CFD Analysis of Pressure and Flow Characteristics of theHuman NoseBJS - NC08A Major Qualifying Project Report:Submitted to the facultyof theWORCESTER POLYTECHNIC INSTITUTEIn partial fulfillment of the requirements for theDegree of Bachelor of SciencebyKalen SmithMarch 17thProf. Brian James Savilonis, Advisor
Table of ContentsACKNOWLEDGEMENTS3ABSTRACT4TABLE OF GRAPHS5CHAPTER 1: INTRODUCTION8CHAPTER 2: BACKGROUND RESEARCH92.1 OVERALL RESEARCH102.2 TURBULENCE MODELING27CHAPTER 3: METHODOLOGY33CHAPTER 4: RESULTS, ANALYSIS AND DISCUSSION394.1 STEADY STATE RESULTS394.2 UNSTEADY STATE RESULTS71CHAPTER 5: CONCLUSION104BIBLIOGRAPHY108APPENDIX: ADDITIONAL CFD RESULTS ON DVD
AcknowledgementsI would like to extend my appreciation to the following individuals without whom this project would nothave been possible.Professor Brian James Savilonis – Professor at Worcester Polytechnic InstituteDr. David B. Wexler, M.D. – Otolaryngologist at Fallon Clinic
AbstractCFD analysis is an alternative method to the construction of an experimental model formodeling airflow through human nasal cavities. In this study, a mesh was created from an MRI, andimported into Fluent for analysis. Simulations were run with various steady and unsteady models, whichwere then compared with each other. Steady state models were also run with both laminar and kepsilon turbulence models. For this patient and these particular flow patterns and geometry, the flowwas determined to be quasi-steady since the steady and unsteady models correlated well with eachother. Also, it was found that the choice of which viscous model to use (in this case, k-epsilon vs.laminar) was found not to play a significant role in the outputs. Finally, the results of this study weresomewhat different from the results determined experimentally.
Table of GraphsGraph 1: The grid and cutting planes selected. The white lines shown represent the cutting plane that was used to create side views 34Graph 2: Pressure drop of laminar model at flow rate of 100 cc/s 41Graph 3: Pressure contours over nostrils of laminar model at 100 cc/s .42Graph 4: Side view pressure contour of laminar model at 100 cc/s 43Graph 5: Velocity contours near nasal inlet of laminar model at 100 cc/s45Graph 6: Velocity contours near turbinate region of laminar model at 100 cc/s46Graph 7: Side view velocity vector plot of laminar model at 100 cc/s46Graph 8: Pressure drop graph of k-epsilon model at 100 cc/s48Graph 9: Pressure contours of nostrils of k-epsilon model at 100 cc/s49Graph 10: Side view pressure contours of k-epsilon model at 100 cc/sGraph 11: Velocity contours near turbinate region of k-epsilon model at 100 cc/s5050Graph 12: Velocity contours near inlet of k-epsilon model at 100 cc/s52Graph 13: Side view of velocity vectors of k-epsilon model at 100 cc/s53Graph 14: Turbulence intensity contours of k-epsilon model at 100 cc/s54Graph 15: Pressure drop of k-epsilon model at 270 cc/s56Graph 16: Side view of pressure contour of k-epsilon model at 270 cc/s57Graph 17: Velocity contours near inlet region of k-epsilon model at 270 cc/s57Graph 18: Velocity contours near turbinate region of k-epsilon model at 270 cc/s58Graph 19: Side view vector plots of k-epsilon model at 270 cc/s59Graph 20: Turbulence intensity plots of k-epsilon model at 270 cc/s59Graph 21: Pressure drop of laminar model at 270 cc/s61
Graph 22: Side view pressure contour plot of laminar model at 270 cc/s62Graph 23: Velocity contours near inlet region of laminar model at 270 cc/s63Graph 24: Velocity contours near turbinate region of laminar model at 270 cc/s63Graph 25: Velocity vector plot of laminar model at 270 cc/s64Graph 26: Pressure drop of k-epsilon model at 500 cc/s65Graph 27: Pressure contour of nostrils of k-epsilon model at 500 cc/s65Graph 28: Side view pressure contour plot of k-epsilon model at 500 cc/s66Graph 29: Velocity contour near inlet of k-epsilon model at 500 cc/s67Graph 30: Velocity contour near turbinate region of k-epsilon model at 500 cc/s68Graph 31: Side view velocity vector plot of k-epsilon model at 500 cc/s69Graph 32: Turbulence intensity contour of k-epsilon model at 500 cc/s70Graph 33: Pressure drop of laminar model at 500 cc/s71Graph 34: Pressure contours of nostrils of laminar model at 500 cc/s71Graph 35: Side view pressure contour of laminar model at 500 cc/s72Graph 36: Velocity contour near inlet region of laminar model at 500 cc/s73Graph 37: Velocity contour near turbinate region of laminar model at 500 cc/s73Graph 38: Velocity vector plot of laminar model at 500 cc/s74Graphs 39: Pressure drops of unsteady flow at 0.4 seconds77Graphs 40: Pressure drops of unsteady flow at 0.8 seconds78Graphs 41: Pressure drops of unsteady flow at 1.2 seconds78Graphs 42: Pressure drops of unsteady flow at 1.6 seconds78Graphs 43: Pressure drops of unsteady flow at 2.0 seconds78Graphs 44: Pressure drops of unsteady flow at 3.5 seconds78Graphs 45: Side view contour plots of unsteady flow at 0.4 seconds.81
Graphs 46: Side view contour plots of unsteady flow at 0.8 seconds.82Graphs 47: Side view contour plots of unsteady flow at 1.2 seconds.82Graphs 48: Side view contour plots of unsteady flow at 1.6 seconds.82Graphs 49: Side view contour plots of unsteady flow at 2.0 seconds.82Graphs 50: Side view contour plots of unsteady flow at 3.5 seconds.82Graphs 51: Velocity vectors of unsteady at flows 0.4 seconds.83Graphs 52: Velocity vectors of unsteady at flows 0.8 seconds.83Graphs 53: Velocity vectors of unsteady at flows 1.2 seconds.84Graphs 54: Velocity vectors of unsteady at flows 1.6 seconds85Graphs 55: Velocity vectors of unsteady at flows 2.0 seconds.85Graph 56: Velocity vectors of unsteady at flows 3.5 seconds.86Graphs 57: Turbulence plots of unsteady at times of 0.4 seconds87Graphs 58: Turbulence plots of unsteady at times of 0.8 seconds87Graphs 59 Turbulence plots of unsteady at times of 1.2 seconds88Graphs 60: Turbulence plots of unsteady at times of 1.6 seconds89Graphs 61: Turbulence plots of unsteady at times of 2.0 seconds89Graphs 62: Turbulence plots of unsteady at times of 3.5 seconds90
Chapter 1: Introduction:Developing an understanding of the pressure and flow behavior that take place during normalbreathing conditions has been a subject of much discussion and research. Traditionally, this has beenaccomplished by building a model based on the nose of a human cadaver or CAT scan or MRI images of ahuman nose, and making a fluid pass through the model in a manner such that it resembles the flowconditions exhibited during the normal breathing cycle. Measurements are made, and the empiricaldata are analyzed to estimate the flow characteristics. Performing such an experiment is believed toyield fairly accurate results, provided that the experiment is properly constructed. However, due to thecomplexity of such an experiment, there are a number of things that could go wrong, and therefore theresults may not accurately reflect the actual breathing conditions that the researchers are trying toemulate.An alternative to constructing a physical experiment is to perform a Computational FluidDynamics (CFD) analysis. This involves taking a meshed geometry, and using a CFD software package tocreate a simulation resembling the real world flow. There are several differences between using CFDAnalysis and a physical experiment to analyze any type of flow. First of all, the CFD approach relies oncomputer tomography (CT) scans, which can yield consistent results whenever a simulation is run.However, no matter how advanced the CFD package is, the CFD approach is limited in that it cannot takeinto account all the physical variables that play a role in a real world simulation. The CFD results tend togive results that suggest how the flow is expected to behave, rather than how it actually does behave ina real world simulation. As a result, there is often a discrepancy between the results found in a physicalexperiment and those found during a CFD analysis. Therefore, in order to determine the validity of CFD
analysis, it is often necessary to compare the results from CFD simulations with those foundexperimentally.However, CFD analysis has many benefits. It is much cheaper and faster to create a CFDsimulation than to complete an experiment. Therefore, it is possible to run many more CFD models inthe cost and time frame of a project than would be possible by running a physical experiment. Becauseof the time and cost savings, using the CFD approach allows the possibility to analyze problems orresearch questions which require substantial amounts of data. CFD results can also be easier to analyzethan experimental data. With CFD, it is possible to observe detailed physical behavior taking place atany location within the flow, which makes it easy to determine the effects of geometry on flow behavior.CFD can also analyze physical properties which would be difficult to analyze experimentally, such asturbulence, shear stress and the flow patterns of streamlines.
Chapter 2:BackgroundIn a study by Croce , the experimenters sought to determine pressure-flow relationships forairflow through a subject’s nasal passages. This was done by creating an in vitro experiment as well as anumerical (CFD) analysis, and the results from each of these approaches were compared with those ofthe other and the literature from previous experiments. The most important purpose of this study wasto determine the validity of using CFD software to monitor or analyze nasal airflow.The first step in conducting the experiment was the creation of a plastinated model. After themodel had been constructed, the next step for the researchers was to determine the types of gases thatneeded to be used and begin simulating the flow through the model. During the process of inspiration,a potentiometer was used to control flow rates, and a pneumotograph was used to measure pressure.Expiration was studied by connecting the compressed air tap to a valve.The numerical study was conducted by taking 3-D scanned images of the plasticized model.During a process known as segmentation, the experimenters could isolate the various parts of the modelthat they were interested in. This involved tracking every individual pixel and making sure that eachpixel could be identified with the location it refers to in the actual geometry. After this, the mesh wascreated with about 100,000 triangular faces and 830,000 tetrahedral cells. This mesh was tested, but itwas found that due to large pressure gradients in the flow fields, a finer mesh of 1,350,000 tetrahedralcells was needed.After the mesh was created, the experimenters were ready to begin the simulation. It waspredicted that they would have flow that was steady, incompressible and laminar. The FLUENT software
package was used for this purpose. Because of the way that the mesh was segmented, a segregationsolver was applied. A pressure difference was applied and standard no-slip conditions were assumed atthe boundaries. Boundary conditions were established at the front of the nose and outlet of the model.Finally, the simulation was ready to begin and data collected which is shown in Figure 1.Figure 1: Pressure drops vs. Reynolds Number for both nostrils. 1After data was collected from both the experimental and numerical models, the data from thetwo models were compared with each other. The data from the experiments and numerical modelswere very similar, except when analyzing very high flows. Even at these high flow rates, the databetween the two experiments did not differ by more than 17%. The fact that the data of the twomodels corresponded well together seemed to justify that a numerical model could yield results in syncwith those of similarly constructed experimental models.When analyzing the data relating to the pressure changes, it was observed that the right nostrildisplayed more negative pressure than the left nostril, which can be a result of the fact that there was alarger flow rate on the left side of the nose. This can be shown by the contours in Figure 2.
Figure 2: Contour plots of Croce study1
The reason for the discrepancy between the experimental and numerical models at high flowrates may be due to the higher turbulence level while the numerical model was based on laminar flowcalculations. This study suggests that flow becomes turbulent when a flow rate of 250 ml/s is exceeded.This transition value is slightly smaller than the one the experimenters were comparing to in theliterature. One possible explanation to this difference lies in the fact that in the computerized modelthe researchers were able to look at “cuts” spaced 0.5 mm away from each other, while those of thephysical models were spaced 2 to 4 mm apart. The velocities that were obtained in this research weresimilar to those of the experiment conducted by other researchers. The maximum velocity of thisexperiment was about 3.1 m/s, and those of several other researchers were approximately 4 m/s, butthis is to be expected since the areas they were calculating over were slightly larger (quantitative valuesfor these areas are not described in the article).Until the development of CFD analysis, it was very difficult to analyze the effects of sniffing onnasal airflow. In a study by Zhao , a CFD approach was taken to determine the effects of sniffing onboth airflow and odorant transport in nasal passages of both human and rat noses, however in critiquingthis article, emphasis was placed on the analysis of nasal airflow. A 3D nasal model was constructedfrom a CT scan which included both nostrils and the nasopharynx. Inspiratory steady-state airflowswere simulated with both laminar and turbulent models. In this study, one of the main objectives wasto estimate the amount of turbulence in the airflow. The turbulence was approximated by computingthe turbulenc