You are on page 1of 13

First Draft of the paper

Carlos Duran-Hernandez∗ Rene Ledesma-Alonso


Department of Computing, Department of Industrial
Electronics and Mechatronics and Mechanical Engineering
Universidad de las Americas Puebla Universidad de las Americas Puebla
San Andres Cholula, Puebla, 72810 San Andres Cholula, Puebla, 72810
Mexico Mexico
Email: jose.duranhz@udlap.mx Email: rene.ledemsa@udlap.mx

Gibran Etcheverry ∗
Department of Computing,
Electronics and Mechatronics
Universidad de las Americas Puebla
San Andres Cholula, Puebla, 72810
Mexico
Email: gibran.etcheverry@udlap.mx

Cardiovascular diseases are related to circulatory system, death every 40 seconds in USA [1]. Within the heart,
and they represent one of the main causes of death around the the cardiac cycle is divided into two stages, systole and
world. Within the CardioVascular System, the heart valves diastole. Considering systole as the initial step, this stage
play a crucial role because they are responsible of letting the is characterize by the contraction of right and left ventricles.
blood flow around the entire body. If one of these valves Until the pressure inside of these ventricles is high enough,
fails, serious pathologies may happen and deteriorate the aortic and pulmonary valves remain opened ejecting the
condition of a person. The study of artificial cardiac valves blood to the aortic and pulmonary arteries respectively.
by means of a FSI has represented an important approach to Diastole stage happens after the ventricles are relaxed then
design new and better models. Among other alternatives, the tricuspid and mitral valves open due to the pressure inside
lumped models are able to replicate the systemic tree the atria, filling both ventricles with blood once again [1].
by using passive circuit elements. The disadvantages of When aortic valve is opened, mitral valve must remain close
previous models are the high computational time required to avoid the flow going into the atrium and conversely. If
for simulating and the vast number of parameters to deal one of these stop working properly, pathologies mentioned
with and their respective adjustment for each scenario. Due before may happen. Due to this reason and thinking in
to these reasons, the content of this paper shows a well enhance the health condition of a person these valves are
established method but not applied in this area known as usually replaced by prosthetic heart valves. If one of these
System Identification. This method is able to estimate the valves is not capable of doing a full close or open properly
same outputs that can be measure with FSI and Lumped (regurgitation and stenosis, respectively) can avoid the heart
Models but overcoming their respective disadvantages. to let the blood enter or leave to continue with the blood flow
Results shown significant correlated results to the selected through the entire body, causing a problem that might end in
outputs Opening Area and Pressure Difference. death [2].
Regarding prosthetic heart valves, studies in many fields
Nomenclature have been done. According to [3] Biological Heart Valves
Ω f Fluid Domain. (BHV) has outstanding mechanical and haemodynamic
Ωs Solid Domain. properties because they are made of organic tissues but their
lifespan is short. On the other hand, Mechanical Heart
Valves (MHV) produce high stress levels that may lead to
1 Introduction
red blood cells destruction but their design helps to minimize
Atrial fibrillation, cardiac arrest, strokes and heart
back-flow and they have a much longer life time [2, 4].
failures among others disorders represent approximately one
Considering that they are made of ”unnatural” materials to
the body they present stagnation points, thrombus formation
∗ Address all correspondence related to ASME style format, figures and
and cavitation at the hinges and pivots [5]. However, the
correspondence for other issues to this author.
1 Copyright
c by ASME
main advantage of MHV is the easiness of modeling, design model by conducting a mechanical simulation (ICFD and
and fabrication, with continuous improvements in geometry FSI), then a parametric study is conducted to create a data
or mechanical properties, allowing MHV to obtain better matrix containing 45 combinations with changes in the
performances after every iteration. Nowadays the creation solid and fluid domain. Next, a not explored alternative
of these mechanical valves is intended to be from new in this cardiac valves modeling area is proposed by means
bio-compatible materials seeking to reduce damage in blood of a System Identification (SI) method capable of using
cells and stress levels. transfer functions to estimate the expected outputs without
Considering that the design of MHVs is based on needing a high computation cost or a large number of
mathematical modeling, they can be tested by using parameters to characterize the phenomena. Results are then
numerical simulations. This approach yields data, by validated by estimation cases outside the initially proposed
simulating the motion of the fluid and computing the data matrix, these cases are compared against their respective
rotation of the valve leaflets, which are impossible to FSI simulations.
measure in a real patient. The methodology employed The content in this paper shows in Section 2 the general
to simulate the performance of valves is known as methodology is explained for all the steps conducted in the
computational Fluid-Structure Interaction (FSI), and consists simulations. Section 3 shows results of every subsection,
of a combination of Incompressible Computational Fluid i.e., the meshes selected, the first FSI result, the parametric
Dynamics (ICFD) to calculate the fluid motion and a force study output curves and then the SI model with their
balance to compute the leaflets rotation. This alternative respective coefficients for each variation. Finally, Section
allows us to modify and create new geometries by using 4 and 5 present conclusions and discussion of future work,
a Computer-Aided Design (CAD) software without any respectively.
complication and they can provide a vast amount of
information depending on the necessities, such as stresses 2 Methodology
acting on the surfaces or any other force interfering with the When an object is immersed into a channel, a chamber,
fluid/structure. Studies on the flow across MHVs have been a wind tunnel or even within our bodies, for instance
carried out in [6–9]. Nevertheless, a numerical simulation when a blood cell or clot moves through the heart or
of flexible leaflet materials is completely different, since along the smallest arteries in the body, a FSI phenomenon
their motion depends strongly on the interaction with the occurs between the object and the surrounding fluid. As
fluid. In this case, the dynamics of the mechanical valve mentioned in Section 1, computational FSI is one of the
can be simulated using time-dependent Finite Element primary alternatives to comprehend how cardiac valves
Method (FEM) combined with ICFD. Many works using behave under certain fluid conditions. Contrary to a ICFD
this strategy are found in literature using different techniques experiment where inlet velocity is fixed to a constant value
[10, 11], for instance an Arbitraty Lagrange-Eulerian the entire period, in FSI, the object has its own velocity and
(ALE) [12] method, the Fictitious Domain Method (FDM) displacement as consequence of the initial configuration that
and the Immersed Boundary Method (IBM) in [2, 13, is set to be a moving part and not a static object. As a first
14] and [15, 16], respectively. Another interesting step, a simplified 2D cardiac valve model has been studied
technique specially adapted for cardiac valves study is by means of a numerical simulation, which corresponds to
proposed in [17, 18], implementing a new method called the geometry and flow conditions presented in a previous
Immersogeometric Analysis (IGA) based on the augmented experimental work [5]. The leaflets used in the experimental
Lagrangian equation. work have the same cross-sectional area along all the width
Other approach to represent the Cardiovascular System of the channel, as consequence of this, our case can be
(CVS) are the well known lumped models. By using passive simplified into a 2D case, neglecting not only the width of
elements such as resistors, capacitors and inductors, the the channel but the width of the valves as well (see Fig. 1).
Windkessel Model (WM) is capable of reproducing the This consideration affects directly in the computational time
peripheral resistance (resistance to blood flow in veins), the invested during calculations.
blood flow resistance due to the aortic or mitral valve, the Parameters used in the experiment can be classified
arterial compliance (elasticity of arteries) and the inertia of as a) Fluid domain (Ω f ) ∈ {U f ,V, f , ρ f , µ}, where the
blood flow, respectively. Different variations of these models elements are inlet velocity, stroke volume, frequency, fluid
are presented in [1, 19–24]. One disadvantage with these density and dynamic viscosity, respectively, and b) Solid
models is the inability to analyze variations in the blood domain (Ωs ) ∈ {d, l, E, ρs , ν}, where the elements are leaflet
flow distribution and wave transmission phenomena [1]. thickness, length, elastic modulus, material density and
This limitation is solved by separating into multiple blocks poisson ratio, respectively. The channel is presented in
to avoid considering the entire systemic tree, allowing to terms of semi-height, where h = 15mm, w = 50mm, and the
set different variables between different segments and each cross-section of 20h x 2h x w. The leaflets have 0.4mm
one behaves as an independent lumped model. Another and 26.3mm for thickness and length, respectively. The
limitation is the need to identify and calculate all the main purpose of this first simulation is to obtain the most
parameters to adjust to a particular physiological condition. approximated results than the experimental ones, for this
In view of previous considerations, the aim of this reason, parameters of both domains (Ω f ) and (Ωs ) listed in
paper is to introduce the study of a simplified cardiac valve Table 1 are the same to those established in [5].

2 Copyright
c by ASME
Non-Slip Condition Wall 1
0.8
Fast Fourier Transform

Amplitude
0.8 0.4
h Outlet

Velocity [U/U max]


Inlet Flow
Wall Pressure 0.6
0

h Wall -0.4
0 10 20 30 40 50
0.4 Frequencies [Hz]
10h 10h

Non-Slip Condition Wall 0.2 Unitary Pulse g(t)


Inlet Velocity Curve

Fig. 1. Layout of the system: h represents semi-height, w width of 0


the test channel (neglected in this case). Top and bottom wall, as well
as the valves, are configured as a non-slip condition objects. Original -0.2
0 0.2 0.4 0.6 0.8 1
found in [5]
Time [t/T]
.
Fig. 2. Inlet velocity curve configured to the FSI simulation. Systolic
and diastolic stages are shown. αγ is a value that controls the
Table 1. List of parameters used for the first FSI simulation
magnitude of velocity during diastole.
Fluid Value Units Solid Value Units

Uf 0.096412 m/s E 2.15x106 Pa


In order to avoid back-flow, it is necessary to implement
V 3.8x10−5 m3 ν 0.47 - two extra parameters. The α value is a constant
f 0.666 Hz ρs 1070 kg/m3 representing the ratio between systole and diastole stages
ρf 1000 kg/m3 d 0.0004 m (systole/diastole) and is equal to 0.5384 and γ allows to
µ 0.01 Pa · s l 0.026 m control the magnitude of this relation. The magnitude of
the velocity is obtained by initially considering the following
volume formula:
Z T
In a previous work [25], drag and lift coefficients were
V =A Us/d · g(t)dt (2)
measured to compare them with theoretical data by means 0
of an ICFD method. Results obtained using LS-Dyna are
close correlated to theoretical curves found in literature. where g(t) = 1 during ts (represents a unitary pulse, see
It is also observed that taking advantage of the software Fig. 2, red dotted curve) and g(t) = −α during td , A is the
capabilities to split one task into many small tasks by using cross-sectional area A = hw. By solving the integral divided
Massively Parallel Processing (MPP), helps to substantially into the two stages, we obtain the following:
reduce more than 80 % of the computational time used than
employing a Shared Memory Processing (SMP) approach. V V
For this reason, the approach used for this simulation is MPP Us = Ud = Us = Ud (3)
Ats Aαtd
as well.
For this test, measurements are focused in how much
2.1 Validation Test
and how long are the valves opened, namely Opening Area
(Aop ) and Pressure Difference (∆D) located at two different As a means to guarantee efficient simulations, it is
points P0 and P1 at −5h and 5h from the center in the x − necessary to perform a validation test by changing and
plane, respectively. evaluating different mesh sizes not only for the solid body
but the fluid domain as well. Determining which mesh
The cardiac cycle is represented by a pulsatile flow curve
can lead to the expected results not only depends on the
that in the present simulation does not corresponds to a real
accuracy of the same but also not to consume too much
cardiac cycle curve. In this case, the curve is designed to
time in execution and to complete calculations. Time
behave similarly as the system configured in [5], with aim of
step size equally affects the final results, choosing a big
obtaining similar experimental results in Aop and ∆D. Systole
value leads to inaccuracies in the model, and selecting a
stage (ts ) is usually constituted as 2T
5 and diastole stage (td ) small value produces accurate calculations but requires a
as the remaining time in the period. The normalized shape
high computational time procedure. For this reason, three
for the inlet curve is formed by Eq. (1) and it is divided in
different validation test are proposed. First test consists
five stages depending on the time t (3 stages correspond to
on selecting the most appropriate mesh size for the solid
systole and two for diastole), where A, B,C, D are parameters
domain by carrying out a structural analysis. Second test
to modify amplitude, slope, phase and a non-zero center
involves a FSI configuration to select the mesh size for the
amplitude, respectively (see Fig. 2, blue curve).
fluid domain. Finally, third test considers both solid and fluid
meshes selected previously but with different time steps sizes
y(t) = Atanh(Bt +C) + D (1) in a simulation time of T = 1.502s.

3 Copyright
c by ASME
2.1.1 Solid Mesh
The selection of the mesh size in the Solid domain (Ωs )
is performed by conducting a structural analysis using the
superior simplified valve model. Initially, a force is exerted
into the tip of the valve during T seconds. From t = 0 to
t = 1s, the force linearly increases from zero to |Fx | and
F5
then it remains constant until T = 1.5s. Input parameters
are the following: Applied Force Fx = 150mN, Simulation
time T = 1.5s, Elastic Modulus (E) = 2.15x106 Pa, Poisson
Ratio ν = 0.47 and density of the fluid ρs = 1070 mkg3 . All S5
nodes of the superior tips are fixed to the top wall and have
zero displacement and velocity. Output values measured
in this test are: Von Misses Stress, execution time and Fig. 3. Selected Meshes. Top image represents a part of the fluid
displacement measured that has to be lower than h to prevent domain mesh where sections A and B can be appreciated. Bottom
node collisions. Nine different mesh sizes (see Fig. A1 in image represents a small part of the solid domain mesh.
complementary appendix) are tested in order to identify the
most appropriate mesh to implement for future simulations. The fluid domain (Ω f ) is divided into two nodes sections:
Configuration S9 is considered as the more complex and a section A representing the nodes located before and after
finest mesh in this test. All the remaining configurations are the valves, these zones contains fewer nodes than the other
compared against S9 measuring their absolute and relative section because the information measured in this zone is not
error with respect to it. Table 2 contains the information of that relevant for this study. On the other hand, a section B
all meshes and their respective results. Displacement of the is created by selecting a rectangular area that represents the
valve is measured from the top wall until the lowest tip of the intermediate zone where valves are placed; in this particular
valve. According to Table 2, S9 is the finest mesh with the zone there is a significant quantity of nodes in order to obtain
most considerable number of elements and nodes, this lead to and measure the most accurate values of the phenomena
obtain more precise results in the Von Misses Stress (VMS) occurring in this area. Eight different configurations are
and displacement measures. Nevertheless, S9 is formed by a tested, and they can be seen in Fig. A2 (in complementary
really heavy mesh, that if it is merely consider this structural appendix). Five of them contains the section B with the
without an ICFD or a FSI analysis, it required 20 minutes to rectangular area, and in the rest of them there is only an
finish calculations only for T = 1.5s. increment in the number of nodes for the entire fluid domain.
Seeking to find a reliable result correlated to the ones Input fluid mechanical properties for this simulation are the
found in S9, the first two configurations S1 and S2 can following: (Ω f ) = { 0.09641 m/s, 3.8 x10− 5m3 , 0.666
be omitted because the displacement obtained is practically Hz, 1000 kg/m3 , 0.001 Pa · s } and (Ωs ) = { 0.0004 m,
twice than S9 and the VMS is less than half of S9. On 0.26 m, 2.15 x106 Pa, 1070 kg/m3 , 0.47 }. Finally, the
the other hand, S3 shows a lower value in VMS, and the output parameters to measure and to compare are: maximum
displacement observed is greater than h, for these reasons, velocity in X − plane, pressure before and after the valves
it is also rejected. Regarding computational time, S7 and S8 (at -5h and 5h distance from the center), VMS at the superior
are rejected because they equally have a high execution time valve and execution time. Considering a period T = 1.5015s,
compare to other alternatives. Even when S4, S5 and S6 have these FSI simulations are carried out only for a t = 0.4s.
similar results, S6 is eliminated due to the topology of the Table 3 contains the information of all meshes and their
mesh, trying to keep one unique shape on the mesh and not results.
combining them with others. Finally, S4 and S5 represent Following the same criteria than before, the last
the best options considering that the values measured are configuration is also selected as the most precise due to the
close enough to S6 and S7; the small difference in the number of elements and nodes on it. Section B in these tests
displacement measured represent a key factor to consider represents a crucial part in calculations, and this section is
when choosing the final model. After several attempts to merely included from configurations F1 to F5. F1 contains
simulate a FSI, it is observed that the closest the valves are the coarser mesh, having only 10 nodes in height and 50
to each other (specifically the nodes on the tips) the greater in semi-length for the section A and it also has the biggest
the probabilities are than the simulations fails to continue internal mesh size for the section B. On the other hand, F5
and to finish. Once both valves are placed in the domain, contains twice the nodes in height and semi-length than F1
the distance between superior and inferior valve tips using but it has the same internal mesh size. From F6 to F8
S4 is only 3.8mm, whilst using S5 the gap is 6mm. Taking there is no section B because they already have three or
in consideration these gaps the final model to use in future four times the number of nodes than previous configurations.
simulations is configuration S5 (see Fig. 3, bottom). Regarding the measurements, configurations F3, F4, F6 and
F7 are neglected due to the significant computational time
2.1.2 Fluid Mesh required to finish calculations. Results of configurations F1
The second validation test is similar to the previous one and F2 are close to F5 even when they have less amount of
with the main difference that this test uses a FSI approach. nodes but similar time of execution . Nevertheless, absolute

4 Copyright
c by ASME
Table 2. Solid Mesh Validation Test Comparisons. Configuration S5 is the mesh selected. |Fx | = 150mN
Mesh No. Shell No. VMS Max Value Displacement Execution Absolute Relative
Case
Size Elements Nodes [v-m] [m] Time Error Error

S1 0.0005 108 122 3.25E+04 0.0229 12 sec 4.696E+04 0.591


S2 0.0004 134 148 4.05E+04 0.0221 12 sec 3.896E+04 0.490
S3 0.0002 528 411 5.61E+04 0.0151 23 sec 2.336E+04 0.294
S4 0.00015 1056 720 7.04E+04 0.0131 28 sec 9.060E+03 0.114
S5 0.0001 2588 1574 7.03E+04 0.0120 58 sec 9.136E+03 0.115
S6 S5 + Triangle Mesh 7764 4154 7.06E+04 0.0120 2 min 44 sec 8.860E+03 0.112
S7 S5 + Hexagonal Mesh 7764 8307 7.01E+04 0.0122 4 min 50 sec 9.360E+03 0.118
S8 S5 + Hexagonal + Triangle 31056 16071 7.70E+04 0.0123 11 min 9 sec 2.470E+03 0.031
S9 S5 + Hexagonal + Hexagonal 31056 32133 7.95E+04 0.0122 20 min 56 sec 0.000E+00 0.000

Table 3. Fluid Mesh Validation Test comparisons. Configuration F5 is the mesh selected.
Nodes Nodes Number Max Pressure Pressure Max Execution Absolute Relative
Case Section A Section B Surface Velocity X Before After VMS
Time Error Error
(Height x Semi-length) Internal Mesh Nodes [m/s] [Pa] [Pa] [v-m]

F1 10 x 50 5.00E-04 1320 2.15E-01 34.40 7.31 8.276E+04 1 hr 41 min 2.661E+03 0.03320


F2 10 x 50 4.00E-04 1320 2.24E-01 32.30 6.98 7.939E+04 1 hr 30 min 7.132E+02 0.00890
F3 10 x 50 2.00E-04 1320 2.22E-01 37.40 8.02 7.933E+04 2 hr 53 min 7.730E+02 0.00964
F4 10 x 50 1.00E-04 1320 2.15E-01 32.60 7.22 7.807E+04 7 hr 54 min 2.033E+03 0.02536
F5 20x100 5.00E-04 1500 2.18E-01 31.70 7.24 8.067E+04 1 hr 47 min 5.735E+02 0.00716
F6 30x200 NO 1920 2.18E-01 33.80 7.43 8.454E+04 4 hr 57 min 4.441E+03 0.05542
F7 30x300 NO 2320 2.18E-01 35.40 7.63 7.802E+04 4 hr 12 min 2.081E+03 0.02596
F8 40x500 NO 3140 2.26E-01 34.80 7.65 8.010E+04 9 hr 12 min 0.00 0.00

and relative errors of F5 against F8 are lower than the others velocities or Es, time step TS2 may be unable to complete
two options. Concluding, future models are constructed by the simulation time and it requires another value to finish.
using the fluid configuration F5 (see Fig. 3, top). For this reason, the conclusion of this test is to exclusively
provide a threshold of functional values to select within the
range of 0.0004 < tsz < 0.0009 for future simulations in the
2.1.3 Time Step Size data matrix. Values within this range are able to complete
Once the mesh size for both domains is selected, the or at least achieve more than 50% of the period configured
conclusive test is to choose the most appropriate value for for the simulation time. This 50% of simulation time is
the Time Step Size (tsz ). Once more, a FSI simulation is enough to observe the behavior of the valve when opening
conducted for a period of T = 1.502s. If a high value is and closing.
selected, the results obtained are inaccurate. On the other
side, the lower the value the greater the computational time
and accuracy of the model will have. Initially, the established 2.2 Parametric Study
threshold is 0.0001 < tsz < 0.008; however, after some After the first validation tests give the general idea
attempts all cases using a tsz greater than 0.001 are unable to on how FSI works, it is necessary to comprehend how a
exceed 3 minutes of execution time. Therefore, the threshold cardiac valve behaves under different conditions, not only
is shrunk to 0.0001 < tsz < 0.001, and the results can be by observing changes in the fluid but the solid properties as
appreciated in Table 4. Output parameters measured are well. This approach is proposed to be done by conducting
velocity in X − plane and vorticity in Z − plane. a parametric study. Two elements are established to be
The results are compared with the finest tsz = 0.0001 changed in this parametric study. The first one is the
in different time moments. Nevertheless, selection of the elastic modulus E within Ωs and the second represents the
time step is directly affected on the specific phenomenon frequency of the fluid f within Ω f . The proposal is to obtain
analyzed. By carrying out this validation test, it can be 45 different combinations retrieved from five variations in
concluded that the same time step size does not work for E and nine variations in f . The models obtained from the
all cases but only for a few of them. Reaching a complete validation tests are utilized for this parametric test, however,
simulation for this specific E and period T was only possible considering than previous validation test are only tested for
by selecting T S2. The period configured is unreached any one specific E, a new structural analysis is done for each of
other time step size. In other scenarios with different periods, the new four variations in E.

5 Copyright
c by ASME
Table 4. Time Step validation test, using configuration S5 and F5 previously selected. The simulation time is set to T = 1.5015s.
Time Step Simulation Vorticity Z Velocity in X Vorticity Z Velocity in X Vorticity Z Velocity in X Execution Absolute Relative
TS Size Time Reached at 0.2 s at 0.2 s at 0.278 s at 0.278 s at 0.825 s at 0.825 s Time
Error Error
[s] [s] [s−1 ] [m/s] [s−1 ] [m/s] [s−1 ] [m/s] [s]

TS1 1.00E-03 0.236 8.069E+03 0.0978 - - - - 2 hr 37 min 2.685E+03 0.499


TS2 8.00E-04 1.5015 6.188E+03 0.1018 2.320E+03 0.2325 1.148E+03 0.3485 15 hr 46 min 1.661E+03 0.243
TS3 6.00E-04 0.8304 2.325E+03 0.0964 4.562E+03 0.2195 1.278E+03 0.3556 12 hr 24 min 6.158E+03 0.899
TS4 5.00E-04 0.206 1.745E+03 0.0973 - - - - 6 hr 28 min 3.639E+03 0.676
TS5 4.00E-04 0.8252 1.740E+03 0.1004 2.695E+03 0.2306 6.172E+03 0.3523 15 hr 41 min 4.876E+03 0.712
TS6 3.00E-04 0.9897 1.451E+03 0.0954 1.447E+03 0.2197 1.814E+03 0.3721 1 d 3 hr 36 min 3.949E+03 0.577
TS7 2.00E-04 0.1986 1.167E+03 0.1005 - - - - 17 hr 4.217E+03 0.783
TS8 1.00E-04 0.2785 5.384E+03 0.2312 1.463E+03 0.2213 - - 1 d 15 hr 3 min 0.00 0.000

Solid Domain Table 5. Velocities used for the Parametric Study


(Ωs) Velocity
E - Elastic Frequency Period
Modulus Magnitude
[Hz] [s] [m/s]

j=5 1.000 1.000 0.145

Fluid Domain
0.900 1.111 0.130
Aop - Opening
(Ωf) 0.800 1.250 0.116
i=9

Area
0.700 1.429 0.101
f - frequency ΔD - Pressure
U - velocity Difference 0.666 1.502 0.097
0.600 1.667 0.087
Fig. 4. Data Matrix containing 45 combinations between Ω f and 0.500 2.000 0.073
Ωs . The parameters adjusted are the E and the frequency and 0.400 2.500 0.058
velocity of the fluid. 0.333 3.003 0.048

2.2.1 Parametric Structural Analysis for Valves Models


Inlet velocity magnitude of each FSI simulation is
This study is conducted with the aim of obtaining the
calculated by using Eq. (3) and they are registered in Table 5.
necessary models to adopt for each valve configuration in
Time and velocity magnitude of each case are then multiplied
the parametric study. The objective is to find the necessary
to the normalized curve proposed in Fig. 2.
force required to measure a displacement in the valves to
approximately h = 15mm. The configuration of this test 3 Experimental Results
is exactly the same than the solid validation test, with the This section shows the results of the parametric study
unique difference that the force applied in the tip has a period and the parametric structural analysis of each E variation.
of T = 3s. The displacement observed in the y − plane has Moreover, this section includes the new approach proposed
to be the closest but lower to 15mm, otherwise, a collision to handle with these kind of problems, not only the
between them occurs and this avoids the FSI simulation to methodology of the same but the results as well.
properly continue and complete the calculation process. The
3.1 FSI Results
remaining input parameters such as, ρs , ν and inlet force
Using all the models gathered from the validation test,
curve are the same as in the solid validation test. For the first
it is possible to obtain the first attempt of the FSI approach.
analysis the E is chosen to be E = 2.15x106 Pa; for these new
Results observed in this first analysis, are close correlated
tests, the range is established to have two greater and two
with the ones observed in the experimental. In Fig. 5, a
lower values than the original. As result of this, E is varied
complete simulation time is shown and the behavior of the
as follows: 1x105 , 1x106 , 2.15x106 , 1x106.5 (3.16x106 ) and
valves can be appreciated. For all cases the inlet curve
1x107 [Pa].
used is designed in such a way that back-flow behavior is
avoided and only a correct functionality can be observed.
2.2.2 Inlet Velocity Curves According to the results shown in Duran-Hernandez [26],
All the 45 combinations assembled in a data matrix the FSI simulation is successfully replicated, with slight
share the same V , ρ f , µ, ν, ρs , d and l. Considering differences that are produced due to the 2D simplification,
five different values for the E, the data matrix is also however, the behavior observed qualitative represents the
constructed by changing velocity and period of the fluid phenomena desired. Furthermore, it was detected that
into nine different variations. Frequencies selected are: when observing correct functionality in the simulation, two
[0.333, 0.4, 0.5, 0.6, 0.666, 0.7, 0.8, 0.9, 1] Hz. situations may happen: a symmetrical and asymmetrical

6 Copyright
c by ASME
Fig. 5. Results obtained from the first FSI simulation. Nine different stages are shown with respect to Vx .

1
0.8
0.6
0.4
0.2
0
-0.2
1
0.5
0
-0.5
-1
1
0.8
0.6
0.4
0.2
0
-0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [t/T]

Fig. 6. Input and output parameters observed during the first FSI simulation. Top represents Flow Rate, Center is ∆D and Bottom is Aop .
All values are normalized. Green marks represent the nine stages selected in Fig. 5.

return. Symmetrical return is when both valves go back to valve. Input (Flow Rate) and output parameters (Aop and
their initial position, i.e., the closest to y = 0 ( see Fig. 5, ∆D) of Fig. 5 can be seen in Fig. 6. Flow Rate is obtained by
last three stages ). On the other hand, the asymmetrical Eq. (4).
return represents the opposite; when diastole occurs any of m3
the valves do not return to their initial position and remains FR = Uhw[ ] (4)
s
close to the latest position during systole and the other valve
moves up/down until the tip reaches any part of the other

7 Copyright
c by ASME
Table 6. Displacement of each E under different forces applied. Time= 0 sec Time= 1 sec
0.015
Elasticity Force Displacement 0.01
0.005
Modulus Applied Measured 0

Y-Coordinates [m]
[Pa] [mN] [mm] -0.005
-0.01
-0.015
1x105 4.9 0.290 Time= 2 sec Time= 3 sec
0.015
1x106 53.0 0.163 0.01
0.005
2.15x106 104.2 0.162
0
3.16x106 163.0 0.220 -0.005
-0.01
1x107 496.0 0.156 -0.015
-0.01 0 0.01 0.02 -0.01 0 0.01 0.02
X-Coordinates [m]
1.5x106 84.0 0.452
E1e5 E1e6 E215e6 E316e6 E1e7 VE150e6 VE150e7
1.5x107 780.0 0.333

Fig. 7. Displacement observed in valves during T = 3s.

3.2 Structural Analysis for Valves Modeling In order to do so, the alternative approach to deal with
fluid mechanics simulations is proposed by means of SI
This analysis helps to comprehend how the design
methods. These systems allow estimating an output handling
behaves under different force loads with different hardness.
the information of past and future events on the input.
Whether the hardness increases so it does the force required
Working as a black box, these methods construct a general
to achieve the desired displacement in the valve. In
polynomial function based on two sets of coefficients, and
difference to the previous structural analysis conducted in
according to the order of the system k, this polynomial will
the solid validation test, the principal objective is to find what
have bk and ak , the numerator and denominator coefficients,
force is needed in order to obtain a distance from y = 0 to the
respectively.
tip of the valve bigger than 0.15mm. Consequently, once both
The SI model proposed in this paper is an
valves are put together in the domain, the distance between
AutoRegressive eXogenous (ARX) method using a
them is approximately bigger than 0.3mm. Table 6 shows
Single-Input Single-Output (SISO) system that considers
the results obtained after a simulation time of T = 3s in every
each inlet flow rate as one input and each Aop or ∆D as
case. Last two cases represent the configuration for verifying
one output. In order to estimate a general model for all
the model, and these two E are located outside the original
the combinations presented in the parametric study, the
matrix proposed.
following procedure is presented:
The behavior of each case can be appreciated in Fig. 7
along four time stages between the period selected. Finally, + For each E:
Fig. 8 shows these frequencies increment throughout the – The output n only corresponds to the n input.
simulation time. When the lowest E is tested, it merely – Nine models are obtained from one E.
presents a few oscillation moments whereas the highest – Once these nine models are obtained, the average
modulus repeats this oscillation more than 20 times but with value of each coefficient is taken to obtain a
lower amplitude. Whether the E increases, so it does the general polynomial function for this E.
oscillation of the valves and the final location of all them Finally, each one of these five general polynomial
remains between the desired distances. In all cases the settle functions is able of reconstructing each of the nine cases
time is a consequence of the inlet curve used that goes from 0 of their respective E. At the end, we have five transfer
to the applied force in a constant slope during the first second functions for Aop and five for ∆D that follows the next
and stays constant until it reaches the desired period. structure proposed by [27].
3.3 System Identification Method B(s)
ARX(s) = (5)
The essential core of this paper is to present an 1 − [A(s)]
alternative method that helps researchers to replicate the
closest results as possible as if they we were doing a FSI where, B(s) and A(s) represent the continuous time
simulation. The advantages of this method lies in skipping coefficients for the numerator and denominator respectively.
many details required to configure for the simulation The unitary element in the denominator represents an
namely, geometry, boundary conditions, mesh quality and addition of an extra k +1 coefficient. All the A(s) coefficients
its topology and mainly, the highest computational time are multiplied by a negative factor. This means that if the
that these approaches require to accomplish short simulation order system k = 2, the denominator will have a3, −a2, −a1
times. and −a0 coefficients.

8 Copyright
c by ASME
E1e5 E1e6
0.015
E1e5 F= 4.93mN E1e6 F= 53mN
0.005
-0.005
-0.015
Displacement [m]

E215e6 E316e6
0.015
E2.15e6 F= 104.2mN E3.16e6 F= 163mN
0.005
-0.005
-0.015
E1e7 VE150e6 and VE150e7
0.015
E1e7 F= 496mN E1.5e6 F= 84mN
0.005 E1.5e7 F= 780mN

-0.005
-0.015
0 1 2 30 1 2 3
Time [sec]
Fig. 8. Oscillation observed within the valves. Transient and stationary part can be observed. Last figure contains verification test models

Data E215e6 T= 1.000 sec Data E215e6 T= 1.111 sec Data E215e6 T= 1.000 sec Data E215e6 T= 1.111 sec
200 0.04
0 0.02
-200 Original Reconstructed Original Reconstructed 0
-400 -0.02
Data E215e6 T= 1.250 sec Data E215e6 T= 1.428 sec Data E215e6 T= 1.250 sec Data E215e6 T= 1.428 sec
200 0.04
Aop Amplitude

0 0.02
DP Amplitude

-200 Original Reconstructed Original Reconstructed 0


-400 -0.02
Data E215e6 T= 1.502 sec Data E215e6 T= 1.667 sec Data E215e6 T= 1.502 sec Data E215e6 T= 1.667 sec
200 0.04
0 0.02
-200 Original Reconstructed Original Reconstructed 0
-400 -0.02
Data E215e6 T= 2.000 sec Data E215e6 T= 2.500 sec Data E215e6 T= 2.000 sec 0 0.2 0.4 0.6 0.8 1
200 0.04
0 0.02
-200 Original Reconstructed 0
-400 -0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Time/T Time/T
Original Reconstructed Original Reconstructed

Fig. 9. Output ∆D for E = 2.15x106 . Only one period is rejected Fig. 10. Output Aop for E = 2.15x106 . Two periods are rejected
for this case. for this case.

Before obtaining the final coefficients for each E, it slight change in the estimation process and specifying that
is necessary to test if all the nine periods chosen are m = 2 and n = 3 , a biproper transfer function conversion
representing the same event required to measure at the is obtained, this is m = n. This allows obtaining better
output. This is accomplished by obtaining a bode diagram estimations in Aop output. The case of ∆D does not present
and zero and poles diagram for each period and E. By doing a remarkable difference in estimation when applying this
these two diagrams it is possible to reject 1, 2 or 3 periods in change, thus we leave the estimation with the original k = 2
every E. One example of this process is shown in Fig. B1 and the strictly proper transfer function for this output.
and Fig. B2 in the complementary appendix. Once the coefficients are identified in each E, an average
Initially, both cases (Aop and ∆D) are tested with the value of all them is obtained to have a single polynomial
same system order m = n = 2, i.e., two coefficients for bm function for each E. This single polynomial function is tested
and an . This leads to have a strictly proper transfer function on the reconstruction of the original nine output signals that
conversion where m < n. However, by implementing a corresponds to every period. Two sets of results are also

9 Copyright
c by ASME
Table 7. Average coefficients obtained for ∆D. System order m = n = 2. Coefficients follow Eq. (5)

∆D Parametric Study Modulus Verification Modulus


Coefficients 1e5 [Pa] 1e6 [Pa] 2.15e6 [Pa] 3.16e6 [Pa] 1e7 [Pa] 1.5e6 [Pa] 1.5e7 [Pa]

b1 -7.952E+07 -8.362E+07 -7.884E+07 -8.724E+07 -6.728E+07 -8.195E+07 -6.189E+07


b0 7.951E+07 8.359E+07 7.878E+07 8.717E+07 6.718E+07 8.191E+07 6.174E+07

a1 0.0327 0.0126 0.0709 0.1989 0.7043 0.0742 1.0566


a0 0.3883 0.4081 0.3636 0.2223 -0.1399 0.3508 -0.4242

Table 8. Average coefficients obtained for Aop . System order m = 2, n = 3. Coefficients follow Eq. (5)

Aop Parametric Study Modulus Verification Modulus


Coefficients 1e5 [Pa] 1e6 [Pa] 2.15e6 [Pa] 3.16e6 [Pa] 1e7 [Pa] 1.5e6 [Pa] 1.5e7 [Pa]

b2 3.550E+03 6.499E+03 6.987E+03 6.957E+03 6.837E+02 5.810E+03 -8.120E+02


b1 -7.092E+03 -1.308E+04 -1.413E+04 -1.408E+04 -1.504E+03 -1.171E+04 1.412E+03
b0 3.542E+03 6.584E+03 7.145E+03 7.124E+03 8.239E+02 5.901E+03 -5.948E+02

a1 0.9402 1.1460 1.1277 1.1663 1.1371 1.0857 1.2199


a0 0.0594 -0.1656 -0.1652 -0.2119 -0.1885 -0.1094 -0.2984

shown in Fig. 9 and 10. They correspond to the first are additionally included in Fig. 7, Fig. 8 and Table 6. The
E used to configure the FSI simulation and compare the estimated output for Aop and ∆D follow the same tendency
results with the experimental data found in [5]. For this to those in the parametric study. Using different orders
case, the use of the bode and zero-poles diagrams let us improved the signal for Aop but not for ∆D, which uses the
know that the last period in ∆D does not correspond to the original method. Table ?? shows the results obtained for
output behavior expected. In the case of Aop , the last two these simulation tests.
periods are also rejected under the same criteria. The average Results found for the internal desired E located between
coefficients found for each E are listed in Table 7 and Table 8. the two first modulus corresponds to Fig. 11 and Fig. 12 and
The entire results of the parametric study are shown in the those outside the matrix corresponds to Fig. 13 and Fig. 14.
complementary appendix C. Coefficients found for these cases are also shown in Table 7
In the figures showing the ∆D and Aop estimation, it can and Table 8 In general terms, it is clearly evident that this
be appreciated that not all curves in red are reaching the final method is limited when estimating ∆D. Conditions inside the
simulation time; this is because for that specific case the matrix are better estimated for both outputs than the cases
simulation time is uncompleted. However, the tsz used for that require an extrapolation for estimation.
that same case was the only one to be able of accomplishing In this manner, this model is only suitable for specific
the longest simulation time. On the other hand, blue (black thresholds of conditions and parameters. Another limitation
for verification results) curves are the reconstruction curves is the quality of the estimation depends completely on
using the SI model, and the sudden stop in some of the the input signal. The noise represents another factor that
figures is a consequence of the high noise percentage found affects the comparison between the original output with the
in the original output signal. Even after performing a signal estimated signal, whether is a numerical or physiological
filtering, this noise is unable to be rejected. For these reasons, noise; the current model is incapable to estimate under those
estimation is merely conducted until before the significant conditions with accuracy. For these reasons, it is necessary
noise occurs or when the simulation stopped. to perform a signal filtering before generating the model and
calculate the respective coefficients taking in consideration
3.3.1 Extra Cases - Model Validation - that this filter treats only with the noise and not affecting the
Finally, the last step is the model validation by using initial phenomena.
information that is originally unconsidered within the matrix.
In order to do so, a fitting curve that contains previous 4 Conclusions
average coefficients is used to determine the interpolated The main contribution of this paper is to propose
points to calculate the respective coefficients for the desired an alternative to the FSI method essentially involve into
new parameters. solving cardiac valve modeling problems. This alternative
The simulation parameters are the same used in previous proposed is a System Identification method, which is not
results: the selected solid and fluid mesh size for the new only faster but also avoids the necessity of configuring a
desired Es, the order system m = n = 2 for ∆D and m = 2, n = considerable number of parameters as it is required for
3 for Aop . Results obtained during the structural analysis instance in lumped models such as the Windkessel. By

10 Copyright
c by ASME
Data E150e6 T0 833 sec Data E150e7 T0 833 sec
300 300
0 0
-300 -300
-600 -600
Data E150e6 T0 900 sec Data E150e7 T0 900 sec
DP Amplitude

DP Amplitude
300 300
0 0
-300 -300
-600 -600
Data E150e6 T1 2005 sec Data E150e7 T1 2005 sec
300 300
0 0
-300 -300
-600 -600
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Time/T Time/T
Original Validated Original Validated

Fig. 11. Validation results in pressure difference measured for Fig. 13. Validation results in pressure difference measured for
E = 1.5x106 Pa for 3 periods. E = 1.5x107 Pa for 3 periods.

Data E150e6 T0 833 sec Data E150e7 T0 833 sec


0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
Data E150e6 T0 900 sec Data E150e7 T0 900 sec
Aop Amplitude

Aop Amplitude
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
Data E150e6 T1 2005 sec Data E150e7 T1 2005 sec
0.04 0.04
0.02 0.02
0 0
-0.02 -0.02
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Time/T Time/T
Original Validated Original Validated

Fig. 12. Validation results in opening Area measured for Fig. 14. Validation results in opening Area measured for
E = 1.5x106 Pa for 3 periods. E = 1.5x107 Pa for 3 periods.

means of polynomial functions, SI is able to overcome the to carry out for a single cardiac cycle, hence, obtaining an
significant computational cost disadvantage and simulation average value of these signals is not possible and therefore
time required by FSI. The simulations reported represent there is more noise observed in the outputs, mainly in
a simplification of the experimental setup previously the pressure difference. These smoothness variations are
mentioned. The current model is a 2D test that contains attacked through a previous signal filtering to use the cleanest
as outputs the opening area and pressure difference of the signal as possible for the input required by the SI model.
valves. Regarding the parametric study, generating a
The results obtained from the first simulation allowed considerable number of variations helped to understand the
to infer the design of the input curve is decisive in seek relevance of the validation test for the meshes and the time
of obtaining results as closer as possible to those reported step sizes. For each variation, the mesh played an essential
experimentally. The dimensionality simplification equally role in accomplish a complete simulation time; moreover,
produces a minor difference that involves the inlet volumetric it was discovered that using the same time step size does
flow. For our case, it is only transmitted along the length not work for all cases. Therefore, establishing a range of
and height of the channel and not in the width as the time steps sizes with a similar order of magnitude, allowed
experimental setup, giving as a result, small numerical to complete the entire study and to obtain results that were
differences in measurements. In addition, the experimental remarkably similar in order of magnitude in comparison to
is carried out for a large number of cycles, and at the the initially proved case. Taking advantage of the software
end the average value of all cycles is used to obtain a capabilities to split the job into every processor for handling
unique curve representing only one cycle; this also helped a locally parallel task helped to considerably reduce the
to reduce possible measurement noise in the signals. On necessary time to complete all cases. For example, if the
the contrary, the simulations conducted were merely possible simulation with the longest period (T = 3s) is conducted

11 Copyright
c by ASME
by using a regular shared memory process, the simulation approach by equally using a SI model, and this considers
time is reached until 8 or 9 days of calculations, whilst to be able to estimate results when you possess previous
using massive parallel processing the simulation time is knowledge of the expected output. This is in order to
accomplished after 1 to 1.5 days approximately. These recognize a possible standard deviation when you have a
reductions in computational time represent more than 80% measurement of the behavior of an existing valve and then
of the time required by SMP, therefore, when SMP is know if its operation is deviating from the expected value
scarcely able to conclude 1 case in a week, MPP finishes and then adjust what is necessary or to a design change
4 or 5 cases in a similar time. Finally, with respect to the requirement.
results obtained for the SI model from a SISO system, by
verifying each of these polynomial functions, more than
90% of them is capable of estimating the cases needed in References
the parametric study. This method uses coefficients to create [1] Peña Pérez, N., 2016. “Windkessel modeling of the human
transfer functions for every case proposed in methodology, arterial system”. B.S. thesis.
and its computational complexity is low, where the order [2] Stijnen, J., De Hart, J., Bovendeerd, P., and Van de Vosse,
system is equal to k = 2 for both outputs. Whilst doing a F., 2004. “Evaluation of a fictitious domain method for
FSI requires days, this method is able to obtain results of the predicting dynamic response of mechanical heart valves”.
45 combinations in less than 30 minutes; this time includes Journal of Fluids and Structures, 19(6), pp. 835–850.
not only the coefficients calculations but the post-processing [3] Yoganathan, A. P., He, Z., and Casey Jones, S., 2004. “Fluid
mechanics of heart valves”. Annu. Rev. Biomed. Eng., 6,
of all data generated by the FSI simulations, the data matrix
pp. 331–362.
creation and the signal filtering, among other steps as
[4] López-Zazueta, A., Ledesma-Alonso, R., Guzman, J., and
well. Comparing the graphs of each output, the results are Zenit, R., 2011. “Study of the velocity and strain fields
significantly correlated in most cases, however this method in the flow through prosthetic heart valves”. Journal of
only considers linear approaches. Employing a non-linear biomechanical engineering, 133(12), p. 121003.
approach may lead to substantially improve these results. [5] Ledesma-Alonso, R., Guzmán, J., and Zenit, R., 2014.
“Experimental study of a model valve with flexible leaflets
in a pulsatile flow”. Journal of Fluid Mechanics, 739,
5 Discussions pp. 338–362.
Addressing the problem of detecting malfunctions in [6] King, M., Corden, J., David, T., and Fisher, J., 1996. “A
three-dimensional, time-dependent analysis of flow through a
the cardiac valves, the alternatives that have been proposed
bileaflet mechanical heart valve: comparison of experimental
as the replacement of the already existing Biological Heart
and numerical results”. Journal of biomechanics, 29(5),
Valves, the Mechanical valves are the best option due to pp. 609–618.
the great possibility of continuous improvement in design [7] Krafczyk, M., Cerrolaza, M., Schulz, M., and Rank, E. .,
through the use of a CAD software. This approach 1998. “Analysis of 3d transient blood flow passing through an
helps to reduce many of the possible limitations that has artificial aortic valve by lattice–boltzmann methods”. Journal
its counterpart, however this same advantage represents a of Biomechanics, 31(5), pp. 453–462.
limitation when it comes to performing many repetitions of [8] Lemmon, J. D., and Yoganathan, A. P., 2000.
the same design under different conditions in the fluid and “Three-dimensional computational model of left heart
solid domain. The proposed method is capable to estimate diastolic function with fluid–structure interaction”. Journal
the behavior of both selected output phenomena, with the of biomechanical engineering, 122(2), pp. 109–117.
significant advantage that there is no need to consider [9] Bluestein, D., Li, Y., and Krukenkamp, I., 2002. “Free
emboli formation in the wake of bi-leaflet mechanical heart
previous knowledge of the expected output. Nevertheless,
valves and the effects of implantation techniques”. Journal of
it is known that in order to develop a good SI model, it is biomechanics, 35(12), pp. 1533–1540.
necessary to have a previously simulated results database to [10] Baccani, B., Domenichini, F., and Pedrizzetti, G., 2003.
be used to prepare the model and then obtain the polynomial “Model and influence of mitral valve opening during the
coefficients before even being able to validate new design left ventricular filling”. Journal of biomechanics, 36(3),
variations. The method additionally requires the signals pp. 355–361.
to be as clean as possible to get closest to the estimation [11] Khalili, F., Gamage, P. T., and Mansy, H., 2018.
desired, otherwise the noise produces poor estimations. In “Prediction of turbulent shear stresses through dysfunctional
addition, it is necessary to provide a signal that consists bileaflet mechanical heart valves using computational fluid
of a proper number of samples to increase the precision dynamics”. arXiv preprint arXiv:1803.03361.
of the model, and one of the advantages of using a FEM [12] Horsten, J. B. A. M., 1990. “On the analysis of moving heart
valves: a numerical fluid-structure interaction model”. PhD
software is that you can easily adjust the number of samples
thesis, Technische Universiteit Eindhoven.
wanted at the output. Regarding validation, the accuracy [13] De Hart, J., Peters, G. W., Schreurs, P. J., and Baaijens,
of estimation decreases when the parameters handled are in F. P., 2000. “A two-dimensional fluid–structure interaction
places where an extrapolation is required; on the contrary, model of the aortic value”. Journal of biomechanics, 33(9),
when interpolation is used in intermediate areas of the data pp. 1079–1088.
matrix, the estimation is still remarkably accurate. As [14] De Hart, J., Peters, G., Schreurs, P., and Baaijens, F., 2003. “A
possible future work, it is proposed to tackle a diverse three-dimensional computational analysis of fluid–structure

12 Copyright
c by ASME
interaction in the aortic valve”. Journal of biomechanics,
36(1), pp. 103–112.
[15] Peskin, C. S., 1972. “Flow patterns around heart valves: a
numerical method”. Journal of computational physics, 10(2),
pp. 252–271.
[16] Wang, Z., Yuan, Q., Shen, B., Tang, D., and Zhang,
X., 2016. “Mathmatic abstraction for fluid-structure
interaction analysis of bioprosthetic heart valves with
immersed boundary method”. DEStech Transactions on
Engineering and Technology Research(mcemic).
[17] Kamensky, D., Hsu, M.-C., Schillinger, D., Evans, J. A.,
Aggarwal, A., Bazilevs, Y., Sacks, M. S., and Hughes,
T. J., 2015. “An immersogeometric variational framework
for fluid–structure interaction: Application to bioprosthetic
heart valves”. Computer methods in applied mechanics and
engineering, 284, pp. 1005–1053.
[18] Xu, F., Morganti, S., Zakerzadeh, R., Kamensky, D.,
Auricchio, F., Reali, A., Hughes, T. J., Sacks, M. S., and Hsu,
M.-C., 2018. “A framework for designing patient-specific
bioprosthetic heart valves using immersogeometric
fluid–structure interaction analysis”. International journal
for numerical methods in biomedical engineering, 34(4),
p. e2938.
[19] Catanho, M., Sinha, M., and Vijayan, V., 2012. “Model of
aortic blood flow using the windkessel effect”. University of
California of San Diago, San Diago.
[20] Westerhof, N., Lankhaar, J.-W., and Westerhof, B. E., 2009.
“The arterial windkessel”. Medical & biological engineering
& computing, 47(2), pp. 131–141.
[21] Xu, P., Liu, X., Zhang, H., Ghista, D., Zhang, D., Shi, C., and
Huang, W., 2018. “Assessment of boundary conditions for
cfd simulation in human carotid artery”. Biomechanics and
modeling in mechanobiology, pp. 1–17.
[22] Pant, S., Corsini, C., Baker, C., Hsia, T.-Y., Pennati, G.,
Vignon-Clementel, I. E., of Congenital Hearts Alliance
(MOCHA) Investigators, M., et al., 2016. “Data assimilation
and modelling of patient-specific single-ventricle physiology
with and without valve regurgitation”. Journal of
biomechanics, 49(11), pp. 2162–2173.
[23] Huang, H., Shu, Z., Song, B., Ji, L., and Zhu, N., 2019.
“Modeling left ventricular dynamics using a switched system
approach based on a modified atrioventricular piston unit”.
Medical engineering & physics, 63, pp. 42–49.
[24] Pironet, A., Dauby, P. C., Chase, J. G., Docherty, P. D.,
Revie, J. A., and Desaive, T., 2016. “Structural identifiability
analysis of a cardiovascular system model”. Medical
engineering & physics, 38(5), pp. 433–441.
[25] Duran-Hernandez, C., Ledesma-Alonso, R., Etcheverry, G.,
and Perez-Santiago, R., 2018. “Aerodynamic coefficient
calculation of a sphere using incompressible computational
fluid dynamics method”. Technology, Science, and Culture:
A Global Vision, pp. 105–112.
[26] Duran-Hernandez, C., Perez-Santiago, R., Etcheverry, G.,
and Ledesma-Alonso, R., 2019. “Modeling of a simplified 2D
cardiac valve by means of system identification”. In Mexican
Conference on Pattern Recognition, Springer, p. Accepted
Paper.
[27] Ljung, L., 1987. System identification: theory for the user.
Prentice-hall.

13 Copyright
c by ASME

You might also like