Human Computational Fluid Dynamics:
Analysis of Nose Flow
Wolfgang Schröder, Andreas Lintermann,
Lennart Schneiders, Jerry Grimmen
Institute of Aerodynamics
RWTH Aachen University
JARA High Performance Computing
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Anatomy of the Nasal Cavity
Functions
Sense of Smell (Regio olfactoria)
Tempering Air (Turbinates)
Isolation (Airfilled Cavities)
Moistening (Goblet Cells)
Resonance Organ (Paranasal Sinuses)
Cleaning Air (Ciliated Epithelium)
Physiological Data
Physiological Respiration through Standard Nose
(R. Hincliff, D. Harrison)
minute ventilation [1/min ]
ventilation frequency [1/min ]
tidal volume
[ml ]
medium
respiration
maximum
respiration
6-8
15
400-500
50-70
25
2000-2800
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Computer Tomography of
the Human Nasal Cavity
Human  Engineering
Surface Extraction by Computer
Tomography
Marching
Cube
Algorithm
• 300 Cuts, 1mm Spacing
• Unstructured Surface
• DICOM Format
• 749.681 Nodes
• 512  512  2 Bytes per Cut
• 1.499.065 Triangles
Silicone Nose Model
Grids
clean + upper and lower turb.
Numerical Method
Navier-Stokes equations, 3D, time dependent
Approximation: Finite Volume Method
• second-order accuracy for non-Euler terms
• AUSM (Advective Upstream Splitting Method) for Euler terms
• time integration via 5-step Runge-Kutta method of second-order
accuracy
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Horseshoe Vortex
The Horseshoe Vortex
Vortex Breakdown
vortex
free
stagnation
point
Vortex Breakdown (cntd.)
Inhalation: Streamlines
upper and lower turbinate and spurs
Comparison Numerics and Experiments
Inhalation
cross section 1
num.
exp.
cross section 2
num.
exp.
Comparison Numerics and Experiments
Exhalation
cross section 1
num.
cross section 2
exp.
num.
exp.
Scheme for Human Respiration Cycle
Inhalation/Exhalation Process
Pressure Loss vs. Mass Flux
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Engineering  Human
Human Nasal Cavity via CT-Images
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Grid refinement data structure
l0 (80 cells)
l1 (O(81) cells)
l2 (O(82) cells)
l3 (O(83) cells)
l0 (1 cell)
l1 (8 cells)
l2
Octree structure with parent-child relation
Boundary refinement
l
l+2
l+1
l
the refined boundary is smoothed by ensuring a level difference of 1
Number of offspring reduction
l (M)
l+1
l+t
l+1 (M)
l+ t
Moving subtrees to the upper level
Splitting of subtrees
level l
level l + 1
levels l+2 … l+t
a copy of the split subtree (n) is introduced to the process
Mesh Generation: Sphere
Mesh Generation: Dinosaur
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Varying Boundary Cells I
• Cut cells may become arbitrarily small
• Result in numerical instability
• Explicit time integrators require a very small time step to remain stable
 Small cells must be removed
• Abrupt changes of the discrete operators result in perturbations
• Smooth transition of leastsquares stencils required
• Disappearing cells?
Varying Boundary Cells II
• Cut cells may become arbitrarily small
• Result in numerical instability
• Explicit time integrators require a very small time step to remain stable
 Small cells must be removed
• Abrupt changes of the discrete operators result in perturbations
• Smooth transition of leastsquares stencils required
• Disappearing cells?
remain as ghost nodes on
the boundary
Emerging and Merging Cells
nt
nt
(n + 1) t
(n + 1) t
Discrete Operator Weighting Functions
L. Schneiders et al., JCP 235: 786-809 (2013)
Transversely Oscillating Circular Cylinder I
Re = 185, yB = A cos (2fet) , A = 0.2D , fe = 0.8 f0 ,
Sr = f0 D/u = 0.195
locally refined mesh
vorticity contours
(cyl. at tdc)
Transversely Oscillating Circular Cylinder II
cell-merging method vs. weighting-function formulation (,
w)
Dancing Cylinders
Vorticity distribution
Folie Schneiders
Mesh Generation: Nasal Cavity
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Flow through the Human Nasal Cavity
Three Nasal Cavities
good
poor
fair
“Good” Geometry: Streamlines
turbinate
“Fair” Geometry: Streamlines
“Poor” Geometry: Streamlines
“Good” Geometry: Wall-Shear Stress
“Fair” Geometry: Wall-Shear Stress
Comparison of Three Geometries
heating
pressure loss
good
fair
poor
good
fair
poor
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Coming Up
• Introduction
• From the Human Nose to the Engineering Model
* laminar or turbulent
* steady or unsteady
• From the Engineering Model to the Human Nose
* General Description
* Mesh Generation
* Accuracy Issues
* Results of the Human Nose: Comparison of 3 Geometries
• Conclusion
Conclusion
The Good: Massively parallel grid generation on HPC systems;
numerical and experimental tools to automatically
analyze local and global phenomena are available
The Bad: Analysis is costly
The Ugly: Uncertainty is high due to too little knowledge on
bio-medical structures, mucous membrane,
tissues, particles etc.