Sie sind auf Seite 1von 172

Mathematical Models and Simulation of the

Human Cardiovascular System

Alexandra Bugalho de Moura

CEMAT, Instituto Superior Técnico, Technical University of Lisbon, Portugal


and
Instituto Superior Politécnico de Tecnologias e Ciências, Luanda, Angola

ESSIM, Dresden, August 2012

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 1 / 110


Outline

Outline
Introduction
Blood flow in arteries
Local blood flow dynamics
Mathematical modeling of blood flow
Mathematical modeling of arterial wall dynamics
Mathematical modeling of blood-wall interaction
FSI numerical results
Global flow dynamics: boundary conditions on local models
Geometrical multiscale approach
The model reduction procedure
Blood pulse wave in large arteries: 1D models
Coupling the 3D FSI and 1D models
Blood pressure in compartments: 0D models
Coupling 3D and 0D models
Coupling 1D and 0D models
Vizualization: Hemodynamic indicators
References
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 2 / 110
Introduction

The cardiovascular system


I TASK: Suplying the organs with blood
I Mainly composed by: heart, arteries and veins

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 3 / 110


Introduction

The cardiovascular system

I TASK: Suplying the organs with blood


I Mainly composed by: heart, arteries and veins

I large (systemic) circulation: carries oxygenated blood to the various organs


I small (pulmonary) circulation: between the heart and lungs

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 3 / 110


Introduction

Complexity

I Mechanical
I Fluid (blood)
I Solid (heart and vessel walls)
I Biochemical (arterial and venous concentrations, transport and exchanges
with surrounding tissue)
I Regulation (chemoreflexes and baroreflexes - nervous system)
I Electrical (heart functioning)

I Diversity of function and geometry (from large arteries to capilaries)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 4 / 110


Introduction

Computational simulations
I Mathematical models in life sciences
I Efficient and accurate numerical
algorithms
I Increase of computer power

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 5 / 110


Introduction

Computational simulations
I Mathematical models in life sciences
better understanding of
I Efficient and accurate numerical =⇒ the circulatory system
algorithms
funcionality
I Increase of computer power

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 5 / 110


Introduction

Computational simulations
I Mathematical models in life sciences
better understanding of
I Efficient and accurate numerical =⇒ the circulatory system
algorithms
funcionality
I Increase of computer power

I provides quantitative data such as


I wall shear stress: influences the orientation and deformation of endothelial cells

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 5 / 110


Introduction

Cardiovascular diseases
I social-economic impact
I demand from medical doctors for scientifically rigorous and quantitative
investigations
I computational simulations are non-invasive and low cost

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 6 / 110


Introduction

Cardiovascular diseases: atherosclerosis


I inflamation and rigidity of arterial wall =⇒ stenosis: narrowing of the
I fat accumulation in the arterial wall vessel section

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 7 / 110


Introduction

Cardiovascular diseases: atherosclerosis


I angioplasty: widening narrowed arteries
I drug eluting stent

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 8 / 110


Introduction

Cardiovascular diseases: cerebral aneurysms


I saccular dilations of the artery wall
I usually occur near or in the Circle of Willis
I risk of rupture yielding dangerous
hemorhage

“It is estimated that 5% of the population has some type of aneurysm in the
brain. The incidence of ruptured aneurysm is approximately 10 out of 100,000
people per year. About 10% of patients who have one aneurysm will have at least
one more” National Library of Medicine, NIH US, http://www.nlm.nih.gov

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 9 / 110


Introduction

Computational hemodynamics Computational Fluid Mechanics - CFD

I Correlation between local hemodynamics and cardiovascular pathologies

I better understanding the on-rise and the development of (individual)


physiopathologies
I design of medical devices
I predict the outcome of surgical procedures

PREDICTIVE MEDICINE ⇒ HELP CLINICAL DECISION

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 10 / 110


Introduction

Patient-specific data

Maximum intensity projections


Rotational CTA
I voxel resolution of 0.8 mm size
I 5123 grid

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 11 / 110


Introduction

Medical image reconstruction

I image segmentation: e.g. constant threshold


I surface extraction: e.g. marching tetrahedra, radial basis functions (implicit
definition)
I surface smoothing: e.g. bi-Laplacian
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 12 / 110
Introduction

Sensitivity to medical image reconstruction (segmentation)

(A. Gambaruto)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 13 / 110


Introduction

Sensitivity to medical image reconstruction (smoothing)


300 it. bi-Laplacian 20 000 it. bi-Laplacian

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 14 / 110


Introduction

Mathematical models

Mesh generation

Problem setting

Geometry Numerical simulations


reconstruction

Image and medical


data acquisition

FEEDBACK Post-processing visualization

Patient’s real data

Clinical
interpretation

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 15 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?
Non-Newtonian behaviour: blood is not water
I Supension of cells ⇒ What mathematical models should be used?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?
Non-Newtonian behaviour: blood is not water
I Supension of cells ⇒ What mathematical models should be used?
Unsteady problem
I Pulsatile flow ⇒ Can we have real-time simulations?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?
Non-Newtonian behaviour: blood is not water
I Supension of cells ⇒ What mathematical models should be used?
Non steady problem
I Pulsatile flow ⇒ Can we have real-time simulations?
Wave propagation nature of blood flow in arteries
I Wall compliance ⇒ How can we model fluid-structure interaction (FSI)?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?
Non-Newtonian behaviour: blood is not water
I Supension of cells ⇒ What mathematical models should be used?
Non steady problem
I Pulsatile flow ⇒ Can we have real-time simulations?
Wave propagation nature of blood flow in arteries
I Wall compliance ⇒ How can we model fluid-structure interaction (FSI)?
Very complex vessel wall
I Wall of layers ⇒ What mathematical model should be used?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Blood flow in arteries

Blood flow in arteries


Simulations restricted to specific regions of interest
I Complex geometry ⇒ Are local simulations reliable?
Global circulation influences local hemodynamics
I Closed system ⇒ How to prescribe boundary conditions?
Non-Newtonian behaviour: blood is not water
I Supension of cells ⇒ What mathematical models should be used?
Non steady problem
I Pulsatile flow ⇒ Can we have real-time simulations?
Wave propagation nature of blood flow in arteries
I Wall compliance ⇒ How can we model fluid-structure interaction (FSI)?
Very complex vessel wall
I Wall of layers ⇒ What mathematical model should be used?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 16 / 110


Local blood flow dynamics

Describing the dynamics

What are the quantities that we use to descrive the flowing of a fluid?
(Eulerian coordinates)
A vecolity and pressure
B only velocity
C only pressure
D other

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 17 / 110


Local blood flow dynamics

Describing the dynamics

What are the quantities that we use to descrive the flowing of a fluid?
(Eulerian coordinates)
A vecolity and pressure
B only velocity
C only pressure
D other

What are the quantities that we use to descrive the motion of a structure?
(Lagrangian coordinates)
A displacement
B velocity
C displacement and velocity
D other

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 17 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The Navier-Stokes equations

∂u

 ρ
 + ρ (u · ∇) u− div σ(P, u) = 0
∂t


div u = 0

u → velocidade
P → pressão
σ(P, u) → Cauchy stress tensor

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 18 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The Navier-Stokes equations

∂u

 ρ
 + ρ (u · ∇) u− div (−PI + τ (u)) = 0
∂t


div u = 0

u → velocidade
P → pressão
Constitutive equations:
σ(P, u) = −PI + τ (u) τ (u) → extra stress tensor

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 18 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The Navier-Stokes equations

∂u

 ρ
 + ρ (u · ∇) u + ∇P− div τ (u) = 0
∂t


div u = 0

u → velocidade
P → pressão
Constitutive equations:
σ(P, u) = −PI + τ (u) τ (u) → extra stress tensor

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 18 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The Navier-Stokes equations

∂u

 ρ
 + ρ (u · ∇) u + ∇P−2µ div Du = 0
∂t


div u = 0

u → velocidade
P → pressão
Constitutive equations:
σ(P, u) = −PI + τ (u) τ (u) → extra stress tensor

Newtonian fluid:
1
τ (u) = 2µ ∇u + ∇T u = 2µDu

Du → strain rate tensor
2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 18 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The shear-thinning behaviour of blood

I The main mechanical properties of


blood are induced by the RBC.

I The aparent viscosity of blood


decreases with increasing shear rate

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 19 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Generalized Newtonian fluids


r
1
γ̇ = (∇u + ∇uT ) : (∇u + ∇uT )
2
τ = µ(γ̇)(∇u + ∇uT ) γ̇ → shear rate

µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇)

Carreau viscosity model:


0.7
F (γ̇) = (1 + (λγ̇)2 )(n−1)/2
0.6
Experimental data

0.5 Carreau
Cross
I Non-linear least squares fitting of
Viscosity (Poise)

0.4 Oldroyd
experimental data
0.3
(Prof. Kameneva, U. Pittsburgh)
0.2

0.1
I µ0 = 0.456 Poi
0
−3 −2 −1 0 1 2 3
I µ∞ = 0.032 Poi
10 10 10 10 10 10 10
Shear rate (s−1) I λ = 10.03 s
I n = 0.344

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 20 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Generalized Newtonian fluids 0.7

0.6
Experimental data

T
τ = µ(γ̇)(∇u + ∇u ) 0.5 Carreau
Cross

Viscosity (Poise)
0.4 Oldroyd

µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇) 0.3

0.2

0.1

0
−3 −2 −1 0 1 2 3
10 10 10 10 10 10 10
Shear rate (s−1)

Model Viscosity Model Model constants for blood


Power law k γ̇ n−1 k = 0.145, n = 0.628

µ0 = 0.456, µ∞ = 0.032
Carreau F (γ̇) = (1 + (λγ̇)2 )(n−1)/2
λ = 10.03s, n = 0.344

µ0 = 0.618, µ∞ = 0.034
Cross F (γ̇) = (1 + (λγ̇)m )−1
λ = 7.683s, m = 0.810

1 + log(1 + λγ̇) µ0 = 1.10, µ∞ = 0.035


Yeleswarapu F (γ̇) =
1 + λγ̇ λ = 45.23s

1 + (λ1 γ̇)2 µ0 = 0.426, µ∞ = µ0 λ21 λ−2


2
Oldroyd µ(γ̇) = µ0
1 + (λ2 γ̇)2 λ1 = 1.09s, λ2 = 3.349s

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 21 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The fluid equations

∂u


 ρ + ρ (u · ∇) u + ∇P−2 div(µ(γ̇)Du) = 0 in Ω


 ∂t


 div u = 0 in Ω




u = h on Γ1


 σ · n = −pn + τ · n = q on Γ2 , Γ3 , Γ4



 u = 0 on Γw




u(0, ·)

= u0 in Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 22 / 110


Local blood flow dynamics Mathematical modeling of blood flow

The Reynolds number

I non-dimensional Navier-Stokes equations

∂u 1
(
+ (u · ∇) u + ∇P − (2µ div Du) = 0 in Ω
∂t Re
div u = 0 in Ω

ρ|u|D
Re =
µ

I low Reynolds number ⇒ high viscosity (Stokes equations)


I high Reynolds number ⇒ turbulence

I Blood ⇒ Re ≈ 200 (laminar flow)

I Poiseuille flow (constant pressure gradient in pipes)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 23 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method:

f (tn + ∆t) − f (tn )


f 0 (tn ) = lim
∆t→0 ∆t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method:

f (tn+1 ) − f (tn )
f 0 (tn ) = lim
∆t→0 ∆t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (explicit):

f (tn+1 ) − f (tn )
f 0 (tn ) ≈
∆t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations


∂un+1 un+1 − un

∂t ∆t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations


∂un+1 un+1 − un

∂t ∆t

∂u

 ρ
 + ρ (u · ∇) u + ∇P − µ∆u = 0
∂t


div u = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations


∂un+1 un+1 − un

∂t ∆t

∂un+1

 ρ
 + ρ (un+1 · ∇) un+1 + ∇Pn+1 − µ∆un+1 = 0
∂t


div un+1 = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations


∂un+1 un+1 − un

∂t ∆t

un+1 − un 
 
 ρ
 + ρ (un+1 · ∇) un+1 + ∇Pn+1 − µ∆un+1 = 0
∆t

div un+1 = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: Euler’s method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .


I Euler’s method (implicit):

f (tn+1 ) − f (tn )
f 0 (tn+1 ) ≈
∆t

I Aplying to the Navier-Stokes equations


∂un+1 un+1 − un

∂t ∆t
 ρ ρ
 un+1 + ρ (un+1 · ∇) un+1 + ∇Pn+1 − µ∆un+1 = un
∆t ∆t


div un+1 = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 24 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Other time discretizations...

Backward difference formula (BDF) order 3:

∂un+1 11un+1 3un 3un−1 1un−2


≈ − + −
∂t 6∆t ∆t 2∆t 3∆t

  
11ρ 3un 3un−1 1un−2
un+1 + ρ (un+1 · ∇) un+1 + ∇Pn+1 − µ∆un+1 = ρ − +


6∆t ∆t 2∆t 3∆t



div un+1 = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 25 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Time discretization: non-linearities

Convective term:

 (un+1 · ∇) un+1 implicit
(un+1 · ∇) un+1 ≈ (un · ∇) un explicit
(un · ∇) un+1 semi-implicit

Extra-stress tensor (semi-implicit treatment):

τ k+1 = µ(γ̇ k+1 )(∇uk+1 + (∇uk+1 )T ) ≈ µ(γ̇ k )(∇uk+1 + (∇uk+1 )T ).

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 26 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

−∆u = f, em Ω
n
∂2u ∂2u
∆u = + 2 Ω
∂x 2 ∂y

y
x Γ

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 27 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

−∆u = f, em Ω n

y
x Γ
Multiplying the equation by an arbitrary function v (test function)

−∆u v = f v

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 27 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

−∆u = f, em Ω n

y
x Γ
Multiplying the equation by an arbitrary function v (test function) and integrating in Ω
Z Z
− ∆u vdΩ = f vdΩ
Ω Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 27 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

−∆u = f, em Ω n

y
x Γ
Multiplying the equation by an arbitrary function v (test function) and integrating in Ω
Z Z
− ∆u vdΩ = f vdΩ
Ω Ω

Integrating by parts
Z Z Z
− ∆u vdΩ = ∇u : ∇vdΩ − (∇u · n) vdΓ
Ω Ω Γ

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 27 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)


−∆u = f, em Ω n
u = 0, em Γ

y
x Γ

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 27 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

 Find u ∈ V such that


 −∆u = f, em Ω
⇐⇒ Z Z
u = 0, em Γ ∇u · ∇vdΩ = f vdΩ, ∀v ∈ V .

Ω Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 28 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)

 Find u ∈ V such that


 −∆u = f, em Ω
⇐⇒ Z Z
u = 0, em Γ ∇u · ∇vdΩ = f vdΩ, ∀v ∈ V .

Ω Ω

Consider Vh ⊂ V , such that dim Vh < ∞ (Galerkin approximation)


Z Z
Find uh ∈ Vh such that ∇uh · ∇vh dΩ = f vh dΩ, ∀vh ∈ Vh .
Ω Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 28 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)
dim V
dim Vh < ∞ ⇒ it is enough to consider the elements of the base Vh , {φi }i=1 h :
Find uh ∈ Vh such that
Z Z
∇uh · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
Ω Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 29 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)
If uh ∈ Vh , it can be written as linear combination of the elements of the base:
dim Vh
X
To find uh = uj φj such that
j=1

dim Vh
Z ! Z
X
∇ uj φj · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
Ω j=1 Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 29 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)
If uh ∈ Vh , it can be written as linear combination of the elements of the base:
dim Vh
X
To find uh = uj φj such that
j=1

dim Vh
Z ! Z
X
∇ uj φj · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
Ω j=1 Ω

dim Vh Z Z
X
uj ∇φj · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
j=1 Ω Ω
| {z } | {z }
Aij bi
dim Vh
X
uj Aij = bi , i = 1, . . . , dim Vh
j=1

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 29 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Method (FEM)


(a glance)
If uh ∈ Vh , it can be written as linear combination of the elements of the base:
dim Vh
X
To find uh = uj φj such that
j=1

dim Vh
Z ! Z
X
∇ uj φj · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
Ω j=1 Ω

dim Vh Z Z
X
uj ∇φj · ∇φi dΩ = f φi dΩ, i = 1, . . . , dim Vh .
j=1 Ω Ω
| {z } | {z }
Aij bi
dim Vh
X
uj Aij = bi , i = 1, . . . , dim Vh
j=1

To find U such that it is the solution of the linear system

AU = B, A = {Aij } , U = {uj } e B = {bi }


Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 29 / 110
Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Space (FEM)


(a glance)

Appropriate choice of the finite dimensional space!

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 30 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Space (FEM)


(a glance)
Appropriate choice of the finite dimensional space!

n I triangulation

Ω I parameter h linked to the size of the mesh


I convergence: lim ku − uh k = 0
h→0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 30 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Space (FEM)


(a glance)
Appropriate choice of the finite dimensional space!

n I triangulation

Ω I parameter h linked to the size of the mesh


I convergence: lim ku − uh k = 0
h→0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 30 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: Finite Element Space (FEM)


(a glance)
Appropriate choice of the finite dimensional space!

n I triangulation

Ω I parameter h linked to the size of the mesh


I convergence: lim ku − uh k = 0
h→0

Γ
n

I Each function vh ∈ Vh is characterized by its value on the


nodes of the mesh Ni

0, i=6 j
φj : such that φj (Ni ) =
1, i =j
, j = 1, . . . , dim Vh . Ω
dim Vh = number of internal nodes of the mesh. Γ
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 30 / 110
Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: functional setting

 Z 
p
Lp (Ω) = f : Ω −→ R : |f (x)| dω < ∞ , 1 6 p < ∞,

and
L∞ = {f : Ω −→ R : ∃C > 0 s.t. |f (x)| 6 C a.e. in Ω}

Banach space:
Z 1/p
p
kf kLp (Ω) := kf kΩ,p = |f (x)| dω

kf kL∞ (Ω) := kf kΩ,∞ = inf {M : |f (x)| 6 M a.e. in Ω}

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 31 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: functional setting


Z
L2 (Ω) is an Hilbert space with inner product: (f , g )L2 (Ω) = f (x)g (x) dω

 Z 
L20 := q ∈ L2 (Ω) : q(x) dω = 0 → isomorfic to L2 (Ω)/R

I Sobolev spaces

W k,p = {f ∈ Lp (Ω) : D α f ∈ Lp (Ω), |α| 6 k} ,


 1/p
X
kf kW k,p (Ω) =  kD α f kpLp (Ω)  , 1 6 p < ∞,
|α|6k

kf kW k,∞ (Ω) = max kD α f kL∞ (Ω) .


|α|6k

I Hilbert space: W k,2 (Ω) := H k (Ω)


I Hγ1 (Ω) ⊂ H 1 (Ω): functions with zero trace on γ ⊂ ∂Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 31 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations
I momentum equation

∂u


 + u · ∇u + ∇p − ν∆u = 0
 ∂t



 u = 0, on γ
−pn + ∇u · n = h, on ∂Ω \ γ

P µ
p= ν=
ρ ρ

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations

I momentum equation
v ∈ Hγ1 (Ω)
Z  
∂u
+ u · ∇u + ∇p − ν∆u ·v = 0
Ω ∂t

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations

I momentum equation
v ∈ Hγ1 (Ω)
Z  
∂u
+ u · ∇u + ∇p − ν∆u ·v = 0
Ω ∂t

Z Z Z
∇·v = pv · n − pdivv,
Ω ∂Ω Ω
Z Z Z
ν ∆u · v = ν (∇u · n) · v − ν ∇u : ∇v.
Ω ∂Ω Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations

I momentum equation
v ∈ Hγ1 (Ω)
Z  
∂u
+ u · ∇u + ∇p − ν∆u ·v = 0
Ω ∂t

Z Z Z Z Z
∂u
·v+ ((u · ∇) · u) · v − pdivv + ν ∇u : ∇v = (ν∇u · n − pn) · v
Ω ∂t Ω Ω Ω ∂Ω

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations
I momentum equation
v ∈ Hγ1 (Ω)
Z  
∂u
+ u · ∇u + ∇p − ν∆u ·v = 0
Ω ∂t

Z Z Z Z Z
∂u
·v+ ((u · ∇) · u) · v − pdivv + ν ∇u : ∇v = v·h
Ω ∂t Ω Ω Ω ΓN

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations
I momentum equation
v ∈ Hγ1 (Ω)
Z  
∂u
+ u · ∇u + ∇p − ν∆u ·v = 0
Ω ∂t

Z Z Z Z Z
∂u
·v+ ((u · ∇) · u) · v − pdivv + ν ∇u : ∇v = v·h
Ω ∂t Ω Ω Ω ΓN

I continuity equation
q ∈ L20 (Ω)
Z
div u = 0 ⇒ divu q = 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 32 / 110


Local blood flow dynamics Mathematical modeling of blood flow

Spacial discretization: variational formulation of the


Navier-Stokes eqations
Find, ∀t > 0, u ∈ Hγ1 (Ω) and p ∈ L20 (Ω) s.t.
   Z
∂u
, v + a(u, v) + c(u, u, v) + b(v, p) = v · h, ∀v ∈ Hγ1 (Ω),


∂t

Γ n


∀q ∈ L20 (Ω),

b(u, q) = 0,

with
Z Z Z
a(u, v) = ν ∇u : ∇v, c(w, u, v) = (w·∇)u · v, b(v, p) = − p divv
Ω Ω Ω

I the bilibear (a, c) and trilinear (b) forms are linear


I a is eliptic: ∃C > 0 s.t. |a(u, u)| > C kuk2H 1 (Ω)
γ

b(v, q)
I inf-sup condition: ∃β > 0 s.t. inf sup >β
q∈L20 (Ω) v∈H 1 (Ω)N
0
|v|H 1 (Ω)N kqkL2 (Ω)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 33 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

Structure equations: 3D elasticity


Elastic material (St Venant-Kirchhoff material) in large displacements
Lagrangian formulation

0
η
0 P(η) = (I + ∇0 η)S(η)
Γs,in Ωs0 Γs,a
z
S(η) = λ tr(E)I + 2µE
0
Γs,w
1 
0
Γs,ext E = ∇0 η + ∇T0 η + ∇T0 η∇0 η
2

Eξ E
Lamé constants: λ = , µ=
(1 + ξ)(1 − 2ξ) 2(1 + ξ)
E → Young modulus (elasticity)
ξ → Poisson ratio (ξ = 0.5 → incompressible material)

∂2η
ρw − div0 (P) = 0 on Ω0s
∂t 2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 34 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

The 3D FSI model: structure equations


Elastic material in large displacement, Lagrangian formulation

0
η
0 P(η) = (I + ∇0 η)S(η)
Γs,in Ωs0 Γs,a
z
S(η) = λ tr(E)I + 2µE
0
Γs,w
1 
0
Γs,ext E = ∇0 η + ∇T0 η + ∇T0 η∇0 η
2

∂2η


 ρ w − div0 (P) = 0 on Ω0s



 ∂t 2


P · n0 Γ0s,ext



 = 0 on
 (P · n ) · τ = 0, η · n Γ0s,out

0 a a = 0 on
 η = 0 on Γ0s,in
Γ0w




 interface conds. on



in Ω0s



 η(0, ·) = η0
in Ω0s

∂t η(0, ·) = η̇ 0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 35 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

Other structural models

simple models: S(ηr ) = Φ


br

pressure-area algebraic relation −→ S(ηr ) :=


Eh
R02 (1 − ξ 2 )
ηr = bηr

∂ 2 ηr
independent rings model −→ S(ηr ) := ρw h
∂t 2
+ bηr

∂ 2 ηr ∂ 2 ηr
generalized strings model −→ S(ηr ) := ρw h
∂t 2
− KGh 2 + bηr
∂z

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 36 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

Time discretization: Newmark method

I I =]0, T ]. N points: 0 = t0 < t1 < . . . < tN−1 < tN = T , ∆t = ti+1 − ti .

 n+1
 η̇ = η̇ n + δt (1 − γ) η̈ n + δt γ η̈ n+1 ,

δt 2
 η n+1 = η n + δt η̇ n + (1 − 2β) η̈ n + δt 2 β η̈ n+1 ,

2
1 1
I mid point scheme (second order accurate): γ = ,β =
2 4

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 37 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

Spacial discretization: FEM

I Find η(t) ∈ H 1 (Ω0s ), such that η(0) = η 0 , and


∂2η
Z Z Z
ρw 2
ξ dω + P(η) : ∇ξ dω = Φξ
b dγ, ∀ξ ∈ HΓ1in (Ω0s )
Ω0s ∂t Ω0s Γ0w

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 38 / 110


Local blood flow dynamics Mathematical modeling of arterial wall dynamics

Spacial discretization: FEM

I Find η(t) ∈ H 1 (Ω0s ), such that η(0) = η 0 , and


∂2η
Z Z Z
ρw 2
ξ dω + P(η) : ∇ξ dω = Φξ
b dγ, ∀ξ ∈ HΓ1in (Ω0s )
Ω0s ∂t Ω0s Γ0w

Vh (Ω0s,h ) ⊂ H 1 (Ω0s ) Ω0s,h = Thel,0 , Thel,0 a partition of Ω0s

I Find, ∀t ∈ I , η h (t) ∈ Vh (Ω0s ), such that η h (0) = η 0,h , and,


∀ξ h ∈ Vh (Ω0s ) ∩ HΓ1t (Ω0s ),
in

∂ 2 ηh
Z Z Z
ρw ξ dω + P(η h ) : ∇ξ h dω = Φ
b h ξ dγ,
Ω0s ∂t 2 h Ω0s Γ0w
h

η 0,h and Φ
b h suitable approximations of the initial and forcing terms.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 38 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

A glance to the fluid-structure interaction (FSI) model


I boundary conditions at the physical interface - the wall
I pressure wave

FLUID STRUCTURE
INTERACTION
⇐⇒
displacement (new domain)

Interface conditions

normal stresses

implicit coupling
(iterative procedure)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 39 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

ALE (arbitrary Lagrangian-Eulerian) coordinates


I We would like to have Lagrangian coordinates on the structural wall, and
Eulerian coordinates in the remaining domain

At : Ω0 → Ωt , At (x̂) = x(t, x̂), x̂ ∈ Ω0 ,


retrieving the desired computational domain as Ωt = At (Ω0 )
I Harmonic extension

4At in Ωt ,

 = 0,
At on ∂Γtw ,

= η,
 At · n = 0, ∂At
·τ on Γtin ∪ Γtout ,

= 0,
∂n

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 40 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: the fluid equations

generalized Newtonian Fluid, ALE formulation

∆At in Ω0

η Ωt  = 0,
t
t Γa
Γin z
interface conds. on Γ0w

Ω0 Γw0
∂At
Γwt w= , Ωt = At (Ω0 )
∂t

∂u

 ρ + ρ (u−w · ∇)u + ∇P − 2 div(µ(γ̇)D(u)) = 0 in Ωt
∂t x̂



div u = 0 in Ωt




interface cond. on Γtw

∀t ∈ (0, T ], with initial cond. u = u0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 41 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: strutural equations


Elastic material in large displacements, Lagrangian coordinates

0
η
0
P(η) = (I + ∇0 η)S(η)
Γs,in Ωs0 Γs,a
z
S(η) = λ tr(E)I + 2µE
0
Γs,w
1
∇0 η + ∇T T

0 η + ∇0 η∇0 η
0
Γs,ext E =
2

∂2η


 ρw − div0 (P) = 0 on Ω0s
∂t 2

interface cond. on Γ0w

∀t ∈ (0, T ], with proper initial conditions η and ∂t η

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 42 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: fluid-structure interface conditions


η

Γw0
n0
Γwt

We set At = η on Γ0w , and


I continuity of the velocity
∂η
u ◦ At = , on Γtw
∂t
I equilibrium of the normal stresses
 
− (det ∇0 η) σ(u, p) ∇0 −T η · n0 = σ s (η) · ns , on Γ0w

with σ(u, p) = −pI + 2µ(γ̇)D(u).

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 43 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: mathematical analysis

Is the FSI problem well posed?


I Does the solution exists?
I Is it unique?
I Does it depend continuously on the data?

Open problem!!
[D. Coutand, Y. Maday, C. Grandmont, B. Desjardens, M. Esteban, H. Beirão da Veiga, ...]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 44 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The 3D FSI model: energy estimate


[Formaggia, Moura, Nobile, M2AN,2007]

ρ ρw λ
E3D = ||u||2L2 (Ωt ) + ||∂t η||2L2 (Ω0s ) + µ||E||2L2 (Ω0s ) + || tr(E)||2L2 (Ω0s )
2 2 2

E3D (t) + 2ν||D(u)||2L2 (Ωt ) 6 E3D (0)

I non-homogeneous natural conditions at inflow (r) and outflow (q)

t t t
Z  Z Z  2ν t
2 2 2
E3D−AL (t) + ν kD(u)kL2 (Ωt ) dt 6 E3D−AL (0) + C
e krkL2 (Γt ) dt + C kqkL2 (Γt ) dt e ρ
0 0 in 0 a

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 45 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The 3D model: other structural models

simple models: S(ηr ) = Φ


br

pressure-area algebraic relation −→ S(ηr ) :=


Eh
R02 (1 − ξ 2 )
ηr = bηr

ρ
E3D−AL (t) := 2
kuk2L2 (Ωt ) + b2 kηr k2L2 (Γ0 )
w

∂ 2 ηr
independent rings model −→ S(ηr ) := ρ̃w
∂t 2
+ bηr

∂ηr 2

ρ ρ̃w
E3D−IR (t) := 2
kuk2L2 (Ωt ) + 2 ∂t 2 + b2 kηr k2L2 (Γ0 )
L (Γ0w ) w

∂ 2 ηr ∂ 2 ηr
generalized strings model −→ S(ηr ) := ρ̃w
∂t 2
− a
∂z 2
+ bηr

∂ηr 2 ∂ηr 2

ρ ρ̃w
E3D−GS (t) := 2
kuk2L2 (Ωt ) + 2 ∂t 2 + a
2 ∂z 2 + b2 kηr k2L2 (Γ0 )
L (Γ0w ) L (Γ0w ) w

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 46 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: energy estimate


For bounded viscosity laws of the general form

µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇),


where F (·) is a continuous monotonic function such that limγ̇→0 F (γ̇) = 1 and limγ̇→+∞ F (γ̇) = 0,
the 3D FSI coupled problem with homogeneous Dirichlet boundary conditions u = 0 on Γtin and Γtout , satisfies
the energy inequality
d
(E (t)) + 2µ∞ kD(u)k2L2 (Ωt ) 6 0
dt
from which we obtain the energy decay property:
Z t
E (t) + 2µ∞ kD(u)k2L2 (Ωt ) dt 6 E (0)
0

where E (0) is a constant depending only on the initial data u0 , η 0 and η̇ 0 .

I dissipation only comes from the fluid viscous effects


I the power exchanged by the fluid and the structure balance at the interface

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 47 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: numerical approximation

Time discretization
Fluid: Implicit Euler’s method or BDF
Structure: Newmark method (mid point)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 48 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: numerical approximation

Time discretization
Fluid: Implicit Euler’s method or BDF
Structure: Newmark method (mid point)

Spacial discretization
FEM (variational formulation)
I Non-conforming fluid and solid discrete spaces
I Dirichlet-Neumann coupling (classic)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 48 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: numerical approximation

Time discretization
Fluid: Implicit Euler’s method or BDF
Structure: Newmark method (mid point)

Spacial discretization
FEM (variational formulation)
I Non-conforming fluid and solid discrete spaces
I Dirichlet-Neumann coupling (classic)

The FSI problem is highly non-linear


I appropriate (time) discretization of the coupling conditions
I implicit coupling

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 48 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: the added-mass effect


I numerical instabilities under certain combinations of physiological parameters:

I for a given geometry, as soon as the density of the structure is lower then a
certain threshold
I for a given structure density, as soon as the length of the domain is greater
then a certain threshold

I Added-mass: inertia added to a system because an accelerating or


decelerating body must move some volume of surrounding fluid as it moves
through (incompressibility constraint)

I Instabilities are found when:

ρw h
<1
ρλadd
L
λadd = πR
 : increases with the length of the domain (slender domains)
πth L

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 49 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: Blood

I incompressible
I arteries are slender
I blood density: ρ ≈ 1, 03 g /cm3
I wall density: ρw ≈ 1, 2 g /cm3

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 50 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: coupling algorithms


I FSI coupling scheme defined through the time discretization of the interface
conditions

I Explicit (loosely or weakly) coupled methods:

ηfn+1 = ηsn
un+1
f = uns
σsn+1 ns = −σfn+1 nf

(at the interface wall)

1 update fluid domain configuration (mesh) and velocities


2 advance fluid in time with Dirichlet condition (velocities)
3 advance solid in time with interface Neumann condition (stresses)
4 go to next time-step

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 51 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: coupling algorithms


I Implicit (strongly) coupled methods:

ηfn+1 = ηsn+1
un+1
f = un+1
s
σsn+1 ns = −σfn+1 nf

(at the interface wall)

1 solve the coupled FSI problem


2 go to next time-step

I monolithic: single solver with the purpose of simultaneously solve fluid and
structure
I partitioned: couples independent fluid and structure solvers

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 52 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: relaxed fixed-point algorithm

I FSI coupling can be stated as a fixed point algorithm, in the form

S ◦ F(η) = η

I interface Dirichlet-Neumann formulation

1 Initialize η 0
2 For k > 0 untill convergence (kη k+1 − η k k < tol)
i solve fluid (including domain update): uk = F(η k )
ii solve solid: η̃ k+1 = S(uk )
iii η k+1 = ωk η̃ k+1 + (1 − ωk )η k

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 53 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: Newton’s method

[Fernandez, Moubachir, 2005]

residual: R(η) = η − S ◦ F(η)

1 initial guess η 0
2 Do, until convergence (kη k+1 − η k k < tol),
i. initial evaluation: η 1 = η 0 − S ◦ F (η 0 ) = R(η 0 ) (solve fluid,including domain
update, and solid, and evaluate residual)
ii. for k > 1: Dη (R(η k ))(η k+1 − η k ) = −R(η k )

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 54 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: Newton’s method


[Fernandez, Moubachir, 2005]

residual: R(η) = η − S ◦ F(η)

1 initial guess η 0
2 Do, until convergence (kη k+1 − η k k < tol),
i. initial evaluation: η 1 = η 0 − S ◦ F (η 0 ) = R(η 0 ) (solve fluid,including domain
update, and solid, and evaluate residual)
ii. for k > 1: Dη (R(η k ))(η k+1 − η k ) = −R(η k )

I Dη (R(η k ))(η k+1 − η k ): Frechet derivative of R computed in η k and applied


to the increment η k+1 − η k (tangent problem).

Dη (R(η k ))z = z − Du,p SDη Fz,

I y = Dη Fz: fluid tangent problem


I δz = Du,p Sy: solid tangent problem

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 54 / 110


Local blood flow dynamics Mathematical modeling of blood-wall interaction

The FSI model: Newton’s method


[Fernandez, Moubachir, 2005]

residual: R(η) = η − S ◦ F(η)


1 initial guess η 0
2 Do, until convergence (kη k+1 − η k k < tol),
i. initial evaluation: η 1 = η 0 − S ◦ F (η 0 ) = R(η 0 ) (solve fluid,including domain
update, and solid, and evaluate residual)
ii. Do, until convergence (kδzk < tol),
a. solve fluid tangent problem: y = Dη F z
b. solve solid tangent problem: δz = Du,p Sy
c. compute: Dη (R(η k ))z = z − δz.

I Dη (R(η k ))(η k+1 − η k ): Frechet derivative of R computed in η k and applied


to the increment η k+1 − η k (tangent problem).

Dη (R(η k ))z = z − Du,p SDη Fz,


I y = Dη Fz: fluid tangent problem
I δz = Du,p Sy: solid tangent problem
I quasi-Newton: simplification of the tangent problem
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 54 / 110
Local blood flow dynamics FSI numerical results

Numerical results (FSI model)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 55 / 110


Global flow dynamics: boundary conditions on local models

Boundary conditions

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 56 / 110


Global flow dynamics: boundary conditions on local models Geometrical multiscale approach

The geometrical multiscale approach


I simulation of large parts of the
arterial tree by using different I allows to study the influence of
models, with different scales and global circulation on the local one
level of accuracy, coupled together

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 57 / 110


Global flow dynamics: boundary conditions on local models Geometrical multiscale approach

The mathematical complexity

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 58 / 110


Global flow dynamics: boundary conditions on local models Geometrical multiscale approach

The mathematical complexity

I 3D ⇒ Navier-Stokes equations or
non-Newtonian rheology, possibly
coupled with a structural model for the
vessel wall.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 58 / 110


Global flow dynamics: boundary conditions on local models Geometrical multiscale approach

The mathematical complexity

I 3D ⇒ Navier-Stokes equations or
non-Newtonian rheology, possibly
coupled with a structural model for the
vessel wall.

I 1D ⇒ A net of systems of hyperbolic


equations describing evolution of mean
pressure and flow rate.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 58 / 110


Global flow dynamics: boundary conditions on local models Geometrical multiscale approach

The mathematical complexity

I 3D ⇒ Navier-Stokes equations or
non-Newtonian rheology, possibly
coupled with a structural model for the
vessel wall.

I 1D ⇒ A net of systems of hyperbolic


equations describing evolution of mean
pressure and flow rate.

I 0D ⇒ System of algebro-ordinary
differential equations, often non linear,
describing mean pressure and flow rate
in time.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 58 / 110


Global flow dynamics: boundary conditions on local models The model reduction procedure

The model reduction procedure

t
Γs,in η Ωt
Ω0s
t
Γin R0 z

Γwt

Fluid-structure interaction problem in a cylindrical-type


geometry representing an artery

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 59 / 110


Global flow dynamics: boundary conditions on local models The model reduction procedure

The model reduction procedure

Γa

t
Γs,in
Ωt
t
Γin z

Γwt

Decompose the geometry into two parts

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 59 / 110


Global flow dynamics: boundary conditions on local models The model reduction procedure

The model reduction procedure

Γa

t
Γs,in
Ωt
t
Γin Ω1D z

Γwt

Devise an appropriate reduced model

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 59 / 110


Global flow dynamics: boundary conditions on local models The model reduction procedure

The model reduction procedure

Ω1D
Ω 3D ?

Find appropriate coupling conditions and numerical strategies

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 59 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: derivation

Derived from the 3D FSI model by

a b z

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 60 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: derivation

Derived from the 3D FSI model by


I Making some simplifying assumptions:
I Cylindrical geometry
I Simplified structural models
I Neglect wall inertia

z
a b

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 60 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: derivation


Derived from the 3D FSI model by
I Making some simplifying assumptions:
I Cylindrical geometry
I Simplified structural models
I Neglect wall inertia
I Integrating over the cross section Σ(z)

Σ (z)

a b z

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 60 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: derivation


Derived from the 3D FSI model by
I Making some simplifying assumptions:
I Cylindrical geometry
I Simplified structural models
I Neglect wall inertia
I Integrating over the cross section Σ(z) : averaged quantities
Z
A(z, t) = dγ
Ωt ∩Σ(z)
Z
Q(z, t) = uz (x, t)dγ
Ωt ∩Σ(z)
Z
1
p(z, t) = p(x, t)dγ
|Σ(z)| Ωt ∩Σ(z)

Ω1D z
a b

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 60 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: the equations

z=a z=b z

∂A ∂Q


 + = 0, z ∈ (a, b), t > 0

 ∂t ∂z

∂ Q2
 
 ∂Q A ∂p Q
−Kr , z ∈ (a, b), t > 0


 + + =
∂t ∂z A ρ ∂z A

endowed with a pressure-area algebraic relation:


∂ψ
p(t, z) − pext = ψ(A(t, z); A0 (z), β(z)), with > 0 and ψ(A0 ; A0 , β).
∂A
√ √ √
A − A0 πh0 E
ψ(A; A0 , β) = β with β =
A0 1 − ν2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 61 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D model: main properties


I describe the wave propagation nature of blood flow
I coupled with the 3D-FSI model can act as absorbing boundary condition.

Full hyperbolic system of equations:


A ∂p
I characteristic speeds: λ1,2 = u ± c, with u = Q/A and c 2 =
ρ ∂A
Z A
c(s)
I characteristic variables: W1,2 = u ± ds = W1,2 (A, Q)
A0 s

In heamodynamics c >> u and the flow is subcritical


(c of the order of 1 − 10 m/s, u of the order of 10−3 − 10−2 m/s)

I An absorbing boundary condition on the


outflow (right) extremity of the 1D model
is obtained imposing W2 (A, Q) = 0.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 62 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

A network of 1D models
I simulation of complex arterial trees by coupling 1D models

I domain decomposition
approach

Stable coupling conditions [L.Formaggia, D. Lamponi, A. Quar-


teroni (2003)]:
n
X
I continuity of the flux: Qi = 0
i=1
I continuity of the total pressure:
ρ 2 ρ 2
p i + |u i | = p j + |u j | , 1 6 i, j 6 n
2 2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 63 / 110


Global flow dynamics: boundary conditions on local models Blood pulse wave in large arteries: 1D models

The 1D Model: energy estimate

z=a z=b z

W2 (t) = g2 (t) at z = b (non reflecting condition).

Z b Z b Z A
ρ
E1D (t) = Au 2 dz + Ψ(A)dz, ∀t > 0, with Ψ(A) = ψ(τ )dτ.
2 a a A0

For regular and small enough data, there exists a positive bounded function G s.t.

d  ρ 
E1D (t) 6 G (g2 (t))+Q(a, t) p(a) + u 2 (a)
dt 2
[Formaggia, Gerbeau, Nobile, Quarteroni, CMAME, 2001]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 64 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Coupling 3D FSI and 1D models


r
η
Γw0

Γin Ω Γa z

z=0 z=a z=b

At the interface between the 1D and the 3D FSI models we impose


I Continuity of the flux Z
u · ndγ = Q(a)
Γa
I Continuity of the normal stresses
Z
1 ρ
(p − 2µD(u)) · ndγ = p(a) + u(a)2
|Γa | Γa 2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 65 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Coupling 3D FSI and 1D models (flow rate problem)


Z
I continuity of the flux : u · ndγ = Q1D
Γ3D
Z
1 ρ 2
I continuity of normal stress : (p − 2µD(u)) · ndγ = p 1D + u 1D
|Γ3D | Γ3D 2

flow rate

mean pressure

Treatment of the defective boundary conditions


flow rate problem: Lagrange multiplier approach
[Formaggia, Gerbeau, Nobile, Quarteroni, SIAM J Num Anal, 2002]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 66 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Coupling 3D FSI and 1D models (mean pressure problem)


Z
I continuity of the flux : u · ndγ = Q1D
Γ3D
Z
1 ρ 2
I continuity of total stress : (p − 2µD(u)) · ndγ = p 1D + u 1D
|Γ3D | Γ3D 2

mean pressure

flow rate

Treatment of the defective boundary conditions


mean pressure problem: P tot n − 2µD(u) · n = P tot
[Heywood, Rannacher, Turek, Int. J. Num. Meth. Fl., 1996]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 67 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

The 3D-1D coupling: energy estimate


[Moura, Janela, Sequeira, JCAM, 2010; Formaggia, Moura, Nobile, 2007]

I F (·) continuous monotonic


µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇) I limγ̇→0 F (γ̇) = 1
I limγ̇→+∞ F (γ̇) = 0

I Dirichlet bound. cond. u = 0 on Γtin and homogeneous cond. on the structure


I apropriate formulation of the Navier-Stokes equations
Z
d  ρ 
(E (t)) + 2µ∞ kD(u)k2L2 (Ωt ) 6 − pn + |u|2 n − 2µ(γ̇)D(u) · n · u dγ
dt Γtout 2

with
ρ ρw λ
E3D = ||u||2L2 (Ωt ) + ||∂t η||2L2 (Ω0s ) + µ||E||2L2 (Ω0s ) + || tr(E)||2L2 (Ω0s ) ,
2 2 2

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 68 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

The 3D-1D coupling: energy estimate


[Moura, Janela, Sequeira, JCAM, 2010; Formaggia, Moura, Nobile, 2007]

I F (·) continuous monotonic


µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇) I limγ̇→0 F (γ̇) = 1
I limγ̇→+∞ F (γ̇) = 0

I Dirichlet bound. cond. u = 0 on Γtin and homogeneous cond. on the structure


I apropriate formulation of the Navier-Stokes equations
Z
d  ρ 
(E (t)) + 2µ∞ kD(u)k2L2 (Ωt ) 6 − pn + |u|2 n − 2µ(γ̇)D(u) · n · u dγ
dt Γtout 2

d
E1D (t) 6 G (g2 (t))+Q(a, t) p(a) + ρ 2

dt 2 u (a)

with
Z b Z b Z A
ρ
E1D (t) = Au 2 dz + Ψ(A)dz, ∀t > 0, with Ψ(A) = ψ(τ )dτ.
2 a a A0

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 68 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

The 3D-1D coupling: energy estimate


[Moura, Janela, Sequeira, JCAM, 2010; Formaggia, Moura, Nobile, 2007]

I F (·) continuous monotonic


µ(γ̇) = µ∞ + (µ0 − µ∞ )F (γ̇) I limγ̇→0 F (γ̇) = 1
I limγ̇→+∞ F (γ̇) = 0

I Dirichlet bound. cond. u = 0 on Γtin and homogeneous cond. on the structure


I apropriate formulation of the Navier-Stokes equations
Z
d  ρ 
(E (t)) + 2µ∞ kD(u)k2L2 (Ωt ) 6 − pn + |u|2 n − 2µ(γ̇)D(u) · n · u dγ
dt Γtout 2

d
E1D (t) 6 G (g2 (t))+Q(a, t) p(a) + ρ 2

dt 2 u (a)

d d
E3D (t) + 2µ∞ ||D(u)||2L2 (Ωt ) + E1D (t) 6 G (g2 (t))
dt dt
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 68 / 110
Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Coupling 3D and 1D FSI models (displacement)

Continuity of the sec-


Continuity of the normal stress
Continuity of the flux ; tion area at the inter-
face

Which boundary
Γout
w conditions for the
structure model?

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 69 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

area ( P = P (A) )

mean pressure

flow rate

I At the interface:
η
I assume only radial movements
R0
I use the pressure-area algebraic ^
R − R0
c. η= (x̂ −ĉ)
relation of the 1D model: R0
√ 2


A0
A= A0 + p
β

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 70 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

area ( P = P (A) )

mean pressure

flow rate

NOTE: Forcing the continuity of the area


through a Dirichlet boundary condition on ; Energy estimate
the structure model

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 70 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

area ( P = P (A) )

mean pressure

flow rate

However, homogeneous Neumann conditions on the structure at the interface give


rise to a stable coupling

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 70 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

area ( P = P (A) )

mean pressure

flow rate

Numerical evidences: as long as an implicit scheme is applied, the coupling with


the imposition of the area is stable. Explicit schemes show instabilities when area
continuity is enforced.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 70 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

area ( P = P (A) )

mean pressure

flow rate

A last note: Simple structural models (like the independent rings or algebraic
relations) do not enforce continuity of the area (they do not need boundary
conditions!).

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 70 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Solving the 3D-1D FSI coupling: explicit coupling

t n+1 3D 1D
)
stress
rma l total
flux (no

normal total stress (flux)


tn 3D 1D

s)
l tota l stres
orma
t n−1 flux (n
3D 1D

n−1 (m)
I given φ3D , while tn−1 < tn , solve the 1D model

I given φn1D solve the 3D problem

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 71 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Solving the 3D-1D FSI coupling: implicit coupling


ess)
t n+1 to tal str
3D o rmal 1D
flux (n

normal total stress (flux)

tn 3D 1D

relaxed fux (relaxed normal total stress)

t n−1 )
stress
3D l total 1D
orma
flux (n

I Set φ(0) = φn−1 and φ(0) = φn−1 , and for k = 1, . . .:


3D 3D 1D 1D

. set φ1D
(k−1)
= ωφ3D
(k−1)
+ (1 − ω)φ1D
(k−1)

. given φ3D
(k−1)
solve 1D →φ(k)
1D

. (k)
given φ1D solve 3D FSI →φ(k)
3D

. verify convergence: kφ3D − φ1D k 6 tol


(k) (k)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 72 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Absorbing boundary conditions


Not coupled Coupled

WTHOUT continuity of the area


EXPLICIT coupling
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 73 / 110
Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Comparing the area with a reference solution (explicit coupling)

algebraic model

independent rings model

linear elasticity model

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 74 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Comparing the area with a reference solution (implicit coupling)

algebraic model

independent rings model

linear elasticity model

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 75 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Forcing the continuity of the area

explicit coupling implicit coupling

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 76 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Area mismatch at the coupling interface

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 77 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Numerical results: generalized Newtonian fluids


Carreau fluid

not
coupled Carreau fluid Newtonian fluid

explicit
coup.
(without
cont. area)

implicit
coup. (with
cont. area)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 78 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

A real geometry: 3D-1D coupling


One dimensional models enable proper transmission of pulse waves avoiding
spurious reflections.
Stand alone 3D FSI (clamped ends) Coupled with a 1D tube at each outflow

I Pressure impulse of 10mmHg at inflow

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 79 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Absorbing boundary conditions

Remember: at the 1D outflow boundary we


impose the absorbing condition:

W2 (Q, p̄) = 0 ⇔
s  2
8β A0 p
p̄ + A0 ·
ρA0 β
s !
A0 p 1/4
p̄ + A0 − A0 =Q
β
Displacement and Pressure
for the 3D-1D coupling and
for the standard traction-free
conditions.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 80 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Simplified absorbing boundary conditions

3D - 1D Coupling:
[+] Allows to integrate 3D (FSI) models into lower order (1D) models that can
represent large parts of the vascular system.
[+] Effectively acts as physiological absorbing boundary condition, partially
filtrating spurious pressure wave reflections.
[–] If the 1D hyperbolic problem is solved explicitly, a CFL condition imposes a
time step much smaller then the one required by the 3D FSI algorithm.
[–] May be nontrivial to implement in many widely used commercial CFD codes.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 81 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Simplified absorbing boundary conditions

3D - 1D Coupling:
[+] Allows to integrate 3D (FSI) models into lower order (1D) models that can
represent large parts of the vascular system.
[+] Effectively acts as physiological absorbing boundary condition, partially
filtrating spurious pressure wave reflections.
[–] If the 1D hyperbolic problem is solved explicitly, a CFL condition imposes a
time step much smaller then the one required by the 3D FSI algorithm.
[–] May be nontrivial to implement in many widely used commercial CFD codes.

Impose condition W2 (Q, p̄) = 0 directly on the 3D FSI model


[Nobile, Vergara, 2008] , [Janela, Moura, Sequeira, IJES, 2010]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 81 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Linear Absorbing Conditions (LAC) (explicit treatment)


[Janela, Moura, Sequeira, IJES, 2010]

I For n > 0 the (constant) pressure p̄ (n) can be set at the outflow sections by
s s !

 p 2
(n) A0 A0 1/4
p
p̄ + A0 p̄ (n) + A0 − A0 = Q (n−1)
ρA0 β β

I Using a Taylor expansion, a first order approximation of p̄ (n) can be given by



ρβ
p̄ (n) ≈ √ 5/4 Q (n−1) (1)
2A0
I This is conceptually equivalent to couple the 3D FSI model with a particular
RL circuit, capable of absorbing pressure waves.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 82 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Linear absorbing conditions

I For physiological values of the the Young modulus E , depending on the flow
rates, the linear approximations given by (1) are accurate

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 83 / 110


Global flow dynamics: boundary conditions on local models Coupling the 3D FSI and 1D models

Benchmarking the new LAC conditions

Pressure distribution and displacement


Percentual error in the computed area of the for different sets of boundary conditions
outflow section

3D-1D Standard Reference LAC

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 84 / 110


Global flow dynamics: boundary conditions on local models Blood pressure in compartments: 0D models

Lumped parameters (0D) models


Analog to electric networks and derived by:
I integrating over the cross section, after considering simplifying assumptions
I further integration in space

Possible representation:
Q2 Q1
Q1 Q2 Q1 Q2
R L R L
P1 P2 P1 P2
C C

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 85 / 110


Global flow dynamics: boundary conditions on local models Blood pressure in compartments: 0D models

Lumped parameters (0D) models


Analog to electric networks and derived by:
I integrating over the cross section, after considering simplifying assumptions
I further integration in space
 dp  dp
1 2
 C
 + Q2 = Q1  C
 − Q1 = Q2
dt dt
 L dQ2 + RQ − p = p
  L dQ1 + RQ + p = p

2 1 2 1 2 1
dt dt

ρµl ρl 3lA0 R0
R= , L= , C = ,
πR04 A0 2Eh

Q2 Q1
Q1 Q2 Q1 Q2
R L R L
P1 P2 P1 P2
C C

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 85 / 110


Global flow dynamics: boundary conditions on local models Blood pressure in compartments: 0D models

Lumped parameters (0D) models


Analog to electric networks and derived by:
I integrating over the cross section, after considering simplifying assumptions
I further integration in space
 dp  dp
1 2
 C
 + Q2 = Q1  C
 − Q1 = Q2
dt dt
 L dQ2 + RQ − p = p
  L dQ1 + RQ + p = p

2 1 2 1 2 1
dt dt

ρµ(t)l ρl 3lA0 R0
R= , L= , C = ,
πR04 A0 2Eh

Q2 Q1
Q1 Q2 Q1 Q2
R L R L
P1 P2 P1 P2
C C

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 85 / 110


Global flow dynamics: boundary conditions on local models Blood pressure in compartments: 0D models

Lumped parameters (0D) models


Analog to electric networks and derived by:
I integrating over the cross section, after considering simplifying assumptions
I further integration in space
 dp  dp
1 2
 C
 + Q2 = Q1  C
 − Q1 = Q2
dt dt
 L dQ2 + RQ − p = p
  L dQ1 + RQ + p = p

2 1 2 1 2 1
dt dt
This two systems represent the same artery yet they generate different coupling
conditions since they have a different set of state variables and forcing terms
Q2 Q1
Q1 Q2 Q1 Q2
R L R L
P1 P2 P1 P2
C C

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 85 / 110


Global flow dynamics: boundary conditions on local models Blood pressure in compartments: 0D models

The 0D net S21


S1 S2
S22

R 211 L 211 R 212 L 212


R 11 L 11 R 12 L 12 R 2 L2
C 21
C1 C 21 C 22
R 221 L 221 R222 L 222

C 22

It gives rise to a ODE system (y is the vector of state variables):


dy


 = Φ(y, t) = A(y(t), t)y(t) + r(t) t ∈ (0, T ]
dt


y(0) = y0 .

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 86 / 110


Global flow dynamics: boundary conditions on local models Coupling 3D and 0D models

Coupling the 3D generalized Newtonian and 0D models


I Continuity of the fluxes Q
I Continuity of the mean pressure p

In a sub-structuring numerical scheme the quantities to be exchanged are governed by


the 0D model

Treatment of the defective boundary conditions


flow rate problem: Lagrange multiplier approach

Treatment of the defective boundary conditions

mean pressure problem: P tot n − 2µD(u) · n = P tot

The well-posedness of the coupled problem is demonstrated through a fixed-point argument


[A. Quarteroni, A. Veneziani (2003)]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 87 / 110


Global flow dynamics: boundary conditions on local models Coupling 3D and 0D models

Coupling the 3D generalized Newtonian and 0D models


Explicit coupling

I periodic pressure source


I 3D Carreau fluid: µ0 = 0.56Pas, µ∞ = 0.0345Pas, λ = 3.313s and n = 0.3568

mean pressure flow rate

[Sequeira, Moura, Janela, 2009]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 88 / 110


Global flow dynamics: boundary conditions on local models Coupling 1D and 0D models

Coupling 0D and 1D models

I Continuity of the fluxes Q


1D 0D
I Continuity of the mean pressure p
R1 R2

If a sub-structuring scheme is adopted at the nu- C

merical level the quantities to be exchanged are


governed by the 0D model: here the 0D model Interface Conditions

requires Q.

The well-posedness of the coupled problem (local in time result) can be


demonstrated by using a fixed-point argument [M. Fernandez, V. Milisik, A. Quarteroni (2005)]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 89 / 110


Global flow dynamics: boundary conditions on local models Coupling 1D and 0D models

Including the brain circulation


[Formaggia, Moura, Nobile, Passerini, APCOM’07]

1D model of the arteries from the


Coupling the 1D extremities with 0D models
heart to the circle of Willis
I coupling with 3-element 0D models of the
type R1 CR2 :
I Continuity of fluxes and mean pressures:

dp 1D DQ1
p 1D +R2 C = pv +(R1 +R2 )Q1D +R1 R2 C
dt dt

pv = 5mmHg pressure at te entrance of the venus


system

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 90 / 110


Global flow dynamics: boundary conditions on local models Coupling 1D and 0D models

Including the brain circulation


[Formaggia, Moura, Nobile, Passerini, APCOM’07]

1D model of the arteries from the


Coupling the 1D extremities with 0D models
heart to the circle of Willis
I coupling with 3-element 0D models of the
type R1 CR2 :
I Continuity of fluxes and mean pressures:

dp 1D DQ1
p 1D +R2 C = pv +(R1 +R2 )Q1D +R1 R2 C
dt dt

pv = 5mmHg pressure at te entrance of the venus


system

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 90 / 110


Global flow dynamics: boundary conditions on local models Coupling 1D and 0D models

Embedding the 3D FSI model into the 1D network


The circle of Willis (CoW) Scheme of the coupling

l.e.c. l.i.c.
vertebral arteries r.i.c.

l.c.c.

Applying a pressure impulse at the inflow boundary points


I Q = 16.3cm3 /s at the inflow of the tube upstream of the carotid (l.c.c.)
I Q = 12.75cm3 /s at the right carotid (on the 1d network of the CoW) (r.i.c.)
I Q = 6.8cm3 /s at the vertebral arteries (on the 1d network of the CoW)

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 91 / 110


Global flow dynamics: boundary conditions on local models Coupling 1D and 0D models

Embedding the 3D FSI model into the 1D network


Visualizing the detailed 3D FSI model and the 1D network of the CoW separately

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 92 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Physiological relevant fluid mechanics parameters


Most commonly used flow parameters associated with ICA are on or near the
vessel wall

I WSS - wall shear stress: τ n − (τ n · n)n (n → outward normal to the wall)


I WSSG - spatial wall shear stress gradients magnitude:
s 2  2
∂WSSζ1 ∂WSSζ2
WSSG = + ,
∂ζ1 ∂ζ2
ζ1 , ζ2 → perpendicular unit tangent vectors to the wall

∂WSSζ1 ∂WSSζ2
I TR - Near wall transport: + (Gambaruto et. al., J. Comp. Phys.)
∂ζ1 ∂ζ2

OSI - oscillatory shear index
I unsteady flow
GON - gradient oscillatory number

I AFI - aneurysm formation indicator.


I Disturbed Flow.
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 93 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Describing the local flow behavior

First order approximation of the velocity field in terms of the spatial coordinates:
∂ui
u = Ax, with Aij = ,
∂xj

The eigenvalues of A satisfy the characteristic equation: λ3 + Pλ2 + Qλ + R = 0,


where P, Q and R are matrix invariants given by:

P = −(A11 + A22 + A33 ), R = det A



A11 A12 A11 A13 A22 A23
Q=
+ + ,
A21 A22 A31 A33 A32 A33

I Q → a measure of rotation rate over strain rate.

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 94 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Particle tracking [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

I Particle tracking using identical seeding location at the aneurysm neck

GsN GsC

GSN GSC
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 95 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Iso-surfaces of Q [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

GsN GsC

GSN GSC
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 96 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Velocity magnitude [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

GsN GsC

GSN GSC

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 97 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

WSS [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

GsN GsC

GSN GSC

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 98 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

WSS and surface shear lines [Gambaruto, Janela, Moura, Sequeira, 2011]

GsN GsC

GSN GSC
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 99 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

WSSG magnitude [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

GsN GsC

GSN GSC

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 100 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

TR (WSSG sum) [Gambaruto, Janela, Moura, Sequeira, Mathematical Biosciences and Engineering, 2011]

GsN GsC

GSN GSC

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 101 / 110


Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Newtonian vs shear-thinning [Ramalho, Moura, Gambaruto, Sequeira, Math Mod Meth Biomed, 2012]

Velocity WSS WSSG


Newtonian
Difference Carreau

max = 2.9 max = 11.5 max = 191.9


(11.6%) (19.1%) (24.7%)
mean = 0.04 mean = 0.003 mean = 11.1
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 102 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Idealized (vel. [cm/s]) [Ramalho, Moura, Gambaruto, Sequeira, 2012]

mean flow = 13,6 cm/s

TF - TF TF - V0 TF - 1D TF - 0D

vel diff max = 16.3 max = 7.4 max = 1.8 max = 1.9
[cm/s] (59.9%) (27.4%) (6.6%) (7.0%)
mean = 0.17 mean = 0.11 mean = 0.06 mean = 0.05
WSS diff max = 23.5 max = 13.6 max = 15.0 max = 15.0
[dyn/cm2 ] (81%) (46.8%) (51,7%) (51.7%)
mean = 4.3e-4 mean = 8.2e-4 mean = 4.4e-4 mean = 4.8e-4
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 103 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators
INFLOW CONDITIONS
unsteady
6 steady

Idealized: steady vs unsteady (hole 1D) 5

flow rate [cm³/s]


4

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7


time [s]

steady unsteady average difference


[cm/s]
velocity

max = 1,4 (5,1%)


mean = 0,15
[dyn/cm2 ]
WSS

max = 1,8 (6,4%)


mean = 0,02
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 104 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators
INFLOW CONDITIONS
unsteady
6 steady

Idealized: unsteady (velocity [cm/s] ) 5

flow rate [cm³/s]


4

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7


time [s]
trac. free
1D
Diff

max = 12,4 (45,6%) max = 7,0 (25,7%) max = 2.8 (10,3%)


mean = 0,22
Alexandra Bugalho de Moura
mean = 0,16 mean=0,23
ESSIM, Dresden, August 2012 105 / 110
Global flow dynamics: boundary conditions on local models Vizualization: Hemodynamic indicators

Patient-specific vs Idealized [Ramalho, Moura, Gambaruto, A. Sequeira, IJNMBE, 2012]

Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 106 / 110


References

References (see references therein)


I A. Quarteroni and L. Formaggia. Mathematical modelling and numerical simulation of the
cardiovascular system. In: Handbook of Numerical Analysis, P.G Ciarlet and J.L Lions Eds.
and N. Ayache editors, Elsevier, Amsterdam, 2002. (MOX report 1)
I M. Fernandéz. Coupling schemes for the incompressible fluid-structure interaction:
implicit, semi-implicit and explicit. SeaMa Journal, 55:59-108, 2011. (INRIA report)
I L. Formaggia and A. Veneziani. Reduced and multiscale models for the human
cardiovascular system. Lecture notes VKI Lecture Series 2003-07, Brussels. 2003. (MOX
report 21)
I J. Cebral et. al. Characterization of Cerebral Aneurysms for Assessing Risk of Rupture By
Using Patient-Specific Computational Hemodynamics Models. American Journal of
Neuroradiology. 26:25502559. 2005.
I Taylor CA, Steinman DA. Image-based modeling of blood flow and vessel wall dynamics:
Applications, methods and future directions. Ann Biomed Eng 2010 Mar; 38(3): 1188-203
I L. Formaggia, A. Moura e F. Nobile. On the stability of the coupling of 3D and 1D
fluid-structure interaction models for blood flow simulations. ESAIM: Mathematical
Modelling and Numerical Analysis (M2AN), 41(4): 743-769, 2007.
I J. Janela, A. Moura e A. Sequeira. Absorbing boundary conditions for a 3D non-Newtonian
fluid-structure interaction model for blood flow in arteries.International Journal of
Engineering Sciences, 48(11): 1332-1349, 2010.
I A.M. Gambaruto, J. Janela, A. Moura e A. Sequeira. Sensitivity of hemodynamics in
patient specific cerebral aneurysms to vascular geometry and blood rheology.Mathematical
Biosciences and Engineering, 8(2): 409-423, 2011.
Alexandra Bugalho de Moura ESSIM, Dresden, August 2012 107 / 110

Das könnte Ihnen auch gefallen