You are on page 1of 34

5 Results

The first stage was a grid independence study of the mesh. However during this phase
instabilities and deviations from the experimental Coefficient of Pressure CP were
experienced despite close agreement to the Coefficient of Drag Cd.

Thus, a mesh

refinement study was performed in attempt to remove these instabilities and inaccuracies.
However the mesh refinement yielded no improvement in stabilisation or accuracy to
experimental data. This lead to an investigation into the y+ values and turbulence models,
which identified that the instabilities and inaccuracy where due a minute first prism height
associated with a y+ value of 1. It was found as the y+ value changed from two to ten, that
stabilisation of the solution was achieved without a reduction in timestep (see 4.1.2 Time
Stepping). In addition to the stabilisation of the solution, the use of the y+ value of ten, gave
CP data in very close agreement with experimental results for zero angle of incidence. Thus,
a y+ value of ten was used for the static angle of incidence and for the Unsteady Reynolds
Averaged Naiver Stokes Equations (URANS) simulations.
The results section of this thesis is presented in the order explained above, i.e.,
1. Grid independence
2. Mesh Refinement
3. Y+ and Turbulence Models
4. Angles of static pitch and yaw
5. URANS

5.1 Grid Independents results


The first phase of the CFD simulation was a grid independence study, conducted to
determine the most efficient mesh whilst still returning accurate results. Two Base meshes
where generated, which are identical except for the topology used for the appendages, i.e.,
1. An O grid on the leading edge and a Y grid on the trailing edge of the appendages,
(referred to in this study as an O-Y topology), see Figure 5-1.
2. A Y grid topology on both the leading and trailing edges of the appendages (referred
to in this study as a Y-Y topology), see Figure 5-2.

63

Figure 5-1 O-Y topology used on appendages

Poor quality
cell

Figure 5-2 Y-Y topology used on appendages

The quality of both the O-Y and Y-Y mesh types were almost identical, with both having a
minimum determinant of approximately 0.42, which exceeds the required value of 0.2.
In additional the minimum angle of both mesh types was 17.28, which is very close to the
preferred value of 18 and well above the required value of 9. As the YY topology was
refined, the angle on the leading edge of the appendages approached 180, resulting in a
poor quality mesh in this critical region. Thus, the YY topology is limited to relatively large
elements in this region, (Figure 5-3).

64

Poor quality
cells

Figure 5-3 Poor element on the leading edge of the Y-Y appendages

The key quality factors are summarised in Table 5-1

Table 5-1 Base mesh quality summary


Key Factor
Minimum volume
Minimum determinant
Minimum angle

Requirement
>0
>0.2
Preferably > 18, definitely > 9

Base mesh Value


1.22 x10-13
0.42
17.28

After mesh quality was checked, the O-Y topology was scaled globally (streamwise, normal
and azmuthal directions) to create the first mesh used in the grid independence study. This
mesh was then scaled globally in increments of 5x105 elements to create a series of 10
meshes ranging in size from 5x105 to 5x106 elements; this process was repeated for the Y-Y
topology. Plotted in Figure 5-4 is the percent error calculated against the experimental Cd
value for each mesh size. As seen from Figure 5-4 both the OY and YY mesh types are grid
independent at approximately 3x106 elements, which is a 50% reduction in mesh size when
compared to Ackermans results. It should also be noted that the OY topology becomes grid
independent with approximately 3.5% error to the experimental data, while the YY topology
becomes grid independent with approximately 1.5% error, however both are within 5% of
experimental results. Table 10-1 and Table 10-2 in Appendix 1; show the complete details
of this study. Due to the refinement restrictions of the YY topology discussed in the previous
section, the OY topology will be used for the remainder of the study.

65

Figure 5-4 Grid independence

Additional validation was performed by comparing CP along the centre line of the vessel.
Figure 5-5 shows a plot comparing both grid independent (3x106 element) mesh types. The
O-Y and Y-Y topology are in agreement with both experimental and Ackermans data until
X/L = 0.92, beyond this point both topologies deviate from the data, (see Figure 5-6). Initially
this was thought to be due to a lack of mesh density in this region, but as will be shown in
the preceding chapters, was a result of the y+ value rather than grid density.

66

Figure 5-5 CP profile along the topside centre line of the Suboff model

Figure 5-6 CP comparison of the OY and YY topology

67

Wake contours where investigated and it was found that the vortices structures exhibited a
large degree of irregularity, (see Figure 5-7) rather than a pair of contra rotating vortices that
was anticipated. In addition, the vortices do not travel the full length of the vessel, thus the
wake is not fully resolved, as shown in Figure 5-8. This supported the initial hypothesis that
the mesh lacked density in these regions.

Figure 5-7 Irregular flow structures at X/L = 0.4 (leeward of the sail)

Figure 5-8 Pressure on the hull and poor flow resolution

All grid independent simulations where performed at the recommended y+ value of less than
2, and hence a very small time step (7x10-5) was used in attempt to stabilise the solution and
ensure convergence, (see section 4.1.3). Despite a very small time step the solution still did
68

not completely converge (residuals were greater than 1x10-4), and the hydrodynamic forces
fluctuated throughout the simulation. However, this was not identified as an issue as these
fluctuations in the hydrodynamic forces had been experienced by Widjaja et al (2007) and
the y+ value used ( 2) was the recommended value. Due to the deviations in the CP when
compared to experimental data, and irregular flow structures, validation of the Coefficient of
Friction (CF) and wake contours were not performed.

5.2 Refined Mesh


As mentioned previously, poor CP results and the failure to fully resolve the flow structures
was initially thought to be due to a lack of grid density in the wake regions. To overcome
these issue a mesh was created where significant refinement was performed in the regions
leeward of the sail extending to the outlet. This included increasing the elements in the
streamwise, normal and azmuthual directions by a 150% as shown in Figure 5-9 and Figure
5-10. The simulation was performed with y+ value of 2 and a timestep of 7x10-4 s.

Regions of low grid density

Figure 5-9 O-Y initial grid independent mesh (2 900 000 elements)
69

Regions of refined density

Figure 5-10 Refined mesh in the wakefield (5 500 000 elements)

Figure 5-11 and Figure 5-12 show the refined mesh following the poor results of initial OY
mesh very closely, this refutes the original hypothesis that the deviation from experimental
data of the CP in the stern region is due to a lack of grid density.

70

Figure 5-11 CP comparison of the initial and refined mesh types

Figure 5-12 CP comparison of the inital and the refined mesh

71

In addition, the Coefficient of Friction in the x direction (CFX) was also checked, which
showed that the CFX for both the refined and the initial study were fairly similar and again
both were in agreement with experimental data until X/L > 0.92, where they deviated from
the experimental results.

Figure 5-13 CFX comparisons

The refinement of the mesh failed to fully resolve CP and CF in the stern region; however
wake contours were compared to determine if there was any improvement in wake
resolution with the refined mesh. A wake survey was conducted at X/L = 0.978 (propeller
plan) at the following:

r/R = 0.25; and

r/R = 0.5 as shown in Figure 5-14.


r/R = 0.5

r/R = 0.25

Figure 5-14 Wake profiles taken at X/L of 0.978 and r/R of 0.25

72

Before wake contours can be shown, an explanation of the sign convention used in this
study is shown in Figure 5-15. However some experimental data for wake contours has
rotated the sigh convention by 90 to that shown below, this is noted below the figure for the
required wake contours.

270

90

180
Figure 5-15 Sign convention used

Shown in Figure 5-16 Figure 5-17 is a plot of the non-dimensionalised velocity in the x and y
directions respectively. In Figure 5-16 it is shown that the velocity in the x direction between
the appendages (45 and 135) is much higher than the experimental data, this shows that
the cross flow and horse shoe vortices systems generated from the appendage tips and
junctions are not being captured as the vortices would reduce the velocity of the flow in this
region.

73

Figure 5-16 VX/V at 0.25r/R mesh comparison

Note: the sign convention changes for the above plot by 90, i.e. Top of sub of model is 90
In the y direction the flow between the appendages seems to be irregular when compared to
experimental results, this indicates instabilities in the solution, these instabilities are also
seen in Figure 5-18 as the flow structures are asymmetrical.

Figure 5-17 Vy/V mesh comparisons

Note: the sign convention changes for the above plot by 90, i.e. Top of sub of model is 90

74

Figure 5-18 Irregular vortices appearing aft of the sail

Increased mesh density in the wake region had no effect in improving the hydrodynamic
coefficients of the model.

Additionally, no significant improvement was found in the

representation of the flow structures in the propeller plane. As seen from Figure 5-19 the
wake contours are represented poorly leeward of the sail as the anticipated horse shoe
vortices dissipate rapidly. The results from this study confirmed that the poor quality solution
and instabilities were not due to a lack of grid density.

75

Figure 5-19 Refined mesh still does not resolve wake

5.3 Turbulence models and y+


As the refinement of the mesh yielded no improvement in solution, an investigation into the
effects of y+ and turbulence models was performed. Table 5-2 summarise the details of the
study showing that the k- and SST turbulence models were applied for a range of y+
values.
Table 5-2 Turbulence model and Y+ effects summary table
Turbulence
Model

Mesh Size
(No.

Y+

Elements)

First prism

Time step

Drag

height (mm)

(s)

(n)

EXP

SST

2 776 420

0.009

K-

2 776 420

0.009

SST

2 776 420

10

0.09

K-

2 776 420

10

SST

2 610 260

K-

Cd

%
Error

115.5

1.22E-03

119.6

1.26E-03

3.52%

0.007

115.6

1.22E-03

0.08%

0.09

0.007

114.7

1.21E-03

0.66%

20

0.18

0.007

109.0

1.15E-03

5.66%

2 610 260

20

0.18

0.007

112.9

1.19E-03

2.21%

SST

2 070 240

50

0.45

0.007

103.8

1.10E-03

10.16%

K-

2 070 240

50

0.45

0.007

113.8

1.20E-03

1.48%

.00007

76

The parameter y+ is the dimensionless distance from the wall to the first node and is
checked using ANSYS CFX Post. However, as y+ is a function of the shear stress, the y+
will vary as the pressure gradients cause changes in velocity. Figure 5-20 and Figure 5-21
show the y+ variations along the Suboff model with maximum values of 1.7 and 14
respectively, although they have been referred to as y+ 2, and y+ 10 in this study.

Figure 5-20 Maximum y+ of 1.7

Figure 5-21 maximum y+ of 14

It was discussed in section 4.1.2 Time Stepping, that as the solution ran for y+ values of 2,
the initial physical time step of 0.007 s had to be reduced to 7x10-5 s to stabiles the solution.

77

Shown in Figure 5-22, at iteration 50 the time step was reduced to 7x10-5, the residuals
stabilised and reduced in magnitude, however they still did not achieve the convergence
criteria of 1x10-4 reduction in residuals.
Reduction in timestep

Figure 5-22 Residuals reduced from a reduction in physical time step

Figure 5-23 shows that despite reducing the time step there is still a significant fluctuation in
the hydrodynamic forces. Plotted in Figure 5-23 is the force in the x direction showing there
is a fluctuation of the approximately 0.5 n. This indicates the solution is not stable and may
be erroneous, as shown earlier in this chapter.

78

Figure 5-23 Fluctuations of the hydrodynamic force - Drag

It was found for y+ values of ten or greater the residuals were stable and decreased in
magnitude with the initial user defined time step of 0.007, (Figure 5-24). In addition, the
hydrodynamic forces completely stabilised as shown in Figure 5-25, with the maximum
hydrodynamic drag fluctuations in the x direction, not exceeding 0.1N.
This indicates that determining the most appropriate time step for various y+ values is
critical. This study has not investigated the effects of time stepping further; however as the
time step can have significant influence on the solution it is be recommended as an area of
future work. The most effective method found for determining the appropriate time step was
to start with a large initial time step, then to refine it as the simulation progresses to achieve
the desired level of stability.

79

Figure 5-24 Residuals for y+ values of 10

Figure 5-25 Hydrodynamic drag in the x direction

Seen in Figure 5-26 as the y+ value approaches ten a significant improvement in


accuracy is achieved for the SST model, then as the y+ value increases beyond this
point the, SST model increases in error. Tu et al (2008) and Versteeg et al (2008)
both explain that as y+ becomes greater than 5, the first node from the wall surface
moves from the viscous sub layer and enters the turbulent region, thus failing to
capture turbulence in that region, which is contradictory to the results shown below
where the most accurate results is at y+ 10 . As expected the k- models error is
fairly consistent over the range of y+ values. Using the k- model ASNSY_CFX

80

failed to generate a solution for a y+ value of 2, thus showing that at a y+ value of 2


the first node height is too small to allow for the generation of a wall function.

Figure 5-26 Effects of y+ on accuracy of Cd

As Tu et al and Versteeg et al recommend a y+ of 1 to fully resolve the turbulent regions of


the flow, an investigation into the wake profiles was performed. As shown in the previous
sections poor agreement with experimental hydrodynamic coefficients is coupled with poor
resolution of the vortices structures associated with the flow, thus both are investigated.
Shown in Figure 5-27 the SST model with a y+ value of 10 has a CP that is in close
agreement with the experimental CP. Additionally as the SST models y+ value increases, the
divergence from the experimental CP increases. The k- model is also in close agreement
with experimental CP data and has little effect with the variance of the y+ values.

81

Figure 5-27 CP comparison for the SST and k- turbulence models with varying y+
values

Analysis of the shear stresses on the surface of the Suboff can give an indication of how well
the viscous sub layer is resolved. Shown in Figure 5-28 is a plot of CFX, along the centreline
of the Suboff model. Windward of X/L = 0.85, all the models are in reasonable agreement
with experimental data, with the SST y+ 50 deviating slightly. Both the k- and SST models,
at y+ values of 10, are in the best agreement with the experimental data, however beyond
X/L = 0.85 all models deviate. Additionally, shown in Figure 5-29 the SST y+ 50 and all k-
models have a smooth curve, while the SST models with y+ values of 2 and 10 have
fluctuating curves, this is indicative of the reduced y+ values capturing the shear stress
fluctuations near the surface.

82

Figure 5-28 Coefficient of friction over the Suboff model

Figure 5-29 CFx plot over the stern region of the Suboff model

Upon analysis of the wake contours developed in the propeller plane, it is shown in Figure
5-30 that again the SST model with a y+ value of 10 is in excellent agreement with the
experimental data. However, the SST model using the recommended y+ value of 2 predicts
a much higher dimensionless velocity over the entire propeller plane, indicating this model is
failing to capture the horse shoe and junction vortices imposed on the free stream velocity.
83

As expected, the SST models with large y+ values and all k- models fail to resolve the wake
due the large node spacing at the wall and the use of wall functions respectively.

Figure 5-30 Vx/V at 0.25 r/R at X/L = 0.978

Note: the sign convention changes for the above plot by 90, i.e. Top of sub of model is 90
Shown in Figure 5-31 is the velocity in the y direction (Vy) made dimensionless to the free
stream velocity, V. At angles of approximately 45 and 135 (between appendages) the SST
model with a y+ value of 2 has a Vy value 100% larger than the experimental results. Again
this shows the poor resolution of the flow structures associated with this flow type. The SST
model with a y+ value of 10 is the most accurate, which is in close agreement with the
experimental data, additionally the accuracy of both the SST and k- models decreases with
increases in y+ values.

84

Figure 5-31 Vy/V at 0.25 r/R at x/l = 0.978 and turbulence model comparison

Note: the sign convention changes for the above plot by 90, i.e. Top of sub of model is 90
Shown in Figure 5-33, Figure 5-34 and Figure 5-35 is a comparison of the wake profiles at
various locations along the Suboff model as depicted in Figure 5-32

X/L = 0.2

X/L = 0.25

X/L = 0.9

Figure 5-32 Locations of wake contours


Figure 5-33 shows the wake contour at X/L = 0.2, slightly fwd of the sail, here VX is
significantly reduced due to a high pressure system established on the leading edge of the
85

sail, and then increases to the free stream velocity above the sail. Surprisingly all turbulence
models and y+ values are in close agreement with each other.

Figure 5-33 Wake contour comparison at X/L = 0.2

X/L = 0.25 is located the tip mid-span of the sail, (Figure 5-34), a low pressure systems
causes the velocity to slightly exceed that of the free stream, this is a result of the pressure
gradients mid-span of the sail (along the sail tip) generating tip vortices. As shown the
velocity only slightly exceeds that of the free stream and will continue to accelerate as it
travels along the appendage.

At this location the appendage tip vortices are only just

beginning to generate, thus the results from the different turbulence models are similar.

Figure 5-34 Wake contour comparison at X/L = 0.25


86

X/L= 0.9 is located on the tip of the stern appendages, which is in the wake of the sail.
Shown in Figure 5-35 is a large degree of turbulent flow generated from the tip and junctions
of the stern appendages. It is shown that at the junction where the wake of sail meets the
free stream velocity there is an increase in velocity beyond the free stream velocity; this is
due to the tip vortices generated from the sail accelerating the turbulent flow from the stern
appendages. Additionally Figure 5-35 shows that for both k- and SST models with high y+
values, the reduction in free stream velocity is only minimal.

This indicates that these

models are not fully resolving the flow structures generated from the sail. The k- model with
a y+ value of 10 shows a larger reduction in the streamwise velocity than both the models
using a high y+ value. The SST model with a y+ value of ten resolves the largest reduction
in streamwise velocity, resolving the flow structures generated from the sail.

Figure 5-35 Wake contour comparison at X/L = 0.9

This chapter has shown that the SST turbulence model with a y+ value of 10 predicts the
hydrodynamic coefficients with close agreement to experimental data; additionally the wake
and velocity contours resolved are also in agreement with experimental data. The increased
y+ value also allows for stabilisation of the solution and an increased physical time step.
This may seem in contradiction to Veersteg et al and Tu et al recommendation that a y+
value of 1 is required for an SST model to resolve the complete turbulent region. However, it
is only the magnitude of the y+ value that is in dispute as this study found that beyond a y+
87

value of ten there is strong evidence to show the turbulent regions are not being fully
resolved, thus the first node from the wall has moved from the viscous to turbulent region at
y+ values greater than 10. It was also shown that for high Reynolds numbers such as used
in this study, that a y+ value of 1 requires a minute first prism height, which requires a very
small timestep to stabilise the flow and minimise discristaction errors.

5.4 Angle of attack


Angle of incidence where investigated at 10 static pitch using the SST model with a y+
value of 10.

As shown in Table 5-3, both pitch and yaw are in good agreement with

experimental results

Table 5-3 Angle of incidence summary table


CD
EXP
9.06E-04
Ackerman 8.58E-04
Current 9.35E-04

CD
EXP
1.11E-03
Ackerman 1.15E-03
Current 1.16E-03

Error
5.17%
3.21%

Error
3.83%
4.36%

10 Yaw
Side Force Coefficient
6.04E-03
6.09E-03
5.80E-03

10 Pitch
CL Coefficient
3.65E-03
3.55E-03
3.38E-03

Error
0.82%
4.01%

Yaw Coefficient
2.03E-03
2.22E-03
2.22E-03

Error
9.28%
9.12%

Error

Pitching Coefficient Error


1.56E-03
-2.79%
1.62E-03
4.34%
-7.36%
1.62E-03
3.94%

Pitch (diving)
CP is also compared along the centrelines circumferentially at 0 and 180 (top and bottom
respectively) of the model, it is shown from Figure 5-36 there is a large pressure difference
in the bow of Suboff for the 0 and 180 lines, however along the body the CP curves come
together until X/L = 0.7 where again they cross, however this time opposite to the bow. This
shows there are higher pressure regions on the fwd bottom and aft top sections of the
model, thus causing a pitching moment.

88

Figure 5-36 CP at 0 and 180 for 10 static pitch (dive)

The high pressure system shown at X/L less than 2, in Figure 5-36 indicates the stagnation
point at the bow is being fully resolved, additionally Figure 5-37 and Figure 5-38 show that
excellent resolution of the contra rotating vortices, generated from the appendages and
cross flows is achieved. Of special interest is the amount of turbulent flow in propeller
domain

Figure 5-37 y+ and vorticity - 10 pitch

89

Vortices generated at the


appendage tip and junction

Figure 5-38 Tip vortices and appendage junction vortices - 10 pitch

Shown in Figure 5-39, and Figure 5-40, the flow over the propeller will vary as the propeller
rotates from 0 to 360. From 0 to 180 (top) the variance in velocity is not significant,
however from 180 to 360 (bottom) along the top surface the difference in streamwise flow
from r/R = 0.2 to r/R = 0.5 is significant and would cause large variance in propeller loadings
as the propeller is rotated 360. This is supported by the velocity profile shown Figure 5-40.

Figure 5-39 Vx/V at 0.25r/R in the propeller plane ( X/L = 0.978)


Note: the sign convention changes for the above plot by 90, i.e. Top of sub of model is 90

90

180

Figure 5-40 Velocity profile at the propeller plane (X/L = 0.978) Pitch 10

Yaw
Figure 5-41 shows a CP plot circumferentially around the hull at increments of 90. It is
shown that there is a significant pressure difference at 90 and 270 for X/L values less than
0.3. The pressure gradient is minimal until X/L reaches 0.2, where the CP curves cross and
a pressure difference exists between X/L values 0.7 and 0.92, thus generating a yaw
moment.

Figure 5-42 shows the effect of the vortices generated from the sail, as the

adverse pressure gradient generated around the stern appendages reduces the pressure
much more significantly on the bottom surface than on the top surface exposed to the sail
vortices.

91

Figure 5-41 Cp plots circumferentially around the Suboff model

Figure 5-42 Cp curves over the stern region of the Suboff model

92

Shown in Figure 5-43 is the stagnation point that generates the pressure difference in the
bow region shown in Figure 5-41. It also shows that on the leeward side (270) of the Suboff
the pressure is higher than at the top and bottom. This is due to the cross flow vortices
generated from the hull, (Figure 5-44). The vortices are generated as the flow is accelerated
over the top and bottom of the Suboff, and then recirculates in the leeward side of the vessel
creating a high pressure system.

Figure 5-43 Stagnation point on the leading edge of the Suboff

Figure 5-44 Hull cross flow vortices creating a high pressure system on the leeward
side of the model
As was shown for pitch, the flow into the propeller plane under a yaw angle of attack is
greatly disturbed and irregular. Figure 5-44, Figure 5-45 and Figure 5-46 shows that on the
windward side of the vessel, the flow at r/R = 0.25 is affected by the vortices generated by
the appendages, while at r/R = 0.5 the flow is not affected by the vortices and is relatively
close to the free stream velocity, while on the leeward side the flow for both r/R =0.2 and 0.5
93

is highly disturbed by the appendage vortices. As the propeller rotates from through 0 to
360, significantly differently flow types will be experienced, (Figure 5-47).

Figure 5-45 Variance in Vx/V through the propeller plan yaw 10

Figure 5-46 Variance in Vy/V through the propeller domain - yaw 10

94

Note: the sign convention changes for the above plots by 90, i.e. Top of sub of model is
270

Figure 5-47 Irregular flow field in the propeller plane Yaw 10 (X/L =0.978)

5.5 Unsteady flow


An initial study into the applicability of using Unsteady Reynolds Averaged Naiver Stokes
(URANS) equations was performed.

URANS apply the time derivative in the RANS

equations and thus are a form of transient runs. The results of this study are summarised in
Table 5-4, Table 5-5 and Table 5-6.

Table 5-4 RANS and URANS comparison at zero AOA


Zero AOA
CD
Error
EXP 1.22E-03
RANS 1.22E-03
0.08%
URANS 1.22E-03
0.16%

95

Table 5-5 RANS and URANS comparisons at 10 yaw


CD
Error
EXP 9.06E-04
RANS 9.35E-04 3.21%
URANS 9.38E-04 3.55%

10 Yaw
Side Force Coefficient
6.04E-03
5.80E-03
5.77E-03

Error %
-4.01%
-4.49%

Yaw Coefficient
2.03E-03
2.22E-03
2.21E-03

Error %
9.12%
8.95%

Table 5-6 RANS and URANS Comparisons at 10 pitch (dive)


CD
Error
EXP 1.11E-03
RANS 1.16E-03 4.36%
URANS 1.16E-03 4.42%

10 Pitch
Side Force Coefficient
3.65E-03
3.38E-03
3.38E-03

Error %
-7.36%
-7.35%

Yaw Coefficient
1.56E-03
1.62E-03
1.62E-03

Error %
3.94%
3.95%

As seen from the above tables there are negligible difference with the use of URANS.
Widjaja used URANS to simulate the non-appended Suboff model and also found the use of
URANS to yield no significant variance in the solution.

96

You might also like