ITTC – Recommended 7.

5 – 03
02 – 01
Procedures Page 1 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

CONTENTS

1 PURPOSE OF PROCEDURE

2 EXAMPLE FOR RANS CFD CODE

2.1 Geometry, Conditions, and Benchmark Data
2.2 Computational Grids
2.3 Verification and Validation of Integral Variable: Resistance
2.4 Verification and Validation of a Point Variable: Wave Profile
3 REFERENCES

INTERIM RECOMMENDED PROCEDURE

Prepared Approved

Resistance Committee of 22nd ITTC 22nd ITTC 1999

Date Date
 4.9 – 03
ITTC 1990
ITTC - Quality Manual 01 – 01
Page 2 of 12
th
19 pp 585 - 587 Effective Date Revision
GENERAL
97-01-20
Guidelines for Uncertainty Analysis of 00
Measurements

Uncertainty Analysis in CFD, Examples for Resistance and Flow

1 PURPOSE OF PROCEDURE dure 7.5-03-01-01 and the equation numbering

is contiguous with QM procedure 7.5-03-01-
Provide an example for the verification and 01.
validation methodology for a RANS CFD
Code and results for steady flor for a
cargo/container ship following the Quality 2.1 Geometry, Conditions, and Bench-
Manual procedures 7.5-03-01-01, “Uncertainty mark Data
Analysis in CFD, Uncertainty Assessment
Methodology” and 7.5-03-01-02, “Uncertainty The geometry is the Series 60
Analysis in CFD, Guidelines for RANS cargo/container ship. The Series 60 was used
Codes.” for two of the three test cases at the last inter-
national workshop on validation of ship hydro-
dynamics CFD codes (CFD Workshop Tokyo,
2 EXAMPLE FOR RANS CFD CODE 1994). The conditions for the calculations are
Froude number Fr = 0.316, Reynolds number
Example results of verification and valida- Re = 4.3x106, and zero sinkage and trim. These
tion are presented for a single CFD code and are the same conditions as the experiments,
for specified objectives, geometry, conditions, except the resistance and sinkage and trim
and available benchmark information. The tests, as explained next. The variables selected
CFD code is CFDSHIP-IOWA, which is a for verification and validation are resistance CT
general-purpose, multi-block, high perform- (integral variable) and wave profile ζ (point
ance computing (parallel), unsteady RANS variable).
code (Paterson et al, 1998; Wilson et al., 1998)
developed for computational ship hydrodynam- The benchmark data is provided by Toda et
ics. The RANS equations are solved using al. (1992), which was also the data used for the
higher-order upwind finite differences, PISO, Series 60 test cases at the CFD Workshop To-
k-ω turbulence model, and exact and approxi- kyo (1994). The data includes resistance and
mate treatments, respectively, of the kinematic sinkage and trim for a range of Fr for the
and dynamic free-surface boundary conditions. model free condition (i.e., free to sink and
The objectives are to demonstrate the useful- trim); and wave profiles, near-field wave pat-
ness of the proposed verification and validation tern, and mean velocities and pressures at nu-
procedures and methodology and establish the merous stations from the bow to the stern and
levels of verification and validation of the near wake, all for Fr = (0.16, 0.316) and the
simulation results for an established bench- zero sinkage and trim model fixed condition.
mark for ship hydrodynamics CFD validation. The data also includes uncertainty estimates,
The section references and are to QM proce- which were recently confirmed/updated by
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 3 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

Longo and Stern (1999) closely following ment ratio rG = 2 . Clustering was used near
standard procedures (Coleman and Steele,
the bow and stern in the ξ−direction, at the hull
1999).
in the η-direction, and near the free surface in
The resistance is known to be larger for the ζ-direction. The y+ values for grids 1-4
free vs. fixed models. Data for the Series 60 were about 0.7, 1, 1.4, and 2, respectively.
indicates about an 8% increase in CT for the About twice the number of grid points in the η-
free vs. fixed condition over a range of Fr in- direction would be required to achieve y+ < 1.0
cluding Fr=0.316 (Ogiwara and Kajatani, for grids 1-4 (i.e., roughly 1,800,000 points on
1994). The Toda et al. (1992) resistance values the finest grid). With grid refinement ratio
were calibrated (i.e., reduced by 8%) for ef- rG = 2 , only grids 1 and 2 were generated.
fects of sinkage and trim for the present com- Grids 3 and 4 were obtained by removing
parisons. every other point from grids 1 and 2, respec-
tively (i.e., the grid spacing of grids 3 and 4 is
twice that of grids 1 and 2, respectively). Grids
2.2 Computational Grids 1 and 2 were generated by specifying the grid
spacing at the corners and number of points
Grid studies were conducted using four along the edges of the computational blocks.
grids (m=4), which enables two separate grid The faces of the computational blocks were
studies to be performed and compared. Grid smoothed using an elliptic solver after which
study 1 gives estimates for grid errors and un- the coordinates in the interior were obtained
certainties on grid 1 using the three finest grids using transfinite interpolation from the block
1-3 while grid study 2 gives estimates for grid faces. Grid 2 was generated from grid 1 by
errors and uncertainties on grid 2 using the increasing the grid spacing and decreasing the
three coarsest grids 2-4. The results for grid number of computational cells in each coordi-
study 1 are given in detail and the differences nate direction at the corners of the blocks by a
for grid study 2 are also mentioned. The grids factor rG. A comparison of the four grids at the
were generated using the commercial code free surface plane is shown in figure 1 along
GRIDGEN (Pointwise, Inc.) with considera- with computed wave elevation contours.
tion to topology; number of points and grid
refinement ratio rG; near-wall spacing and k-
ω turbulence model requirement that first point 2.3 Verification and Validation of Integral
should be at y+<1; bow and stern spacing; and Variable: Resistance
free-surface spacing.
Verification. Verification was performed
The topology is body-fitted, H-type, and with consideration to iterative and grid conver-
single block.. The sizes of grids 1 (finest) gence studies, i.e., δ SN = δ I + δ G and
through 4 (coarsest) are 101x26x16 = 42,016, 2
U SN = U I2 + U G2 .
144x36x22 = 114,048, 201x51x31 = 317,781,
and 287x78x43 = 876,211 and the grid refine-
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 4 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

Iterative convergence was assessed by ex-  ε 21G   0.07 x10 −3 

δ *
=  pG = 
amining iterative history of ship forces and L2 REG1
−   ( 2 )1.6 − 1 
norm of solution changes summed over all grid  G
r 1    (39)
−3
points. Figure 2 shows a portion of the iterative = 0.09 x10
history on grid 1. The portion shown represents ln(ε 32G ε 21G )
a computation started from a previous solution pG =
ln(rG )
and does not reflect the total iterative history. (40)
Solution change drops four orders of magni- ln(0.12 0.07)
= = 1.6
tude from an initial value of about 10-2 (not ln( 2 )
shown) to a final value of 10-6. The variation in
rGpG − 1
( 2 )1.6 − 1
CT is about 0.2%D over the last period of oscil- CG = = = 0.74 (41)
rG Gest − 1 ( 2 ) 2 − 1
p
lation (i.e., UI = 0.2%D). Iterative uncertainty
is estimated as half the range of the maximum where pest=pth=2 was used in equation (41).
and minimum values over the last two periods Uncertainty and error estimates are made next
of oscillation (see figure 2c). Iterative histories both considering CG as sufficiently less than or
for grids 2-4 show iterative uncertainties of greater than 1 and lacking confidence and CG
about 0.02, 0.03, and 0.01%D, respectively. as close to 1 and having confidence, as dis-
The level of iterative uncertainties for grids 2-4 cussed in Section 3.2.3.
are about two orders of magnitude less than the
grid error and uncertainty. The iterative uncer- For CG = 0.74 considered as sufficiently
tainty for grid 1 is one order of magnitude less than or greater than 1 and lacking confi-
smaller than the grid error. For all four grids dence, UG is estimated and not δG
the iteration errors and uncertainties are as- U G = CGδ RE
*
G1
+ (1 − CG )δ RE
*
G1 (42)
sumed to be negligible in comparison to the −3 −3 −3
= 0.07 x10 + 0.02 x10 = 0.09 x10
grid errors and uncertainties for all four solu-
tions (i.e., δI << δG and UI << UG such that δSN UG is 1.8% SG1 .
= δG and USN =UG). For CG = 0.74 considered close to 1 and having
confidence, both and δ G∗ and U GC are esti-
The results from the grid convergence mated
study for CT are summarized in tables 1 and 2. δ G*1 = CGδ RE
*
= 0.07 x10 −3 (43)
The solutions for CT indicate the converging G1

condition (i) of equation (16) with U GC = (1 − C G )δ RE

*
G1
= 0.02 x10 −3 (44)
RG = ε 21 / ε 32 =0.58. The first-order RE estimate
The corrected solution SC is defined with
δ RE [in equation (22)], order of accuracy pG
G1 S = SG1
[in equation (23)], and correction factor CG [in
S C = S G1 − δ G*1 = 4.96 x10 −3 (45)
equation (24a)] are
δ and U G are 1.4% and 0.4% SC, respec-
*
G1 C

tively. In both cases, the level of verification is

relatively small <2%.
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 5 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

the corrected comparison error, as discussed

Table 2 includes results for grid study 2, next.
which are similar to those for gird study 1, but
the values are larger by a factor of about 2, Second using SC, the corrected comparison
except SC which differs by only 0.4%. Also error is calculated from equation (34) as
shown in table 1 are CP and CF. CF comprises E C = D − S C = 5.42 x10 −3 − 4.96 x10 −3
(48)
about 70% of CT and also displays conver- = 0.46 x10 −3 = 8.5% D
gence; however, CP indicates oscillatory con-
The validation uncertainty is calculated from
vergence. Relatively small CG and oscillatory
equation (35) as
CP suggests that the solutions are relatively far
from the asymptotic range. Another reason for U VC = U S2C N + U D2 = 0.14 x10 −3 = 2.6% D (49)
oscillatory CP is that different flow phenomena where U SC N = U GC = 0.4%D. Here again,
may be resolved for the finer than the coarser
grids. EC > U VC such that the simulation results are
not validated. However, validation uncertainty
Validation. Validation is performed using UVC is relatively small and U S C N <<UD more
both the simulation prediction S and the cor- strongly suggests than was the case for E that
rected simulation prediction SC, as summarized EC is mostly due to modeling errors. There-
in table 3. First using S, the comparison error is
calculated from equation (30) with S = SG1 as fore modeling issues should/can be improved
to reduce EC and validate CT at the reduced
E = D − S = 5.42 x10 −3 − 5.03 x10 −3
(46) level UVC =2.6%D in comparison to equation
= 0.39 x10 −3 = 7.2% D
(47).
The validation uncertainty is calculated from
equation (33) as The results from grid study 2 are summa-
U V = U SN
2
+ U D2 = 0.17 x10 −3 = 3.1% D (47) rized in table 4. The results are similar to those
where USN=UG =1.7%D and UD=2.5%D. for grid study 1, but E and EC are smaller and
Comparison error E >UV such that the simula- UV and UVC are larger.
tion results are not validated. USN and UD are of
similar order such that reduction in UV would
require reduction of UD and USN (e.g., use of 2.4 Verification and Validation of a Point
finer grids for USN). E is positive, i.e., the Variable: Wave Profile
simulation under predicts the data. The trends
shown in table 1 suggest Cp too small. Pre- Verification. Verification for the wave
sumably modeling errors such as resolution of profile was conducted as per that described for
the wave field and inclusion of effects of sink- the resistance in Section 4.3 with the distinc-
age and trim can be addressed to reduce E and tion that a point variable is defined over a dis-
validate CT at UV=3.1%D; however, the case tribution of grid points. Interpolation of the
for this reasoning is stronger when considering wave profile on all grids onto a common dis-
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 6 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

tribution is required to compute solution dence, pointwise values for UG are estimated
changes. Since calculation of the comparison and not δG. Equation (26) is used to estimate
error E=D-S is required for validation, wave UG
profiles on grids 1-4 are interpolated onto the  ε 21 
distribution of the data. The same four grids U G = CG  p G 
 r G − 1
were used and, here again iteration errors and  G 
uncertainties were negligible in comparison to (53)
 ε 21G 
the grid errors and uncertainties for all four + (1 − C G ) p 
solutions, i.e., δI << δG and UI << UG such that  r G − 1
 G 
δSN = δG and USN =UG. For <CG>=0.60 considered close to 1 and hav-
ing confidence, pointwise values for both
RG at local maximums and minimums (i.e., δ G∗ and U GC are estimated using equations (25)
x/L = 0.1, 0.4, and 0.65 in figure 3a) and based
on L2 norm solution changes both show con- and (27)
vergence. The spatial order of accuracy for the  ε 
δ G* = CG  p 21 
G
(54)
 rG − 1 
wave profile was computed from the L2 norm 1 G

of solution changes

pG =
(
ln ε 32G / ε 21G
2 2
= 1.4
) (50)
 ε 21 
U G = (1 − CG ) p G 
 r G −1 
(55)
ln(rG )  G 
where < > is used to denote a profile-averaged Equation (10) is used to calculate SC at each
grid point
value and ε 2 denotes the L2 norm of solution
S C = SG1 − δ G*1 (56)
change over the N points in the region, 0 < x/L
<1 The results are summarized in table 5. The
1/ 2 level of verification is similar to that for CT
 N

ε 2 = ∑ ε i2  (51) with slightly higher values. Table 5 includes
 i =1  results for grid study 2, which are much closer
Correction factor is computed from equation to those for grid study 1 than was the case for
(24a) using order of accuracy pG in equation CT.
(50) and pGest = 2.0
−1
( 2 )1.4 − 1
pG
rG
CG = = = 0.60 (52) Validation. Validation of the wave profile
rG Gest − 1 ( 2 ) 2 − 1
p
is performed using both the simulation predic-
The estimates for order of accuracy and correc- tion S and the corrected simulation prediction
tion factor in equations (50) and (51) were SC . Profile-averaged values for both defini-
used to estimate grid error and uncertainty for tions of the comparison error, validation uncer-
the wave profile at each grid point. tainty, and simulation uncertainty are given in
table 6. Values are normalized with the maxi-
For <CG> = 0.60 considered as sufficiently mum value for the wave profile ζmax=0.014 and
less than or greater than 1 and lacking confi- the uncertainty in the data was reported to be
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 7 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

3.7%ζmax. E is nearly validated at about 5%. Wake and Uncertainty Assessment for
The trends are similar to those for CT, except DTMB Model 5512,” Proc. 25th ATTC,
there are smaller differences between the use Iowa City, IA, 24-25 September 1998.
of E and EC.
The point comparison error E=D-S is com- Ogiwara, S. and Kajitani, H., 1994, “Pressure
pared to validation uncertainty UV in figure 3b, Distribution on the Hull Surface of Series
while error EC=D-SC is compared to validation 60 (CB=0.60) Model,” Proceedings CFD
uncertainty UV in figure 3d. In the latter case, Workshop Tokyo, Vol. 1, pp. 350-358.
the validation uncertainty UV in figure 3d is
mostly due to UD. Much of the profile is vali- Paterson, E.G., Wilson, R.V., and Stern, F.,
dated. The largest errors are at the crests and 1998, “CFDSHIP-IOWA and Steady Flow
trough regions, i.e., bow, shoulder, and stern RANS Simulation of DTMB Model 5415,”
waves. 1st Symposium on Marine Applications of
Computational Fluid Dynamics, McLean,
The results from grid study 2 are summa- VA, 19-21 May.
rized in table 7 and included in Figure 3. The
results are similar to those for grid study 1, but Toda, Y., Stern, F., and Longo, J., 1992,
both E and EC and UV and UVC are larger. "Mean-Flow Measurements in the Bound-
ary Layer and Wake and Wave Field of a
Series 60 CB = .6 Model Ship - Part 1:
3 REFERENCES Froude Numbers .16 and .316," Journal of
Ship Research, Vol. 36, No. 4, pp. 360-377.
CFD Workshop Tokyo 1994, 1994, Proceed-
ings, Vol. 1 and 2, 1994, Ship Research In- Wilson, R., Paterson, E., and Stern, F., 1998
"Unsteady RANS CFD Method for Naval
stitute Ministry of Transport Ship & Ocean Combatant in Waves," Proc. 22nd ONR
Foundation. Symposium on Naval Hydro, Washington,
DC.
Longo, J. and Stern, F., “Resistance, Sinkage
and Trim, Wave Profile, and Nominal
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 8 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

Table 1 Grid convergence study for total CT, pressure CP, and frictional CF resistance (x10-3) for
Series 60.
Grid Grid 4 Grid 3 Grid 2 Grid 1 Data
101x26x16 144x36x22 201x51x31 287x71x43
CT 5.72 5.22 5.10 5.03 5.42
ε -8.7% -2.3% -1.3%
CP 1.95 1.63 1.64 1.61 CR = 2.00
ε -16.4% +0.6% -1.8%
CF 3.78 3.59 3.46 3.42 3.42
ε -5.0% -3.6% -1.2% ITTC
% of finer grid value.

Table 2. Verification of total resistance CT (x10-3) for Series 60.

Study RG pG CG UG δ G* U GC SC
1 0.57 1.6 0.74 1.8% 1.4% 0.4% 4.96
(grids 1-3)
2 0.24 4.1 3.1 3.9% 2.4% 1.6% 4.98
(grids 2-4)
%SG.

Table 3. Validation of total resistance for Series 60 – study 1 (grids 1-3).

E% UV% UD% USN%
E=D-S 7.2 3.1 2.5 1.7
EC=D-SC 8.5 2.6 2.5 0.4
%D.

Table 4. Validation of total resistance for Series 60 – study 2 (grids 2-4).

E% UV% UD% USN%
E=D-S 5.9 4.4 2.5 3.7
EC=D-SC 8.1 3.0 2.5 1.5
%D.
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 9 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

Table 5 Profile-averaged values from verification of wave profile for Series 60.
Study RG pG CG UG U GC
1 0.62 1.4 0.60 2.6% 1.0%
(grids 1-3)
2 0.64 1.3 0.57 3.6% 1.4%
(grids 2-4)
%ζmax .

Table 6. Profile-averaged values from validation

of wave profile for Series 60 – study 1 (grids 1-3).
E% UV% UD% USN%
E=D-S 5.2 4.5 3.7 2.6
EC=D-SC 5.5 3.8 3.7 1.0
%ζmax .

Table 7. Profile-averaged values from validation

of wave profile for Series 60 – study 2 (grids 2-4).
E% UV% UD% USN%
E=D-S 5.5 5.1 3.7 3.6
EC=D-SC 6.6 3.9 3.7 1.4
%ζmax .
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 10 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

(a) (b)
0.6 0.6
Y/L

Y/L
0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
X/L X/L

(c) (d)
0.6 0.6
Y/L

Y/L

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
X/L X/L

(e) (f)
0.6 0.6
Y/L

Y/L

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
X/L X/L

(g) (h)
0.6 0.6
Y/L

Y/L

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
X/L X/L

Figure 1. Grids and wave contours from verification and validation studies for Series 60: (a) and (b)
coarsest - grid 4; (c) and (d) grid 3; (e) and (f) grid 2; and (g) and (h) finest - grid 1.
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 11 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

10-2 0.008
U CF
-3 V CP
10 0.006
W CT
P
10-4
Residual

0.004

10-5
0.002
-6
10
(a) 0 (b)

10-7
0 10000 20000 30000 40000 0 10000 20000 30000 40000
Iteration Iteration

0.00505
-3
SU=5.037x10

0.00504

0.00503
CT

0.00502

0.00501
(c) SL=5.013x10
-3

0.005
30000 32000 34000
Iteration
Figure 2. Iteration history for Series 60 on grid 1: (a) solution change, (b) ship forces - CF, CP, and
CT and (c) magnified view of total resistance CT over last two periods of oscillation.
 ITTC – Recommended 7.5 – 03
02 – 01
Procedures Page 12 of 12
CFD, Resistance and Flow
Effective Date Revision
Uncertainty Analysis in CFD 1999 00
Examples for Resistance and Flow

0.01 Grid 1 (287x71x43)

Grid 2 (201x51x31)
Grid 3 (144x36x22)
Grid 4 (101x26x16)
Toda et al. (1992)
ζ/L

-0.01 (a)

0 0.25 0.5 0.75 1

x/L

0.2 E=D-S 0.2 E=D-S

+UV +UV
-UV -UV
0.1 0.1

0 0
E

-0.1 -0.1

(b) (c)
-0.2 -0.2

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

x/L x/L

0.2 EC=D-SC 0.2 EC=D-SC

+UV +UV
-UV -UV
0.1 0.1
EC

EC

0 0

-0.1 -0.1

(d) (e)
-0.2 -0.2

0 0.25 0.5 0.75 1 0 0.25 0.5 0.75 1

x/L x/L
Figure 3. Wave profile for Series 60: (a) grid study; (b) and (d) validation using grids 2-4; and (c)
and (e) validation using grids 1-3.