Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
e-mail: philippe.versailles@mail.mcgill.ca
Jeffrey M. Bergthorson
Associate Professor
Mem. ASME
e-mail: jeff.bergthorson@mcgill.ca
Department of Mechanical Engineering,
McGill University,
Montreal, QC, H3A 0C3, Canada
Optimized Laminar
Axisymmetrical Nozzle Design
Using a Numerically Validated
Thwaites Method
This paper presents the Thwaites method as an accurate and efficient design tool for laminar, axisymmetrical nozzles. Based on historical developments, it is improved to describe
internal flows with highly favorable pressure gradients in cylindrical coordinates. The
calculation of the core flow velocity distribution based on the continuity equation is proposed as a replacement to other sophisticated numerical methods. A remarkably good
agreement is obtained when comparing the results of the current Thwaites method
against those of computational fluid dynamics (CFD) simulations, for which the integral
boundary layer thicknesses are calculated with equations developed from first principles
in the course of the work. This consistency among the results and the low time and
resource costs of the Thwaites method confirm its applicability and usefulness as an engineering design and optimization tool. [DOI: 10.1115/1.4007155]
Introduction
Appropriately designed nozzles are critical for proper performance of large-scale wind tunnels [1], small-scale fundamental
combustion [2,3], or fluids experiments [4], as well as microscale
propulsion devices [5]. Over the years, various methods for
designing nozzle contractions have been used. They ranged from
by eye empirical techniques to others involving the selection of
a given velocity distribution along the axis, from which the stream
function, achieving a reasonable pressure gradient, is chosen as
the optimal nozzle contour. Moreover, panel and finite difference
methods provided more detailed inviscid (potential flow) solutions. These techniques are summarized in Ref. [6]. Bell and
Mehta [6,7] employed a 3D, inviscid, panel code to get the velocity distribution at the wall of contractions with noncircular cross
section that was then fed to the Thwaites method [8] for the prediction of the boundary layer characteristics. This removed the
need to rely on separation criteria, as was needed in earlier techniques. However, the work of Bell and Mehta did not analyze axisymmetrical internal flows, and the results of the numerical
method were solely validated against experimental data obtained
slightly downstream of the contraction outlet. Further studies
investigating the validity of the Thwaites method for the prediction of boundary layer characteristics inside nozzle contractions
are required. A similar methodology to the one of Bell and Mehta
has been used recently [9], with the potential flow solution
obtained using the OpenFoam CFD package, to compare the performance of transformed fifth-order polynomial shapes against the
original, well known, fifth-order profile proposed in Ref. [6]. Such
CFD and panel method computations are useful for case to case
comparison. Nevertheless, they are not practical for the study of
large design spaces, as often encountered in engineering situations, because they would require the generation of numerous 2D/
3D grids and a significant amount of time and resources in analyzing and postprocessing all the possible configurations.
In contrast to such detailed solutions, the Thwaites method has
been applied for the design of axisymmetrical nozzles, with the
2
1
Corresponding author.
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received August 22, 2011; final
manuscript received July 10, 2012; published online September 27, 2012. Assoc.
Editor: Mark F. Tachie.
C 2012 by ASME
Copyright V
2
dx
U dx U dx
qU
(1)
(2a)
sw h
Sk
qU
(2b)
U
d h2
2 Sk k 2 Hk Fk
dx
(3)
(4)
0:45
U6
U 5 dx
(5)
By performing the integral of Eq. (5) for a given U(x), the momentum thickness and the HolsteinBohlen parameter, k, are
obtained at every x-location. Thwaites proposed the fifth order
polynomial of k shown by the gray dash-dot line in Fig. 1 to determine H(k). This correlation is valid down to k 0.09, where
boundary layer separation is expected to occur. This separation
prediction capability of the Thwaites method is very useful in
assessing the performance of nozzles. Once the shape factor H is
obtained, the displacement thickness d* can be calculated and the
boundary layer is fully characterized as a function of x.
The range of validity of the Thwaites correlation for H(k) is restricted to k lower than 0.25, and inaccuracies in Eq. (4) have
been observed. Based on solutions to FalknerSkan velocity profiles, Dey and Narasimha [13] proposed the second order polynomial Eq. (6) in replacement to Eq. (4) and the H(k) correlation
shown by the gray dotted line in Fig. 1. Assuming that the second
order term of Eq. (6) is a small perturbation (DF) applied on the
original linear fit of Thwaites (Eq. (4)), they developed Eq. (7) for
the calculation of the momentum thickness. The first integral in
the bracket of the RHS is identical to the original solution of Eq.
(5), while the second is a usually small correction for highly
favorable pressure gradients. With these modifications, the
Thwaites method is expected to be valid for k up to 0.4, which
corresponds to highly accelerating flows [13].
Fk 0:45 6:0 k 2:0 k2
101203-2 / Vol. 134, OCTOBER 2012
(6)
0:45
hx2
U6
x
0
U5 dx 0:9
x
0
dU
dx
2
1
7
U
x0
!2
U 5 dx0
3
dx5
(7)
For axisymmetrical bodies, Rott and Crabtree [14] developed
Eq. (9) based on Eq. (8), which is the cylindrical equivalent to
Eq. (3). The same H(k) and F(k) correlations as for external planar
flows are assumed to apply, boundary layer separation is again
anticipated at k 0.09, and an additional assumption is that the
boundary layer thickness d must be much smaller than the local
radius of the body r0(x). The expected accuracy of Eq. (9) is
65%.
U d r02 h2
2 Sk k 2 Hk Fk
r02 dx
0:45 x 2
hx2 2
r U 5 dx
r0 U 6 0 0
(8)
(9)
In order to extend the range of validity of Eq. (9) to higher favorable pressure gradients, the strategy of Dey and Narasimha is
applied to axisymmetrical flows in the present work. Keeping
their F(k) correlation Eq. (6), a corrected equation for the momentum thickness in cylindrical coordinates is developed,
x
0:45
r02 U 5 dx
hx2 2
r0 U 6
0
3
!2
x0
x 2
dU
1
2
5
0
2
r U dx dx5 (10)
0:9
r0 U7 0 0
0 dx
The H(k) correlations of Thwaites [12] and Dey and Narasimha
[13] are in close agreement for 0.1 < k < 0.17, but they display
significant differences for k < 0.1, as seen in Fig. 1. A new formula for H(k) is proposed here. To build the current H(k) relation
shown by the black solid line in Fig. 1, the correlation of Thwaites
is used for the low end of k values, owing to its historical widespread utilization, while the correlation of Dey and Narasimha is
used for the highest values of HolsteinBohlen parameter. The
union is at k 0.15, and efforts have been made to ensure that
H(k) is second-order continuous to obtain smooth solutions in
terms of displacement thickness.
the Thwaites method and CFD. Its contour is given by a 7th order
polynomial with first- and second-order derivatives equal to zero
at the inlet and outlet. The exit diameter of 0.01 m (0.400 ) is
selected based on previous well-performing nozzles [15], as well
as to minimize the required gas flow during the experiments. With
the inlet diameter of 0.032 m (1.2500 ), this leads to a suitable contraction ratio of 9.77 for flow relaminarization of disturbances [6].
The last two degrees of freedom of the polynomial are fixed by
adjusting the location of two points (x1, r0,1) and (x2, r0,2). The
nozzle has a length of 0.0762 m (300 ), which is sufficiently short to
minimize the growth of the boundary layer, but long enough to
prevent separation and instabilities in the concave section and to
achieve uniform flow at the outlet [2]. An entrance length corresponding to 20% of the radius at inlet of the contraction, r0(x 0),
is added upstream of the nozzle. This is the optimal distance
between the turbulence management assembly and the entrance of
the nozzle to prevent flow distortion through the last screen and to
minimize boundary layer growth [16]. Figure 2 shows the nozzle
contour without entrance length and the first- and second-order
derivatives of the radius r0 as a function of the axial location x.
3.1 Thwaites MethodIterative Calculations of the Boundary Layer Integral Parameters. In the current study, the
Thwaites method for axisymmetrical internal flows presented in
Sec. 2 is implemented in a Matlab algorithm. The core equation of
the code is Eq. (10), which, once solved, allows finding h(x),
d*(x), as well as other important flow properties. The nozzle
length is discretized with equally spaced nodes located at the
same axial position as for the CFD mesh for easier point-to-point
comparison. A supplementary length upstream of the simulated
nozzle has been added in the Thwaites simulation to account for
the momentum thickness at the nozzle inlet due to the growth of
the boundary layer in the settling chamber. The length is adjusted
such that the momentum thickness at x 0 (see Fig. 2) is the same
for both CFD and Thwaites simulations. In the end, the results of
the Thwaites code are obtained with 1384 equally spaced nodes
along the nozzle. However, preliminary results for a mesh consisting of 232 nodes almost perfectly overlap those obtained with the
finest discretization.
At first, U(x) must be determined in order to solve Eq. (10). In
contrast to previous studies using sophisticated panel or CFD
methods to obtain the potential core flow velocity [6,7,9], the current study assumes that the axial velocity is constant over the
cross section for all axial positions. The local velocity corresponds
to the volumetric flow rate normalized by the local area, and it can
be calculated with Eq. (11), derived from the continuity equation
for incompressible flows. The volumetric flow rate at one axial
position must be set in order to get U(x) at all other locations. For
that purpose, the outlet velocity U(xoutlet) is set to 3 m/s, a relevant
value for the current design.
Ux r0 x2 Uxoutlet r0 xoutlet 2
(11)
Fig. 3
Computational domain
Mesh #
NI
RI
NA
RA
NO
RO
Total number
of cells
Mean difference in
h versus mesh #6
1
2
3
4
5
6
80
160
240
160
240
400
1.100
1.075
1.020
1.054
1.020
1.020
425
850
1275
1913
1950
1275
1
1
1
1
1
1
80
160
240
160
240
400
1.032
1.032
1.010
1.032
1.010
1.010
34,000
136,000
306,000
306,080
468,000
510,000
0.77%
0.36%
0.04%
0.29%
0.04%
represent the number of cells and the successive ratio of cell size
(the finest being at the wall), respectively. The subscripts I, A, and
O stand for the inlet, axis, and outlet edges, respectively, as shown
in Fig. 3. In order to assess mesh convergence, the mean difference over the length of the nozzle in terms of momentum thickness (h) between each grid and mesh #6 is included in the table.
Increasing the number of cells from 34,000 to 136,000 removed
oscillations in the boundary layer thicknesses sampled along the
axial direction and decreased the average difference by a factor
of 2. A smoother curve of H against x was obtained with mesh
#3, and the mean difference decreased by about an order of magnitude in comparison to grid #2. Increasing the number of nodes
on the axis, while keeping the total number of cells almost constant, yielded a higher mean difference for mesh #4. This is due to
the coarser mesh along the radial direction. On the other hand,
keeping the same mesh characteristics on the inlet and outlet
boundaries, but increasing the number of nodes on the axis with
mesh #5 did not change the mean difference in comparison to grid
#3. This indicates that 1275 nodes equally spaced on the axis are
sufficient. Furthermore, the relatively small difference in terms of
momentum thickness between meshes #3 and #6 does not require
the use of the finest mesh. Mesh #3 is then used to produce the
results presented in the remainder of this paper. In order to minimize the error, the CFD simulations were allowed to run until stabilization at minimal values of the scaled residuals of the
continuity, x-momentum, and y-momentum equations. Over most
of the 13,000 iterations required for convergence, the residuals
reduced smoothly and almost linearly (on a log scale) toward
1013, 1016, and 1017, respectively.
For the sake of brevity, images of the CFD solution will be
omitted. However, contours of the main variables (pressures,
velocities, vorticity, derivatives, etc.) have been verified to be free
of oscillations caused by insufficient mesh refinement. The flow
field consists of an accelerating potential core stream, within
which the vorticity is negligible and the total pressure constant,
surrounded by a very thin boundary layer close to the wall. In this
last region, the velocity decreases to zero at the wall due to the
no-slip condition, vorticity is much higher than in the core flow,
and the total pressure decreases toward the wall due to viscous
losses.
Fig. 4 Control volume for the calculation of the integral parameters in internal cylindrical coordinates
3.3 Boundary
Layer
Integral
Parameters
for
Axisymmetrical Internal Flows. Equations for d* and h for planar [8] and external axisymmetrical [11] flows have been developed based on the assumptions that (1) the deflection of the
streamlines due to the presence of the boundary layer does not
change the pressure field significantly in the potential flow and (2)
the velocity of the surrounding free-stream flow U is solely a
function of x. The first assumption has been verified for internal
nozzles with the CFD simulation described in Sec. 3.2. However,
the velocity in the nozzle core potential flow U0(x,r), analogous to
the free-stream velocity over a bluff-body, varies with the radial
location r, as illustrated by the gray velocity profile in Fig. 4.
Since U0 is a function of the radial location, a technique must be
derived to determine d* and h.
Exact equations for calculating the necessary boundary layer
thicknesses from CFD solutions have been developed in the
101203-4 / Vol. 134, OCTOBER 2012
q U0 x; r 2pr dr
q U0 x; r ux; r 2pr dr
Rd
(13)
Since, U0(x,r) is equal to u(x,r) outside of the boundary layer,
Eq. (13) can be simplified to
Rd
U0 x; r r dr
ux; r r dr
(14)
R
Rh
U02 x; r r dr
R
Rd
2
U0 x; r u2 x; r
U0 x; R d U0 x; r ux; r r dr
(15)
The first two terms in the bracket of the RHS are, respectively, the
inflowing and outflowing momentum fluxes through the left and
right boundaries of the control volume, respectively, while the
third term is the momentum flux exiting through the flow rate
m_ out . For ease of integration in the algorithm, Eq. (15) is rewritten
in a form similar to Eq. (14) by again invoking that U0 and u are
equal outside of the boundary layer,
Rh
0
U02 x; r r dr
2
u x; r U0 x; R d
U0 x; r ux; r r dr
(16)
Results
CFD. In these figures, the black solid and gray dashed lines are
the Thwaites and CFD solutions, respectively. The CFD solution
65% is also represented by dotted lines.
As seen in these figures, the agreement between both methods
is very good, within 5% everywhere in the nozzle except in the
last 7 millimeters, where the difference reaches 8.2% and 7.5%
for the momentum and displacement thicknesses, respectively.
This tends to confirm the expected accuracy (65%) of the
Thwaites method in cylindrical coordinates [8], specifically for internal flows. In addition, it validates Eqs. (10), (14), and (16)
developed during the work and demonstrates that the proposed
construction of U0(x,r), Eq. (17), is correct. The discrepancies in
the results might be due to the acceptable, but not exact, relation
for H(k) and F(k) in the Thwaites method, as well as to the u velocity profile that slightly varies radially in the CFD results, as
presented in the following. CFD simulations have been made with
an extension of 0.0254 m downstream of the nozzle in an attempt
to improve the agreement between both methods close to the outlet. However, this did not allow reconciliation of both Thwaites
method and CFD results.
The behavior of the boundary layer thicknesses along the nozzle can be explained from the evolution of the velocity derivative
(dU/dx) and of the favorable pressure gradient (@P/@x), shown
in Figs. 7 and 8, respectively. Exact solutions to the boundary
layer equations for accelerating free-stream velocities in 2D planar and axisymmetrical coordinates show that decreasing the pressure gradient @P/@x to more negative values (thus increasing the
flow acceleration and making the pressure gradient more favorable) reduces the boundary layer integral thicknesses [8,17].
In the first few millimeters of the nozzle, the pressure gradient
is relatively weak. This is because the flow acceleration caused by
the reduction of the radius, given by Eq. (18) for the Thwaites
method and showed by the gray dashed curve in Fig. 7, is relatively small, since r0 (x 0) 0. The growth of the boundary layer
results in a small acceleration of the flow and is the dominant
mechanism of flow acceleration in the entrance length and the first
few millimeters of the nozzle, as shown by the black dash curve
in Fig. 7.
dU
2 dr0
Uxoutlet r0 xoutlet 2 3
(18)
dx due to
r0 dx
radius
As x rises, the radius change of the nozzle is dominant in determining dU/dx (solid line in Fig. 7), and the increasing favorable
pressure gradient results in a decrease of the boundary layer thicknesses along the x-direction, as seen in Figs. 5 and 6. This reduction of the integral parameters is also enhanced in the concave
region, which extends up to x 0.0523 m. It has been shown that,
for a given free-stream velocity with imposed velocity gradient,
increasing the curvature of a 2D plate reduces the boundary layer
thicknesses significantly [18]. Around x 0.066 m, the favorable
pressure gradient reaches its maximum value (highest @P/@x)
and then decreases toward the outlet of the nozzle because of a
significant reduction in the absolute value of r00 (x) in this region.
As the pressure gradient decreases, its coercive effect on the
growth of the boundary layer reduces and d* and h increase toward the nozzle outlet.
The agreement of the Thwaites method with CFD is partly due
to the consistency between U and u, and to the fact that the viscous velocity profile has only a small variation over the radius in
the core potential region at a given axial location. This can be
seen in Fig. 9, which shows that u(x, r 0) is close to u(x,
r R d) almost everywhere in the nozzle. However, even the
small difference between u(x, r 0) and u(x, r R d) (i.e., ra-
4.1 Flow Laminarity. The Thwaites method is valid for laminar flows, and the motive of the present paper is to demonstrate
d 2 r0
dx2
(19)
The performance of the current nozzle in terms of forced transition is shown in Fig. 15 for both computational methods. Again, a
good overlap of the curves is obtained, and the Gortler parameter
is well below the critical value of 53. Therefore, Gortler vortices
should not form for the operating condition studied, and forced
transition of the boundary layer is not expected.
Conclusion
Acknowledgment
The authors are thankful to Paul E. Dimotakis, John K. Northrop Professor of Aeronautics and Professor of Applied Physics at
the California Institute of Technology, for proposing this nozzle
101203-8 / Vol. 134, OCTOBER 2012
Nomenclature
F
G
H
KMIE
LHS
m_
N
NSBL
P
R
r
r0
r00
r000
reff
Red*
Red*crit
Reh
RHS
S
u
U(x)
U0(x,r)
x
Greek Symbols
d
d*
h
k
q
sw
Subscripts
A on the axis
I at the inlet
O at the outlet
References
[1] Liepmann, H. W., 1943, Investigations on Laminar Boundary-Layer Stability and Transition on Curved Boundaries, National Advisory Committee
for Aeronautics, Technical Report No. NACA-WR-W-107, NACA-ACR3H30.
[2] Bergthorson, J. M., 2005, Experiments and Modeling of Impinging Jets and
Premixed Hydrocarbon Stagnation Flames, Ph.D. thesis, California Institute of
Technology, Pasadena, CA.
[3] Bergthorson, J. M., Salusbury, S. D., and Dimotakis, P. E., 2011, Experiments
and Modelling of Premixed Laminar Stagnation Flame Hydrodynamics,
J. Fluid Mech., 681, pp. 340369.
[4] Dowling, D. R., 1988, Mixing in Gas Phase Turbulent Jets, Ph.D. thesis, California Institute of Technology, Pasadena, CA.
[5] Hao, P.-F., Ding, Y.-T., Yao, Z.-H., He, F., and Zhu, K.-Q., 2005, Size Effect
on Gas Flow in Micro Nozzles, J. Micromech. Microeng., 15(11), pp.
20692073.
[6] Bell, J. H., and Mehta, R. D., 1988, Contraction Design for Small Low-Speed
Wind Tunnels, NASAAmes Research Center/Stanford University, Technical Report No. NASA-CR-177488.
[7] Bell, J. H., and Mehta, R. D., 1989, Boundary-Layer Predictions for Small
Low-Speed Contractions, AIAA J., 27(3), pp. 372374.
[8] White, F. M., 1991, Viscous Fluid Flow, McGraw-Hill, New York.
[9] Doolan, C. J., 2007, Numerical Evaluation of Contemporary Low-Speed
Wind Tunnel Contraction Designs, ASME J. Fluids Eng., 129(9), pp.
12411244.
[10] Fujiso, Y., Wu, B., Grandchamp, X., and Hirtum, A. V., 2011, Caracterisation
` La Validation Experimentale De
Dun Convergent Axisymetrique Destine A
Production De Parole, Actes des 9e`mes Rencontres Jeunes Chercheurs en Parole, pp. 14.
[11] Cipolla, K. M., and Keith, W. L., 2003, Momentum Thickness Measurements
for Thick Axisymmetric Turbulent Boundary Layers, ASME J. Fluids Eng.,
125(3), pp. 569575.
[12] Thwaites, B., 1949, Approximate Calculation of Laminar Boundary Layer,
Aeronaut. Q., 1(Part 3), pp. 245280.
[13] Dey, J., and Narasimha, R., 1990, An Extension of the Thwaites Method for
Calculation of Incompressible Laminar Boundary Layers, J. Indian Inst. Sci.,
70(1), pp. 111. Available at http://journal.library.iisc.ernet.in/archives/v9-79/
Vol.70(1-6)1990/Vol.70(1)1990/vol70(1)1990.pdf
[14] Rott, N., and Crabtree, L. F., 1952, Simplified Laminar Boundary-Layer Calculations for Bodies of Revolution and for Yawed Wings, J. Aeronaut. Sci.,
19(8), pp. 553565.
[15] Bergthorson, J. M., Sone, K., Mattner, T. W., Dimotakis, P. E., Goodwin, D.
G., and Meiron, D. I., 2005, Impinging Laminar Jets at Moderate Reynolds
Numbers and Separation Distances, Phys. Rev. E, 72(6), p. 066307.
[16] Mehta, R. D., and Bradshaw, P., 1979, Design Rules for Small Low Speed
Wind Tunnels, J. R. Aeronaut. Soc., 83(827), pp. 443449.
[17] Schlichting, H., 1979, Boundary-Layer Theory, McGraw-Hill, New York.
[18] Ozalp, A. A., and Umur, H., 2003, An Experimental Investigation of the Combined Effects of Surface Curvature and Streamwise Pressure Gradients Both in
Laminar and Turbulent Flows, Heat Mass Transfer, 39(10), pp. 869876.
[19] Wazzan, A. R., Gazley, C., Jr., and Smith, A. M. O., 1979, Tollmien-Schlichting Waves and Transition: Heated and Adiabatic Wedge Flows With Application to Bodies of Revolution, Prog. Aerosp. Sci., 18, pp. 351392.
[20] Saric, W. S., 1994, Gortler Vortices, Annu. Rev. Fluid Mech., 26(1), pp.
379409.