low reynolds number fluid-structure interactions: modeling...
TRANSCRIPT
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Low Reynolds Number Fluid-Structure Interactions:Modeling the Response of Soft Microfluidic Devices
Ivan C. Christov & Tanmay C. Shidhorehttp://tmnt-lab.org/
School of Mechanical EngineeringPurdue University
Minisymposium on Analysis, Control and Inverse Theory of Flows, Material Structures,Acoustics, and Their Interactions
SIAM Conference on Analysis of Partial Differential Equations
Baltimore, Maryland
December 10, 2017
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 1 / 14
PURDUE UNIVERSITY COLLEGE OF ENGINEERING BRAND MANUAL page 31
THE SIGNATUREAlways use the correct artwork for the schools of Engineering signatures. The left side is always the university name as independent usage or the name of the division level for lockup usage with Frutiger Black font. The right side can be changed to the name of the College or divisions with Frutiger Light font.
VISUAL IDENTITY
LOGO USAGELOGO WITHTHE SCHOOL NAME
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Lab on a chip
“The ability to perform laboratory operations on small scales
using miniaturized (lab-on-a-chip) devices has many benefits.
Designing and fabricating such systems is extremely challenging,
but physicists and engineers are beginning to construct highly
integrated and compact labs on chips with exciting functionality
...” – Nature 442 (2006)
But there’s much to learn ...
– Wall Street Journal (Oct. 26, 2015)
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 2 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Fabrication and response of microfluidic devicesTypical lab-on-a-chip device fabrication via replica molding of PDMS.(Sollier, et al., Lab Chip, 2011)
cured polymer is still sticky having some difficulties to crosslink
(E ¼ 1.1 MPa). These data confirm that changing the cross-
linking ratio for PDMS is not a reliable solution towards more
rigid polymer prototyping materials.
As discussed by Inglis,27 recipes for harder PDMS have been
recently developed as alternatives to classical Sylgard 184
PDMS. For example, the h-PDMS by IBM Zurich has been
reported to be 4.5 times stiffer than standard PDMS40,41 but
Inglis reported difficulty in release of these polymers from the
molds.27 Here we tested another PDMS, also from Dow Corning
with an expected higher stiffness (biomedical grade PDMS,
Silastic 7-4860). Unfortunately, this Silastic PDMS is very
viscous, which greatly complicates its processing, and above all
still was measured to have a tensile elastic modulus similar to
1 : 10 PDMS (2.4 versus 2.5 MPa) (Fig. 3.A).
The other prototyping polymers investigated are much stiffer
than PDMS. The Young’s modulus measured for TPE is
"1.2 GPa, similar to thermoplastics like PC (2 GPa) and three
orders of magnitude higher than 1 : 10 PDMS. This is important
since design performance for TPE chips should be replicated
when later transferring to mass producible thermoplastics,
reducing the need to re-optimize designs late in the commer-
cialization process. When cured as indicated in our process flow,
NOA81 exhibits a tensile modulus of 325 MPa, which is lower
than the 1 GPa reported by Bartolo et al.34 but still two orders of
magnitude higher than PDMS. As explained previously, such
a difference is likely due to the reduced curing time that was
required for successful device assembly and sealing. Finally, the
tensile modulus measured for PUMA is 91 MPa, which is lower
than TPE and NOA81. Notably, PUMA was observed to have
different properties in its stress-strain curve than TPE and
NOA81, with a short elastic behavior, for extension only from
0 to 2 mm, followed by a large plastic behavior, where PUMA
pieces undergo irreversible extension up to 60 mm.
Deformation under flow
The low tensile modulus of PDMS offers several advantages for
microfluidic applications as (i) it helps in demolding without
Fig. 2 Protocols for fabrication of PDMS, PUMA, TPE and NOA chips. Our primary test device was a straight and rectangular channel (5 cm long,
60 mm wide, 52 mm deep). A restriction (1000 mm long, 30 mm large) was added 4 cm downstream the inlet in order to validate the replication fidelity for
more complex structures. (i) PDMSmicrofluidic devices were fabricated using standard replica molding processes,26 with a cross linker to polymer ratio
of 1 : 10, and maintaining 2 h as the curing time for repeatable mechanical measurements. As PUMA, NOA and TPE adhere to SU8 and make
demolding impossible, a PDMS master mold with the same polarity as the silicon master was produced as indicated by Kuo et al.,31 using octade-
cyltricholosilane (Sigma-Aldrich, 0.2% v/v in pure ethanol) to passivate the surface. (ii) The PUMA chips were prepared as previously described by
Kuo et al.,32 especially using 1 min of UV exposure for PUMA channels, 5 min for the PUMA-coated glass slide, and additional 5 mins to complete the
bonding after the conformal sealing. (iii) TPE devices were fabricated following the protocol published by Fiorini et al.,28–30 with UV exposure for 70 s in
our conditions (l ¼ 364 nm, 400 W, Dymax Model 2000 Flood). (iv) NOA chips were made using a protocol adapted from Bartolo et al.34 and W€agliet al.,35 especially using 2 s of UV exposure for NOA channels, 3 s for the NOA-coated glass slide, and an additional 3 s to complete the bonding after the
conformal sealing.
3756 | Lab Chip, 2011, 11, 3752–3765 This journal is ª The Royal Society of Chemistry 2011
Dow
nloa
ded
by P
rince
ton
Uni
vers
ity o
n 08
Nov
embe
r 201
2Pu
blish
ed o
n 07
Oct
ober
201
1 on
http
://pu
bs.rs
c.or
g | d
oi:1
0.10
39/C
1LC2
0514
E
View Online
Resulting microchannels are soft. (Gervais, et al., Lab Chip, 2006; Ozsun, et al., JFM, 2013)
images were acquired, filtered and rendered using Imaris 4.2
image analysis software (Bitplane Inc.). Cross-section contours
were obtained using built-in algorithms from the Matlab image
processing toolbox (Mathworks Inc., Natick, MA).
Results and discussion
Numerical simulations
In order to understand the nature of coupled fluid–structure
interactions, simulations focus on predicting the extent of the
channel deformation as well as the resulting pressure distribu-
tion inside the device. The models also reveal the extent of the
flow perturbation due to the channel deformation (Fig. 3).
These results (and eqn (12) describing the average fluid
velocity U(z) at a position z down the channel) show that the
average fluid velocity increases towards the channel outlet as
the channel cross-section decreases, while the volumetric flow
rate remains constant. This acceleration of the flow in a
deforming channel is a potential source of error in predicting
quantities such as the shear stress in a PDMS device for cell
attachment assays.
Confocal imaging and displacement measurements
Channel deformation was verified directly using confocal
microscopy. 3D rendered images of the channels capture the
shape of the channels under various flow conditions (Fig. 4).
The experimental data reveal the much higher deformation
predicted by the geometry change at the inlets of the device
(Fig. 5 and 6, first 3 data points). The rapid tapering of the
channel (Fig. 2) effectively produces an unconstrained
displacement (stress-free) boundary condition at the edges.
Both simulation and experimental data show that this effect
dies away within a few widths inside the channel. This entrance
effect is not taken into account in the simplified model
provided in eqn (10). However, since in most applications,
channels are much longer than wide, neglecting it is usually a
reasonable approximation.
By using the raw data, measurements of the maximum
displacement (located at mid channel width) were performed
and compared to the numerical simulations and scaling model
(eqn (10)). Two types of measurements were performed in both
250 mm and 500 mm wide channels: (i) A flow rate was specified
and the maximum displacement measured for various points
along the channel axis (Fig. 5 and 6). (ii) An axial position was
Fig. 2 Schematics of the microfluidic channel used (channel to scale).
Deflection is measured in the channel’s narrower region (250 mm or
500 mm). A series of triangular ticks (20 mm wide) are patterned at the
channel wall every 500 mm for accurate positioning of confocal images.
Fig. 3 3D simulation of the velocity (color) and pressure (grayscale)
profiles under an imposed pressure drop of 1 bar. The channel
geometry represented is 25 mm 6 500 mm 6 1 cm (Re = 15). Young’s
modulus is assumed to be of E = 1 MPa. The decreasing cross-sectional
area along the channel axis is responsible for the change in the fully
developed flow profile and for the acceleration of the flow.
Fig. 4 3D rendering of the entrance of two microfluidic channels of
the same material and geometry (250 mm wide, 26 mm thick, 1 cm long,
E = 2.2 MPa). Top channel: 300 mL min21 imposed flow rate (Re =
13). The tapering of the channel can be observed by comparing the
cross-sections in the first and second segments. Long dash lines:
topographic displacement curves. Bottom channel: 1 mL min21
imposed flow rate (no displacement reference measurement).
Fig. 5 Maximum displacement vs. axial position (E = 2.2 MPa). m:
26 mm 6 250 mm channel at a flow rate of 300 mL min21. &: 30 mm 6500 mm channel at a flow rate of 800 mL min21. 6: Numerical
simulation under the same material and flow conditions.
504 | Lab Chip, 2006, 6, 500–507 This journal is ! The Royal Society of Chemistry 2006
Dow
nloa
ded
by P
rince
ton
Uni
vers
ity o
n 05
June
201
2Pu
blish
ed o
n 03
Mar
ch 2
006
on h
ttp://
pubs
.rsc.
org
| doi
:10.
1039
/B51
3524
A
View Online
Non-invasive pressure measurements
(a) (b)
xz
y
05
1015
–0.5
00.5
z (m
m)
x (mm)0 5 10 15
–20
0
20
40
3.6 kPa
100 Pa
1.6 kPa
50 kPa
20 kPa
10 kPa
Chip
Bottom wall
ShimChannelzy
(d)
(e) (f)
(c)
xy
100
101
102
102 103 104 105 103 104 105
100
101
102
FIGURE 1. (a) A one-dimensional channel with a deformable top wall. (b) The two-dimensionaldeflection ⇣2d(x, z) of S1 (t = 200 nm, 2h0 = 175 µm) under hydrostatic pressure of p = 2.3 kPaas measured optically. The inset shows a cross-sectional view of the channel. Clamps hold thechip (with the thin membrane at the centre) and the bottom wall together. The two walls areseparated by a precision shim; an O-ring (black ovals) seals the channel. (c) Cross-sectionalline scans from the image in (b) showing the parabolic z-profile of the deformable wall underpressure. (d) The one-dimensional (average) wall deflection ⇣(x) at several different pressuresfor the same sample. The indicated values are gauge pressure values. These are the localconstitutive curves. (e) The peak deflection ⇣p of the PDMS deformable walls as a function ofpressure p. (f ) ⇣p of the SiN deformable walls as a function of p; the dashed lines are the p1/3
asymptotes. Note that the ⇣p and the p axes do not cover the same ranges in (e) and (f ). Errorbars are smaller than the symbols.
pressure distribution in a planar channel flow and validate our measurements againstthe analytic approximation in (1.2) for a long channel. Our method does not dependupon the particulars of the local constitutive relation ⇣(x) = ⇣(p(x)). In other words, itremains independent of the nature of the wall response, providing accurate results forbuckled walls and elastically stretching walls alike.
734 R1-3
For a rigid conduit, q = R−1h ∆p (∼ Poiseuille).
What if flow and deformation are coupled? q = f (∆p)?
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 3 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Problem formulation: Static response
h0
x = 0
z = 0
u(x, z)
ℓy = 0
x
y
z
q
w
x = -w/2 x = w/2
For (h0/`)Re � 1, theNavier–Stokes equations forthe fluid become:
Stokes
{0 = −∇p + µ∇2v ,∇ · v = 0,
BCs
{v = 0 at y = 0, h(x , z).
h(x , z) = h0 + u(x , z).
For the top wall: Kirchoff–Love equation for bending of thin plates,[Et3
12(1− ν2)
]︸ ︷︷ ︸
≡B
∇2‖∇
2‖u = p, u =
∂u
∂x= 0 at x = ∓w
2; z = 0, `.
All other walls are rigid.
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 4 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
AsymptoticsPerturbation expansion, {V ,P,U} ∼
∑i
∑j ε
iδj{V ,P,U}ij ,assuming 0 < h0/`︸︷︷︸
≡ε
� h0/w︸ ︷︷ ︸≡δ
� 1. [Capital letters = dimensionless.]
At the leading order, we get the usual lubrication approximation:
VZ ,00(X ,Y ,Z ) = −1
2
dP00
dZ[H(X ,Z )− Y ]Y .
Then, the flow rate follows
Q = − 1
12
dP00
dZ
∫ +1/2
−1/2H(X ,Z )3 dX .
Still need to know H(X ,Z ) to finish the calculation ... at leadingorder, plate eq. reduces to beam eq. in the streamwise cross-section(H = 1 + βU):
∂4U00
∂X 4= P00(Z ), β :=
uc
h0=
h30∆p
Bδ4= O(1).
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 5 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Flow rate–pressure drop relationPutting all the pieces together:
Q =P(Z )
12(1− Z )
[1 +
β̃
20P(Z )
�����β̃2
630P(Z )2 +
�����
��β̃3
48, 048P(Z )3
]where β̃ = β/24 and ∆P = P(0).
Deformation (β ↑) increases Q for a fixed ∆P, decreases P overall:
0.2 0.4 0.6 0.8�
1
2
3
4
5
(�) Δ�
increasing β
0.2 0.4 0.6 0.8 1.0�
2
4
6
8
10
12
(�) �(�)
increasing β
β= �
β= �
β= �
β= �
Figure: (a) ∆P–Q, different β̃; (b) P(Z ), Q = 1, different β̃.
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 6 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Explicit expression for the velocity fieldEven for β = O(1), keeping only terms up to P2 is a good approx.,
P(Z ) = −10
β̃
[1−
√1 +
12
5β̃Q(1− Z )
].
Hence,
VZ (X ,Y ,Z ) = 6Q
[1 +
12
5β̃Q(1− Z )
]−1/2
×{
1 + β̃P(Z )(X + 1
2
)2 (X − 1
2
)2 − Y}Y .
-0.4 -0.2 0.0 0.2 0.4
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
X
Y
(a)
-0.4 -0.2 0.0 0.2 0.4
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
X
Y
(b)
-0.4 -0.2 0.0 0.2 0.4
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
X
Y
(c)
0.32
0.64
0.96
1.28
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 7 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Two-way fluid–structure interaction simulations
Set up simulations to benchmarkagainst (Ozsun, et al., JFM, 2013):
I Case 1: `×w = 15.5 mm× 1.7 mm,h0 = 244µm, t0 = 200µm(t0/w ' 0.12);E = (1.6± 0.2) MPa, νs = 0.499.
I Case 2: `×w = 15.5 mm× 1.7 mm,
h0 = 155µm, t0 = 605µm
(t0/w ' 0.35);
E = (1.6± 0.2) MPa, νs = 0.499.
FSI simulations usingANSYS/Fluent:
I Partitioned approach w/ “systemcoupling” module.
I BCs as discussed before; iterate tosteady state.
I Large-deformations are considered;
grid-independence was verified.
X
Y
Z
t
h
w
ANSYS®
Fluent
ANSYS®
Mechanical
Fluid Stresses
Incremental
Displacement
System
Coupling
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 8 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Top-wall deformation: FSI simulations vs. thin-plate theoryThe leading-order deformation shape shape is self-similar (in Z ) :
∂4U
∂X 4= P(Z ) ⇒ U(X ,Z )
P(Z )=
1
24
(X − 1
2
)2 (X + 1
2
)2.
FSI simulations of the “S4” case from (Ozsun, et al., JFM, 2013) verify this:
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.001
0.002
0.003
0.004
0.005
U/P
(Z)
z = 2000μm
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.001
0.002
0.003
0.004
0.005
U/P
(Z)
z = 5000μm
50mL/min40mL/min30mL/min20mL/min10mL/minChristov et al.
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.001
0.002
0.003
0.004
0.005
U/P
(Z)
z = 8000μm
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.001
0.002
0.003
0.004
0.005
U/P
(Z)
z = 12000μm
Case 1: t0/w ' 0.12(≈ a thin plate).
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 9 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Flow rate–pressure drop: Experiment, numerics & theory
•s: “S4” data (Ozsun, et al., JFM, 2013); ×s: our FSI simulation.
Dashed line – rectangular conduit (∼Poiseuille), q = h30w∆p/(12µ`).
Solid curve – (new) perturbative theory:
q =h3
0w ∆p
12µ`
[1 +
1
480
w4
Bh0∆p +
1
362, 880
(w4
Bh0
)2
(∆p)2 +1
664, 215, 552
(w4
Bh0
)3
(∆p)3
].
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 10 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Mindlin’s thick-plate-bending theory
Governing equations (Mindlin, ASME J. Appl. Mech., 1951) in terms ofdeformation u, rotation of the normal φ, stress resultants Q and M.
Equilibrium conditions (perturbative approach ⇒ keep only ∂/∂x):
∂Q∂x
+ p(z) = 0,
∂M∂x−Q = 0.
Constitutive relations are
M = B∂φ
∂x,
Q = κGt
(∂u
∂x+ φ
),
where κ is a “shear correction factor,” and G = E/[2(1 + νs)] iselastic shear modulus.
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 11 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Top-wall deformation: Simulations vs. thick-plate theoryFollowing (Mindlin, ASME J. Appl. Mech., 1951) thick-plate theory, we getanother leading-order self-similar (in Z ) deformation shape:
U(X ,Z )
P(Z )=
1
24
(1
4− X 2
)[2(t/w)2
κ(1− νs)+
(1
4− X 2
)].
FSI simulations of the “S5” case from (Ozsun, et al., JFM, 2013) verify this:
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.002
0.004
0.006
0.008
0.010
U/P
(Z)
z = 2000μm
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.002
0.004
0.006
0.008
0.010
U/P
(Z)
z = 5000μm
50mL/min40mL/min30mL/min20mL/min10mL/minThick-plate, κ=1.0
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.002
0.004
0.006
0.008
0.010
U/P
(Z)
z = 8000μm
−0.4 −0.2 0.0 0.2 0.4X
0.000
0.002
0.004
0.006
0.008
0.010
U/P
(Z)
z = 12000μm
Case 2: t0/w ' 0.35(6≈ a thin plate).
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 12 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
Flow rate–pressure drop: Experiment, numerics & theory
0 10 20 30 40 50q (mL/min)
0
5
10
15
20
25
30
35
Δp (kPa)
Experimental data (Ozsun et al.)Simulation dataRigid channelThick-plate (κ=1.0)Fit, Gervais et al.
Case 2: “S5” data (Ozsun, et al., JFM, 2013) vs. our simulations & theory.
Solid curve – (new) perturbative theory for thick plates:
q =h3
0w ∆p
12µ`
{1 +
[1
480+ f1(t/w)
]w4
Bh0∆p +
[1
362, 880+ f2(t/w)
](w4
Bh0
)2
(∆p)2
+
[1
664, 215, 552+ f3(t/w)
](w4
Bh0
)3
(∆p)3
}Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 13 / 14
Introduction Perturbative calculation Computational modeling Extension: Thick plates Summary
SummarySystematically derived flow rate–pressure drop relations for adeformable shallow microchannel, explicitly showing 3 dependencies:
q ≈ h30w∆p
12µ`
[1 + F
(∆p
E,h0
w,t
w
)]
Refs: I.C.C., V. Cognet, T.C. Shidhore, H.A. Stone, “Flow rate–pressure drop relation
for deformable shallow microfluidic channels,” in rev. for J. Fluid Mech., arXiv:1712.02687
T.C. Shidhore, I.C.C., “Static response of deformable microchannels: A comparative
modelling study,” in rev. for J. Phys.: Condens. Mat., arXiv:1709.03002.
Thank you for your attention!
Christov & Shidhore (Purdue) Modeling Low Reynolds # FSIs SIAM PD17 14 / 14
PURDUE UNIVERSITY COLLEGE OF ENGINEERING BRAND MANUAL page 31
THE SIGNATUREAlways use the correct artwork for the schools of Engineering signatures. The left side is always the university name as independent usage or the name of the division level for lockup usage with Frutiger Black font. The right side can be changed to the name of the College or divisions with Frutiger Light font.
VISUAL IDENTITY
LOGO USAGELOGO WITHTHE SCHOOL NAME
Supported by NSF grant CBET-1705637.