiological
r
ng
Hematocrit (40
(40~50%)
50%)
Sa

• The results will be used as the input

p to
numerical platelet deposition model
or/and blood damage model
model.
35
 A brief review of Mixture Theory
y

• X1 and X2 denote the positions of

particles of S1 and S2

t
ec
x1  χ1 ( X1 , t),
t) and x2 = χ 2 ( X2 , t).

oj
• The kinematical quantities

Pr
associated with these motions are

ia
d1 χ1 d2 χ 2 d1 v1

r
v1 = ; v2 = , a1 =
ng
dt dt dt
Massoudi and Rao, 2001
Sa
d2 v2  v1  v2
a2 = , L1 = , L2 = ,
dt  x1  x2

1 1 1 1
D1 = (L1 + L1T ), D2 = (L 2 + L T2 ), W1  (L1  LT 1 ), W2  (L 2  LT 2 )
2 2 2 2

36
 A brief review of Mixture Theory,
cont’
t’

• ρ1 and ρ2 are the bulk densities of the

mixture components.

t
ec
ρ1 =  ρ10 , ρ2 =  ρ20,

oj
Pr
– ρ10 is the pure density of the component

ia
1, ρ2 is the pure density of component 2.

r
ng
– ν is the volume fraction of the component
Sa
1, and φ is the volume fraction of
component 2 2.
– For a saturated mixture φ= 1 - ν.

37
 A brief review of Mixture Theory,
cont’
t’
• The mixture density
density, ρm and the mean velocity vm of
the mixture are defined

t
ec
 m =  1 +  2, ρm v m = ρ1 v1 + ρ2 v 2.

oj
• The individual stress tensors

Pr
T1  (1   )Tf T2  Ts

ia
• A mixture stress tensor

r
ng
Tm  T1  T2
Sa

– The mixture stress tensor reduces to that of a

pure fluid as φ→0 and to that of the solid phase
as ν →0

38
A brief review of Mixture Theory,
cont’
t’

Conservation of Conservation of
Mass Linear Momentum

t
ec
Phase 1:
Ph 1 1 D1v1 = div +
+ div (  1v1) = 0 1 T1 1b1 + fI

oj
Plasma t Dt

Pr
Phase 2:  2 D 2 v 2 = div
+ div (  2 v 2) = 0 2 T 2 +  2b 2 - f I

ia
RBC t Dt

r
ng
Constitutive Equations
Sa

Stress Tensor Stress Tensor Interaction force:

for plasma: T1 for RBC: T2 fI

39
 Constitutive Equations
q for Plasma

• Plasma can be assumed to be a linearly

viscous fluid.

t
ec
oj
T1 = [ - p(1 -  ) + λ (1 -  )tr D1 ] I + 2 (1 -  )D1

Pr
ia
• p is the fluid pressure, μ is the viscosity,
r
ng
D1 is the symmetric part of the velocity
Sa

gradient,, and λ is the second coefficient

g
of viscosity.
D1 
1
2
v1  (v1 ) T 
40
 Constitutive Equation
f RBC Ph
for Phase

• The RBC phase is represented as an

anisotropic non-linear density-gradient-

t
ec
type fluid.

oj
T2  f ( ,  , D2 , D2 )
2

Pr
ia
T2  β o  β1    β 2 trD 2 I  β 3 D 2  β 4     β 5 D 2
2

r
ng
Sa

• ρ2= φρRBC, and the β

β’s
s are material
properties.
D2 
1
2
v 2  (v 2 ) T 
41
 Constitutive Equation
f RBC Ph
for Phase, cont’
t’

• β0 is similar to pressure

t
• β2 corresponds to the second coefficient

ec
of viscosity

oj
Pr
• β1 and β4 are the material parameters

ia
connected with the distribution of the
r
ng
RBCs
RBC
Sa

• β3 is the viscosityy
• β5 is similar to the cross-viscosity of a
Reiner Rivlin fluid
Reiner-Rivlin fluid.
42
 Constitutive Equation
f RBC Ph
for Phase, cont’
t’

• β0 is similar to pressure

t
• β2 corresponds to the second coefficient

ec
of viscosity

oj
Pr
• β1 and β4 are the material parameters

ia
connected with the distribution of the
r
ng
RBCs
RBC
Sa

• β3 is the viscosityy
• β5 is similar to the cross-viscosity of a
Reiner Rivlin fluid
Reiner-Rivlin fluid.
43
 Constitutive Equation
f RBC Ph
for Phase, cont’
t’
• The viscous effects (β3) is assumed to be
predominate over the effects of the gradient

t
ec
of RBC volume fraction,
fraction the second

oj
coefficient of viscosity and the normal

Pr
stresses.
stresses

ia
• β1, β2, β4, and β5 are negligible.
r
ng
• The stress tensor for the RBCs reduces to the
Sa

structure.
T2  0I  β3D2

44
 Constitutive Equation
f RBC Ph
for Phase, cont’
t’
• β0 and β3 are given by Massoudi and Antaki Antaki,
(2008) 10000

t
ec
1000
  p  (  )   (    2
)

oj
0 3
sity (cP) 30

• Shear-thinning effects were incorporated by

Pr
100
adopting a shear-dependent viscosity for the
viscos

ia
RBC phase,10
proposed by Yeleswarapu et al.
r
ng
1
Sa
1  ln(1  Shear
 ) Rate (1/s)
1.00E-06
1.00E-04 1.00E-02 1.00E+00 1.00E+02 1.00E+04

 3 ( )  (    (  0    )
1  Model
Present 
)(   2 )   [2tr (D 22 )]1/ 2
Experiment (Chien et al. Science 1967)

45
 Constitutive Equation
ffor IInteraction
nteraction
t ti Forces
F

• The mechanical interaction force can be

assumed to be of the form (Johnson

t
ec
1990). DIFFUSION

oj
DRAG

Pr
FI = A1 grad φ + A 2 F (φ)( v 2 - v 1 )

r ia
+ A 3 φ(2 tr D12 )-1/4 D1 ( v 2 - v 1 )
ng
 A4 φ(W2  W1 )( v 2 - v 1 )  A5 a vm
Sa

SHEAR LIFT SPIN LIFT VIRTUAL MASS

9 f 3(6.46)  f  f
1/ 2 1/ 2
3 2 3 1  2
A2  A3  A4   f A5  a
2 a2 4 a 4 3 1
46
 Constitutive Equation
f IInteraction
for nteraction
t ti Forces,
F cont’
t’
• No investigations as to what form A1
may have

t
ec
• The remaining coefficients have not

oj
Pr
been extensively studied for general

ia
two-component flows.
r
– Need e
ng
experiments
periments for 1
1-D
D Mi
Mixture
t re Theor
Theory
Sa

• The forms given above are ad-hoc

applications of results that are strictly
valid under more restricted conditions.
47
 Ope
OpenFOAM
O

• FVM-based
FVM based open source code

t
• C++

ec
oj
• Linux environment

Pr
• N bl
No blackkbbox

ia
• p g a new flow
Suitable for developing
r
ng
model
Sa

48
 Time
e Table
ab e

t
2nd 3rd 4th 1st 2nd 3rd 4th

ec
Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,, Quarter,,
2010 2010 2010 2011 2011 2011 2011

oj
1D Mixture Model

Pr
(Five Sub-Studies)
2D/3D Mixture Model

ia
(Expansion Channel,
Exotic Channel)

r
Design Optimization
ng
Sa

49
 Accomplishments
cco p s e ts
Journal papers
• Massoudi, M., Kim, J., Hund, S., Antaki, J., A

t
ec
Mixture Theory formulation for Blood Flow
Flow, in

oj
manuscript

Pr
ia
Conference abstracts and proceedings
r
ng
• Massoudi, M., Kim, J., Hund, S., Antaki, J., A
Sa

Mixture Theoryy formulation for Blood Flow.

The 47th Annual Technical Meeting of
Societyy of Engineering
g g Science, 2010.

50
 Acknowledgements
c o edge e ts

• Dr. Antaki

t
• Dr. Massoudi

ec
oj
• Dr. Kameneva

Pr
• Dr Gandini
Dr.

ia
• Lab mates
r
ng
Sa
Thanks for your time!

Questions?
Q
51