jspd170034 rp1

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/322698005
A Study on the Statistical Calibration of the Holtrop and Mennen Approximate

Power Prediction Method for Full Hull Form, Low Froude Number Vessels
Article in Journal of Ship Production and Design · January 2018

DOI: 10.5957/JSPD.170034
CITATIONS READS
18 7,754
2 authors:
Lampros Nikolopoulos Evangelos Boulougouris

University of Strathclyde University of Strathclyde
19 PUBLICATIONS 129 CITATIONS 236 PUBLICATIONS 2,559 CITATIONS
SEE PROFILE SEE PROFILE
All content following this page was uploaded by Lampros Nikolopoulos on 07 February 2018.
The user has requested enhancement of the downloaded file.

Journal of Ship Production and Design, Vol. 00, No. 0, Month 2018, pp. 1–28
http://dx.doi.org/10.5957/JSPD.170034
A Study on the Statistical Calibration of the Holtrop and Mennen

Approximate Power Prediction Method for Full Hull Form, Low Froude
Number Vessels
Lampros Nikolopoulos* and Evangelos Boulougouris†
*Department of Naval Architecture, Ocean and Marine Engineering, University of Strathclyde, Glasgow, Scotland
†
Maritime Safety Research Centre, Department of Naval Architecture, Ocean and Marine Engineering, University of Strathclyde, Glasgow, Scotland
The article herein presents a statistical calibration study of the approximate power
method of Holtrop and Mennen focused on adapting the method to vessels char-
acterized as “full” hull forms and low design and operating speeds and, thus, low
Froude numbers. The fitting of the method is done by adjusting the constants, co-
efficients, and components of the method’s equations by a systematic variation
process controlled by genetic algorithms. The database that the method is calibrated
against is consisted by model test results from modern (built between 2010 and 2016)
bulk carriers and tankers, the KVLCC2, and the method follows a multistage approach,
calibrating first the model for the prediction of total resistance and applying the self-
propulsion equations afterward. The uncertainty of the new improved method is
assessed and modeled with a nonlinear regression equation to enable the use of the
calibrated method in the early ship design and optimization process.
Keywords: Propulsion; model testing; powering; estimation; design (vessels); bulk

carriers; tankers
1. Introduction computational fluid dynamic solvers that solve Reynolds Averaged

Navier Stokes equations over the hull form in finite volume ap-
One of the most important aspects of the design and study of proaches (Schneekluth & Bertram 1998; Specialist Committee on
vessels and ocean structures is the prediction of resistance and, thus, CFD 2014). Although originally the computational cost was pe-
powering requirements for a range of operational speeds. Given the nalizing its application in early design stages, the advances in
constant societal and legislative pressure for the reduction of the computing hardware and software allowed the integration of
carbon footprint of shipping and maritime activities, the accurate Computational Fluid Dynamics (CFD) in the early ship hull form
and reliable powering prediction from an early stage and tightly design and optimization (Specialist Committee on CFD 2014). Such
integrated within the process itself as the respective optimization methods, however, can be used in applications where variables such
studies is imperative. Although towing tank model testing has been as the principal particulars have been fixed or have a small variance,
since early times the most reliable method for power predictions, with the focus being on the variation in local geometry and topo-
because of its high cost and demand for a fixed and given geometry logical characteristics that have a significant effect on local flow
it can be used only in the basic design stage when the design phenomena (Specialist Committee on CFD 2014). However, when
parameters related to the vessel’s hull geometry are fixed and serves the application in question is a global optimization study, where there
as a final validation and benchmark to be used afterward in the is a large variance in principal particulars (vessel’s dimensions) and
conclusion of the shipbuilding process during sea trials. On the structural and cargo arrangements, there is an apparent need for a
other hand, the last decade has seen the exponential growth of large number of optimization variants and, thus, evaluations (usually
in the order of thousands). The application of CFD in such cases is
Manuscript received by JSPD Committee December 5, 2017; accepted De- impractical as the optimization algorithm will eventually require
cember 28, 2017. hundreds of thousands computational hours even in high-performance
MONTH 2018 2158-2866/18/nnnnnn-0001$00.00/0 JOURNAL OF SHIP PRODUCTION AND DESIGN 1

Table 1 Particulars of vessels examined
Principal particular VSL01 VSL02 VSL03 VSL04 VSL05 VSL06 VSL07
Vessel type KVLCC2 VLCC Newcastlemax Capesize Capesize Ultramax Ultramax

Lwl (m) 335 322 298.61 291 292 198 200
Lbp (m) 334 328 294 286 288 195 195
B (m) 61 60 50 45 45 32.26 32
Draft (m) 20.8 21.6 18.5 18.3 18.15 12.9 11.3
Displacement (m3) 343,176.4 333,410.3 202,174.2 68,864 61,000
Cb 0.8098 0.7989 0.837 0.845 0.8595 0.8486 0.86
LCB (%) 0.035 fwd 0.03188 0.02368 0.0175 0.0162 0.0159 0.0285
Bulbous bow transverse area (m2) 100 123.9 100 100 0 10 63.053
Center of bulbous bow transverse area (m) 7.5 11 7.5 7.5 6 5 6
Cm 0.998 0.998 0.998 0.997 0.9981 0.9981 0.9953
Transom transverse area (m2) 30 30 30 30 30 30 30
Cstern 0 0 0 0 0 0 0
Wetted surface (m2) 29,629.27 28,226.2 20,959.7 10,196.8 9706
Cp 0.8114 0.8005 0.8538 0.8538 0.8538 0.8538 0.864
computers; therefore, empirical or statistical methods are better The resistance decomposition and hull form considerations
suited. The most prominent of these is the approximate power taken and the large number of input variables make the Holtrop
prediction method by Holtrop and Mennen (1982) together with methodology an ideal candidate for an initial or early stage too
its revision (Holtrop 1984). Although this methodology pro- (IMO Level 1 stage) for predicting the powering requirements of a
vides sufficient accuracy, the statistical sample of the hull forms given hull. For applications where the range of design parameters
on which it is based dates back to the 1970s and 1980s. Such hulls, is large1 and subject to change, the methodology can provide good
although roughly similar, have some distinct deviations from accuracy, correct trends, and variant ranking in terms of resistance.
modern commercial vessels. In the article presented herein, the However, for reasons of clarity, the following restrictions should
authors attempt to make a calibration of the Holtrop and Mennen be noted:
methodology via a systematic variation with the use of genetic
1) The Holtrop methodology is based on the regression analysis
algorithms. The calibration is based on a statistical sample of
of 334 ship models but within a range of dimension ratios
model test results of low Froude number (Fn) full hull forms (with
(L/B, B/T, T/L, Cb, Cp, Cwp, etc.) (Holtrop 1984). Although
Cb greater than 0.7) of modern commercial bulk carriers and
the range of applicability is significant and all typical vessel
tankers, and its focus is on the integration of the new coefficients
ratios are covered sufficiently, in cases of new designs where
in a holistic methodology for the optimization of large bulk
their ratios are close to the margins of the ones studied by
carriers. The uncertainty of the new coefficients is also taken into
Holtrop, a drop in accuracy must be taken into account.
account based on the sea trial results of an expanded statistical
2) For nonconventional hull forms and all vessels having hull
sample.
geometries significantly different from those in the original
database used, the accuracy of subject methodology is lim-
2. General ited. In addition, for vessels fitted with hydrodynamic energy
improvement devices (commonly referred to as energy-
The Holtrop and Mennen method (Holtrop & Mennen 1982;
saving devices) that significantly affect various flow phe-
Holtrop 1984) is currently considered as one of the most accurate
nomena at the stern of such vessels, the prediction error is
and efficient methods for the estimation of the resistance and
increased as such cases are not included in the original da-
propulsion power requirements of conventional monohull vessels at
tabase used by Holtrop and Mennen.
the initial stages of design. It is an empirical method consisting of
equations for the various resistance components that derive from the Following the development of the methodology and a second
statistical analysis and regression of a database with a large number publication from Holtrop (1984) on its updated coefficients, its
of model test results. It was developed in the early 1970s by Jan use is widespread both in a plethora of initial design applications
Holtrop and Frits Mennen when they were working in the MARIN and in ship operation and performance simulation modules
research center and towing tank. They focused on developing a (for some recent examples, see Cichowicz et al. 2015; Lu et al.
modern way to carry out data analysis with a focus on the ex- 2015; Mao et al. 2016; Vettor & Guedes Soares 2016). In-
trapolation of model tests. Initially, the method was designed for terestingly though, despite the aforementioned and the develop-
internal purposes so MARIN could make more accurate predictions. ment of other statistics-based methods for ship propulsion power
The two started with system analysis and then reanalyzed model prediction, the literature available at the time of writing lacks
tests and full-scale trials. The target was to have a component-wise
prediction method that would show the difference between model
and full-scale tests to serve the extrapolation of model experiments 1
Such as in the concept and preliminary ship design stages where the principal
(Holtrop-Mennen founders 2010, MARIN Report). particulars of the vessel are not fixed.
2 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

Fig. 1–11 Distribution of the EHP difference (%) over different speeds—laden condition
studies on the systematic calibration of the Holtrop and Mennen vessels. The vessels collected are existing vessels or those
methodology based on statistical samples from model tests, sea under construction of full hull form that represent modern ship
trials, or CFD results. design trends. Their principal characteristics which were used
The study presented herein aims to assess the deviation of the as input for the Holtrop powering prediction are presented in
Holtrop and Mennen prediction when compared with model test Table 1. T1
results, in Section 3, for a database of full hull form and low Fn From Table 1, we can observe that the vessel model test database
modern existing vessels. Afterward, in Section 4, the various con- that serves as the calibration basis is made of full hull forms of bulk
stants and coefficients used in the Holtrop and Mennen method- carriers and tankers including also vessels that were able to transit
ology are adjusted through a multistage approach for both bare hull the old Panama Canal (PANAMAX beam dimension), resulting in
resistance and propulsion power prediction. The error of the new adjustments of their length-to-beam ratio.
methodology is modeled with a nonlinear regression formula and In Figs. 1–11, the percentage of difference between the measured F1
presented in Section 5, and last, the results and future work are effective horsepower (EHP) during the model tests of the afore-
discussed in Section 6. mentioned vessel cases and the predicted EHP according to Holtrop’s
method is depicted. This difference is calculated over the entire range
of speeds available in the model tests to have a larger number of points
3. Vessel model test database (speed, power) that will be used for the calibration studies, and to
assess the effect of speed on the prediction accuracy.
To assess the level of accuracy of the methodology in question, For proper interpretation of the aforementioned figures, it is
a database was built comprising model test results of different assumed that a positive (þ) difference percentage corresponds to an
MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 3

Fig. 1–11 (Continued )
overestimation of the resistance by the Holtrop and Mennen overestimation linearly decreases up to a Fn in the range of
method, whereas a negative () difference corresponds to a re- 0.12–0.17 (depending on each vessel case) where there is a
spective underestimation. From the results, the following can be transition to underestimation of the resistance by the Holtrop
noted. and Mennen method, whereas in even higher speeds this can
be up to maximum 5%. The trend of the linear increase in the
underestimation percentage by increasing speed is still very
3.1. For the design and scantling conditions strong and evident as in the previous cases.
3) For the KVLCC2 case: The Vessel 01 graph corresponding
A very high sensitivity and nearly linear correlation to the op-
to the distribution of the EHP difference of the KVLCC2
erating speed are evident, resulting in an increased underestimation
vessel between the Holtrop and Mennen prediction and its
of resistance by increasing the Fn.
model test results is very interesting to comment. At very
1) For 5 of 11 cases (45.5%), the Holtrop and Mennen method low Fn of approximately 0.002, the Holtrop and Mennen
underestimates the resistance for the entire speed range, method underestimates the resistance, whereas at 0.04 it
ranging from 1% to 16%. marginally overestimates it. The overestimation percentage
2) For the other 5 cases, in low Fn, the Holtrop and Mennen increases by increasing the Fn up to a maximum of 15% at
method overestimates the resistance by maximum 10% in the 0.055. From this Fn, the overestimation percentage decreases
lowest respective speed, but by increasing Fn the relative up to 0.13 where it is practically zero. When compared with the

Fig. 12–17 Distribution of the EHP difference (%) over different speeds—ballast condition
other vessel cases in this study, this peculiar behavior at low pressure, and wave resistance. Unfortunately, the other vessel
speeds up to 0.06 Fn, according to the authors’ understanding, cases studied herein did not include tests at such low Fn during
can be attributed to the overestimation of various resistance their model testing, to observe the behavior of different hulls
components such as the influence of bulbous bow, viscous and geometries at low Fn.

Fig. 18–27 Shaft power prediction error (%) distribution over different speeds—laden condition
3.2. Ballast condition 3.3. Laden condition

In contrast to the laden conditions, in the ballast condition 1) The strong correlation to the Fn is still evident. However, for
for all vessels, a significant underestimation of a higher rela- the case of SHP, it is nonlinear (in contrast to the case of EHP),
tive magnitude is evident. Similarly to theladen condition, the illustrating a logarithmic trend for VSL03, 04, 05, 06 and 07,
underestimation percentage increases almost linearly by the in- whereas for VSL02 it is additionally fluctuating.
crease in the Fn from about 20% at the lower range up to about 50% 2) For 4 of 10 total cases (40%), there is an underestimation of
for the high range. The only nonlinear increase is observed for the SHP for the entire Fn range which increases with in-
VSL04 and VSL05 and at high Fn for VSL03 where a slight pa- creasing Fn. The underestimation ranges from 1% at lower
rabola is observed. The steepness of the curve is vessel dependent, speeds to maximum 10% at higher speeds (15 knots).
but for all cases similarities can be observed. 3) For 6 cases (60%), there is an overestimation of maximum 17%
Similarly to the EHP (and thus resistance) difference mentioned at low Fn, which decreases with increasing Fn up to the range of
F4 earlier, from Figs. 12–17, 18–27, and 28–34 one can see the dis- 0.135–0.175, from which point there is an underestimation of the
tribution of the error between the Holtrop and Mennen methodology SHP increasing with the increase in the Fn.
and model test prediction for the shaft horsepower (SHP) in the
laden and ballast conditions, respectively. 3.4. Ballast condition
The assumption for positive and negative differences is the same
as that in the case of EHP. From the aforementioned figures, the In the ballast condition, for all cases the same underprediction
following interesting result can be drawn. found in the EHP estimation was also evident here but on a smaller

magnitude ranging from 10% at small Fn up to 40% for high Fn optimization are summarized in Table 2 for the laden condition and T2
(close to 0.175). The trend of increasing underprediction by in- Table 3 for the ballast condition, respectively. T3
creasing speed is the same as that in the case of EHP.
From the aforementioned analysis, we can herein consider that 2) Optimization target
the Holtrop and Mennen methodology is accurate for EHP pre- For each vessel case, the difference between the Holtrop pre-
diction but lacks accuracy in SHP prediction, and EHP and SHP diction and the respective model test result for each speed run (based
prediction (off design condition) in ballast conditions. Furthermore, on the model tests) is calculated and its minimization is set as the
the evident correlation to Fn (thus vessel speed) and inaccuracy target of the optimization run. The difference is dependent on the
according to the authors’ perception can be attributed to the dif- calibration stage and can either be the difference in the effective
ferent flow development and phenomena which cannot be captured power or the difference in the propulsion factors, namely, thrust
by an empirical method. The trends are consistent, thus underlining deduction, relative rotative efficiency, and wake field fraction,
the result robustness. respectively. It should be pointed out that the absolute value
(i.e., unsigned magnitude) of the differences was used instead of the
signed difference because using the latter might lead to results with
4. A multistage evolutionary approach larger overall errors if the positive and negative differences cancel
each other.
To be able to control the process of calibration and maintain the
same ranking and composition of the total power requirements, a 3) Design engine
multistage approach was adopted for the calibration, with each stage
F5 being treated as a typical optimization problem as depicted in Fig. The design engine applied can be either the NSGA II (Non-
35. The constraints are part of the optimization, penalizing the dominating Sorting Genetic Algorithm) (Deb et al. 2002) or the
nonfeasible solutions and probably pushing for more feasible ones. MOSA (Multi-Objective Simulation Annealing) (Ulungu et al.
As in every optimization problem, the following components 1999) algorithms. These are employed in the CAE software
were considered: CAESES (ex-Friendship Framework) where the simulation and
variation are programmed.
1) Optimization variables
4) Constraints
The variables of this problem were the selected constants and
coefficients for each formula of each resistance component as The only constraint set was an upper limit on the optimization
formulated in the original Holtrop publication. The methodology target, which is rejecting effectively combinations of the co-
was parametrically programmed and the constants became the efficients, resulting in an average EHP and average SHP difference
variables of the new problem. The selected variables for the (depending on the calibration stage) greater than 15%. This was

Fig. 28–34 Shaft power prediction error (%) distribution over different speeds—ballast condition

the initial solution/state is critical for evolutionary algorithms such
as the NSGA II, contributing to more efficient convergence and a
better final solution.
The results were sorted and a combination was selected from
the lowest average absolute difference basis. The resulting
values improved the effective power curve prediction error, as
can be seen from Figs. 34–47 and 48–54. In all the graphs, the F6
Fig. 35 Generalized optimization problem workflow
gray, triangle marker curve represents the error distribution after
calibration and the black, circle marker curve represents the error
distribution corresponding to the original coefficients of the
method.
4.1.1. Interpretation of the results. When looking at the results

for both laden and ballast conditions, it is evident that the simulation
had an obvious positive effect as in all vessel cases the prediction
was improved, given the fact that the all EHP differences were
reduced, whereas all the trends in terms of deviation over speed
were accurately maintained. In fact, if one observes the form of the
curves, it is evident that the calibrated curves are in fact “trans-
lations” of the original publication curves toward the X axis, which
was after all the objective of the herein presented study.
4.1.2. Laden conditions. From Figs. 37–47, the distribution of F7
Fig. 36 Stages and tools used for each test the EHP difference over different speeds for the laden conditions is
depicted with the following observations:
1) For 1 of 11 cases (VSL03—design condition), the translation
of the deviation distribution was such that for the entire Fn
done to apply a restriction and because this is an evolutionary range, there is an overestimation of the resistance which,
algorithm, the latter is pushed to converge to the target. however, decreases (on a third power basis) with increasing
speed following the same trend as the equivalent curve before
the calibration. The maximum overestimation is 10%, but
4.1. Stage 1: bare hull resistance calibration by increasing the Fn it is reduced to almost 0.5%.
2) Another interesting case would be that of VSL05 design
The first calibration stage is for bare hull resistance and power condition, where for low speeds there is an underestimation
requirement. The parameters that underwent systematic variation which decreases by increasing curves starting from 1% at low
were from 1 to 49 for laden conditions and 1–54 for ballast con- Fn and up to a minimum of 2% overestimation and then
ditions depicted in Tables 2 and 3 with the same range of variance. sharply passes again to the underestimating region with the
The variation engine employed was the NSGA II algorithm in the underestimation increasing sharply by increasing Fn up to a
CAESES platform with a chosen set of 10 generations of variants maximum of 13%.
with each generation having 100 variants, thus resulting in 1000 3) For 8 of 11 cases, the effect of the apparent translation of the
variants, in other words 1000 combinations of the 49 parameters. deviation distribution curve is that there is an overestimation
For this stage, the optimization target was set to be the absolute at low speeds of maximum 10% which decreases steeply
value of the percentage of difference in EHP (kW) between the by increasing the speed up to a transition speed within the
model test estimation and each “Holtrop variant” prediction, with range of 0.125–0.175 Fn, from which point there is a transit to
minimum being the desired merit. This was done for all 7 vessel the underestimation area with a steep increase in underesti-
cases in all speed-power points of each model test, resulting in 111 mation by increasing the speed up to a maximum of 20%
differences calculated at points referred to as “calibration points.” underestimation.
Furthermore, for the ease of selection process, the average value of
the differences of the calibration points per vessel was calculated as
it can assist in the sorting of the best variants during postprocessing.
4.1.3. Ballast conditions. From Figs. 48–54, the distribution of F8
Within this spirit, the average value for the methodology in total (as
the EHP difference over different speeds for the ballast conditions is
average of the averages of absolute differences) was calculated to
depicted with the following observations:
serve also as a constraint. This constraint imposed was set as having
not more than 15% of average deviations. 1) For 1 of 7 cases (VSL05), the deviation distribution is located
The optimization was on a chain basis for three optimization at the underestimating region for all Fn, starting from a
loops, meaning that the best variant from the first run was set as the minimum of 2% at low Froude numbers which increases on a
initial solution for the second and the best variant from the second power rate by increasing speed and up to a maximum of 20%
run was set as the initial solution for the third run. This was done as at high Froude numbers.

Table 2 Holtrop constants used as variables in optimization problem and their respective value range—laden condition
Calibration variables—laden (design and scantling) condition
Value in
original Value Value
Holtrop calibration calibration
No. Constant in Holtrop formula publication minimum maximum
Wave making resistance
1:07961
T
L-1 c1 ¼ 2; 223; 105pc7 3:78613 p pð90 iE Þ1:37565 2,223,105 2,200,000 2,300,000
B
1:07961
T
L-2 c1 ¼ 2; 223; 105pc7 3:78613 p pð90 iE Þ1:37565 3.78613 2.0 4.70
B
1:07961
T
L-3 c1 ¼ 2; 223; 105pc7 3:78613 p pð90 iE Þ1:37565 1.07961 0.6 1.30
B
1:07961
T
L-4 c1 ¼ 2; 223; 105pc7 3:78613 p pð90 iE Þ1:37565 1.37565 2.0 0.80
B
pffiffiffiffiffi
L-5 c2 ¼ expð1:89p c3 Þ 1.89 3.0 0.9
0:56pABT 1:5
L-6 c3 ¼ pffiffiffiffiffiffiffiffi 0.56 0.20 0.90
BpT pð0:31p ABT þ TF hB Þ
0:56pABT 1:5
L-7 c3 ¼ pffiffiffiffiffiffiffiffi 0.31 0.01 0.80
0:56pABT 1:5
L-8 c3 ¼ pffiffiffiffiffiffiffiffi 1.5 1.20 3.0
AT
L-9 c5 ¼ 1 0:8p 0.8 0.20 1.50
BpT pCM
L3
L-10 c15 ¼ 1:69385 þ c15new p 1.69835 2.0 1.10
V
L3
L-11 c15 ¼ 1:69385 þ c15new * c15new ¼ 0 c15new ¼ 1 c15new ¼ 2
V 1
L ∇3 B
L-12 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 0.0140407 0.005 0.10
T L1 L
L ∇3 B
L-13 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 1.75254 1.10 2.50
T L1 L
L ∇3 B
L-14 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 4.79323 3.70 5.70
T L L
L-15 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 1.20 4.0
L-16 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 0.1 0.50 0.04
L-17 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 4.0 1.20
L-18 c16 ¼ 1:73014 0:7067pCP þ c16Cnew pCP 2 þ c16Dnew pCP 3 1.73014 1.20 2.50
L-19 c16 ¼ 1:73014 0:7067pCP þ c16Cnew pCP 2 þ c16Dnew pCP 3 0.7067 0.20 1.20
L-20 c16 ¼ 1:73014 0:7067pCP þ c16C new *C P 2 þ c16Dnew pCP 3 c16Cnew ¼ 0 c16Cnew ¼ 2 c16Cnew ¼ 2
L-21 c16 ¼ 1:73014 0:7067pCP þ c16Cnew pCP 2 þ c16Dnew *C P 3 c16Dnew ¼ 0 c16Cnew ¼ 2 c16Cnew ¼ 2
L
L-22 λ ¼ 1:446pCp 0:03p þ cnew 1.446 0.60 2.0
B
L
L-23 λ ¼ 1:446pCp 0:03p þ cnew 0.03 0.01 0.10
B
L
L-24 λ ¼ 1:446pCp 0:03p þ cnew 0 2 2
B
(continued )

Table 2 Continued
Value in
L-25 d ¼ 0:9 0.9 1.50 0.30
L-26 Rw ¼ c1 pc2 pc5 p∇pρpgpexpfm1 pFnd þ m2 pcosðλpFn2 Þg 2 0 1
Frictional resistance—form factor
L-27 c12 ¼ ðT =LÞ0:2228446 0.2228446 0.05 0.80

( 0:92497 )
B 0:521448 0:6906
L-28 ð1þk1 Þ¼ c13 p 0:93 þc12 p pð0:95CP Þ pð1 CP þ0:0225pLCBÞ 0.92497 0.40 1.50
LR
( 0:92497 )
B 0:521448 0:6906
L-29 ð1 þ k1 Þ ¼ c13 p 0:93 þ c12 p pð0:95 CP Þ pð1 CP þ 0:0225pLCBÞ 0.521448 0.90 0.10
LR
( 0:92497 )
B
L-30 ð1 þ k1 Þ ¼ c13 p 0:93 þ c12 p pð0:95 CP Þ0:521448 pð1 CP þ 0:0225pLCBÞ0:6906 0.6906 0.6906 1.50
LR
L-31 c13 ¼ 1 þ 0:003pC stern 0 3 3
Resistance due to bulbous bow

pffiffiffiffiffiffiffiffi
0:56p ABT
L-32 PB ¼ 0.56 0.10 1.0
TF 1:5phB
0:11pexpð3pPB 2 ÞpFni3 pABT 1:5 pρpg
L-33 RB ¼ 0.11 0.05 0.30
1 þ Fni2
L-34 RB ¼ 3 5.0 1.50
1 þ Fni2
L-35 RB ¼ 2 1.2 4
1 þ Fni2
L-36 RB ¼ 3 1.5 4.0
1 þ Fni2
L-37 RB ¼ 1.5 1.20 3.0
1 þ Fni2
L-38 RB ¼ 2 1.50 4.50
1 þ Fni2
v
L-39 Fni ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffi 2 1.50 4.50
gpðTF hB 0:25p ABT Þ þ 0:15pv2
Resistance due to transom immersion
L-40 c6 ¼ 0:2pð1 0:2pFnT Þ 0.2 0.01 0.30
Model ship correlation

rffiffiffiffiffiffiffi
L
L-41 CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew 0.006 0.001 0.10
7:5
L
7:5
L
7:5
L
7:5
(continued )

Table 2 Continued
Value in
L-45 L 7.5 3.0 10.0
CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew
7:5
L
L-46 CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew 4 2.0 6.0
7:5
L
L-47 CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew - - -
7:5
L
L-48 CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew cAnew ¼ 0 cAnew ¼ 2 cAnew ¼ 2
7:5
L
L-49 CA ¼ 0:006pðL þ 100Þ0:16 0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew cBnew ¼ 0 cBnew ¼ 2 cBnew ¼ 2
7:5
Propulsion factors
AE
L-50 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.9922 - -
A0
AE
L-51 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.05908 - -
A0
AE
L-52 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.07424 - -
A0
AE
L-53 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.0225 - -
A0
L B D2
L-54 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.001979 0.0001 0.1
Bpð1 CP Þ L BpT
L B D2
L-55 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 1.0585 0.4 1.9
Bpð1 CP Þ L BpT
L B D2
L-56 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.00524 0.02 0.001
Bpð1 CP Þ L BpT
L B D2
L-57 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.1418 0.3 0.01
Bpð1 CP Þ L BpT

L-58 L CV 0.0661875 - -
wHOLTROP ¼ c9 pCV p p 0:0661875 þ 1:21756pc11 p
TA ð1 CP1 Þ
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
B 0:09726 0:11434
þ 0:24558p þ þ 0:75pCStern pCV þ 0:002pCStern
Lpð1 CP1 Þ 0:95 CP 0:95 CB

L-59 L CV 1.21756 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
L-60 L CV B 0.24558 - -
wHOLTROP ¼ c9 pCV p p 0:0661875 þ 1:21756pc11 p þ 0:24558p
TA ð1 CP1 Þ Lpð1 CP1 Þ
0:09726 0:11434
þ þ 0:75pCStern pCV þ 0:002pCStern
0:95 CP 0:95 CB
L-61 L CV 0.09726 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
(continued )

Table 2 Continued
Value in

L-62 L CV 0.11434 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
3
L-63 TA 0.08333333 - -
c11 ¼ 0:0833333p þ 1:33333
D
3
TA
L-64 c11 ¼ 0:0833333p þ 1:33333 1.3333 - -
D
1:6 16
L-65 wSCHNEEKLUTH ¼ 0:5pCP p p 0.5 0.1 0.9
D L
1 þ 10 þ
T B
1:6 16
L-66 wSCHNEEKLUTH ¼ 0:5pCP p p 1.6 0.1 3
D L
1 þ 10 þ
T B
1:6 16 16 8 20
L-67 wSCHNEEKLUTH ¼ 0:5pCP p p
D L
1 þ 10 þ
T B
L-68 wKRUGER ¼ 0:75pCB 0:24 0.75 0.5 1.5
L-69 wKRUGER ¼ 0:75pCB 0:24 0.24 0.1 0.5
L-70 wHECKSCHER ¼ 0:7pCP 0:18 0.7 0.25 1.5
L-71 wHECKSCHER ¼ 0:7pCP 0:18 0.18 0.3 0.3
L-72 wTROOST ¼ 0:25 þ 2:5pðCB 0:6Þ2 0.25 0.1 0.4
L-73 wTROOST ¼ 0:25 þ 2:5pðCB 0:6Þ2 2.5 0.1 4
2) For the remaining 6 of 7 cases, because of the translation of however, at the nominal deviation, the “translation” observed in the
the error distribution curve, an overestimation at low speeds is deviation graphs and described earlier, it has improved from an
observed which is decreasing by increasing speed and at a average underestimation of 1.4% to an overestimation of 1.9%
transition speed of the region of 0.15 Fn changes to the which is preferable in ship design studies to have a safer design
underestimation region which is increasing by increasing margin for sizing the propulsion plant. For the laden conditions, one
speed up to a maximum of 20%. can also observe that despite an improvement in prediction in some
3) For all cases and because the same trends of the curves before vessels of the herein presented study, the database is very distinct
the calibration are kept, the correlation between deviation and (VSL02, VSL03, VSL04, and VSL05); for other vessels in the same
speed is very close to being linear and has a high steepness, database and for the same selected optimization variant, the pre-
which is in contrast to the laden condition which follows a diction error is higher when compared with the respective error
second to third power fit correlation. corresponding to the original Holtrop coefficients. This highlights a
sensitivity of the method which is expected, given the already low
From these observations, it can be deducted that the EHP cali-
level of prediction error (%) for the original Holtrop coefficients in
bration is successful and an accurate prediction is possible for early
the laden conditions.
ship design studies (especially in case systematic, global optimi-
In the ballast conditions, the average absolute deviation was
zation studies are to be conducted).
reduced from 31.8% to 9.78%, which makes it a considerable
improvement that constitutes the basis for the next Stage 2 cali-
T4 T5 4.1.4. Consolidated results. Tables 4 and 5 summarize the EHP bration for the delivered horse power (SHP).
difference between the Holtrop prediction and the model test results
before and after the calibration for laden and ballast conditions, 4.2. Stage 2: shaft power calibration
respectively. For the laden conditions (both design and scantling),
the improvement overall seems marginal as the deviation decreased After the calibration of the constants that are parts of the
from 5.7% to 4.7% in terms of absolute deviation. If one considers, equations that predict the bare hull resistance and, in turn, the

Table 3 Holtrop constants used as variables in optimization problem and their respective value range—ballast condition
Calibration variables—ballast condition
Value in
Wave making resistance
1:07961
T
B-1 c1 ¼ 2223105pc7 3:78613 p pð90 iE Þ1:37565 2,223,105 2,100,000 2,250,000
B
1:07961
T
B-2 c1 ¼ 2223105pc7 3:78613 p pð90 iE Þ1:37565 3.78613 3.6 3.8
B
1:07961
T
B-3 c1 ¼ 2223105pc7 3:78613 p pð90 iE Þ1:37565 1.07961 1.0 1.2
B
1:07961
T
B-4 c1 ¼ 2223105pc7 3:78613 p pð90 iE Þ1:37565 1.37565 1.5 1.0
B
pffiffiffiffiffi
B-5 c2 ¼ expð1:89p c3 Þ 1.89 2.5 1.5
1:5
0:56pABT
B-6 c3 ¼ pffiffiffiffiffiffiffiffi 0.56 0.40 0.60
0:56pABT 1:5
B-7 c3 ¼ pffiffiffiffiffiffiffiffi 0.31 0.20 0.40
0:56pABT 1:5
B-8 c3 ¼ pffiffiffiffiffiffiffiffi 1.5 1.2 3
AT
B-9 c5 ¼ 1 0:8p 0.8 0.7 0.9
BpT pCM
3
L
B-10 c15 ¼ 1:69385 þ c15new p 1.69835 1.80 1.50
V
L3
B-11 c15 ¼ 1:69385 þ c15new * c15new ¼ 0 c15new ¼ 1 c15new ¼ 3
V
1
L ∇ 3 B
B-12 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 0.0140407 0.01 0.02
T L L
1
L ∇ 3 B
B-13 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 1.75254 1.60 1.90
T L L
1
L ∇ 3 B
B-14 m1 ¼ 0:0140407p 1:75254p þ 4:79323p c16 4.79323 4.60 4.90
T L L
B-15 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 1.2 3
B-16 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 0.1 0.8 0.01
B-17 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 3.5 1.5
B-18 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 8.07981 4.0 12.0
B-19 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 13.8673 25.0 10.0
B-20 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 6.984388 3.0 11.0
B-21 c16 ¼ 8:07981pC P 13:8673pCP 2 þ 6:984388pCP 3 1 0.9 1.6
B-22 c16 ¼ 8:07981pCP 13:8673pC P 2 þ 6:984388pCP 3 2 1.8 2.9
B-23 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pC P 3 3 2.6 3.5
(continued )

Table 3 Continued
Value in
L
B-24 λ ¼ 1:446pCp 0:03p þ cnew 1.446 1.40 1.50
B
L
B-25 λ ¼ 1:446pCp 0:03p þ cnew 0.03 0.001 1.0
B
L
B-26 λ ¼ 1:446pCp 0:03p þ cnew 0 2.0 2.0
B
B-27 d ¼ 0:9 0.9 2.0 0.1
B-28 Rw ¼ c1 pc2 pc5 p∇pρpgpexpfm1 pFnd þ m2 pcosðλpFn2 Þg 2 3.0 1.50
Frictional resistance—form factor

2:078
T
B-29 c12 ¼ 48:20p 0:02 þ 0:479948 48.20 25.0 60.0
L
2:078
T
B-30 c12 ¼ 48:20p 0:02 þ 0:479948 2.078 0.50 4.0
L
2:078
T
B-31 c12 ¼ 48:20p 0:02 þ 0:479948 0.479948 0.35 0.60
L
( 0:92497
B
B-32 ð1 þ k1 Þ ¼ c13 p 0:93 þ c12 p pð0:95 0.92497 0.30 1.50
LR )
CP Þ0:521448 pð1 CP þ 0:0225pLCBÞ0:6906
( 0:92497
B-33 B 0.521448 3.0 1.50
ð1 þ k1 Þ ¼ c13 p 0:93 þ c12 p pð0:95
LR )
CP Þ0:521448 pð1 CP þ 0:0225pLCBÞ0:6906
( 0:92497
B
B-34 ð1 þ k1 Þ ¼ c13 p 0:93 þ c12 p pð0:95 0.6906 0.20 0.90
LR )
CP Þ0:521448 pð1 CP þ 0:0225pLCBÞ0:6906
B-35 c13 ¼ 1 þ 0:003pCstern 0.003 0.001 0.70
B-36 c13 ¼ 1 þ 0:003pC stern 0 0.10 2.0
Resistance due to bulbous bow

pffiffiffiffiffiffiffiffi
0:56p ABT
B-37 PB ¼ 0.56 0.10 0.90
TF 1:5phB
B-38 RB ¼ 0.11 0.01 0.30
1 þ Fni2
B-39 RB ¼ 3 5.0 1.0
1 þ Fni2
B-40 RB ¼ 2 3.0 1.50
1 þ Fni2
B-41 RB ¼ 3 1.80 4.0
1 þ Fni2
B-42 RB ¼ 1.5 1.2 3.0
1 þ Fni2
(continued )

Table 3 Continued
Value in
B-43 0:11pexpð3pPB 2 ÞpFni3 pABT 1:5 pρpg 2 1.60 2.60
RB ¼
1 þ Fni2
v
B-44 Fni ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffi 2 1.40 2.60
gpðTF hB 0:25p ABT Þ þ 0:15pv2
Resistance due to transom immersion
B-45 c6 ¼ 0:2pð1 0:2pFnT Þ 0.2 0.01 2.0
Model ship correlation
B-46 CA ¼ 0:006pðL þ 100Þ0:16 0.006 0.001 0.50
L
0:00205 þ 0:003p pCB 4 pc2 pð0:04 c4 Þ þ cAnew pFncBnew
7:5
B-47 CA ¼ 0:006pðL þ 100Þ0:16 0.16 0.50 0.01
L
7:5
B-48 CA ¼ 0:006pðL þ 100Þ0:16 0.00205 0.06 0.10
L
7:5
B-49 CA ¼ 0:006pðL þ 100Þ0:16 0.003 0.001 0.10
L
7:5
0:16
B-50 CA ¼ 0:006pðL þ 100Þ 7.5 6.0 11.0
L
7:5
B-51 CA ¼ 0:006pðL þ 100Þ0:16 0.04 - -
L
7:5
B-52 CA ¼ 0:006pðL þ 100Þ0:16 cAnew ¼ 0 cAnew ¼ 0:20 cAnew ¼ 1:20
L
7:5
B-53 CA ¼ 0:006pðL þ 100Þ0:16 cBnew ¼ 0 cBnew ¼ 1:0 cBnew ¼ 1:0
L
7:5
B-54 c4 ¼ c4new p TLF c4new ¼ 1 c4new ¼ 0:01 c4new ¼ 5
Propulsion factors
AE
B-53 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.9922 - -
A0
AE
B-54 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.05908 - -
A0
AE
B-55 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.07424 - -
A0
AE
B-56 ηR ¼ 0:9922 0:05908p þ 0:07424pðCP 0:0225plcbÞ 0.0225 - -
A0
L B D2
B-57 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.001979 0.0001 0.9
Bpð1 CP Þ L BpT
L B D2
B-58 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 1.0585 0.5 2.5
Bpð1 CP Þ L BpT
L B D2
B-59 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.00524 0.1 105
Bpð1 CP Þ L BpT
L B D2
B-60 tHOLTROP ¼ 0:001979p þ 1:0585p 0:00524 0:1418p 0.1418 0.0001 0.9
Bpð1 CP Þ L BpT
(continued )

Table 3 Continued
Value in

B-61 L CV 0.0661875 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
þ 0:24558p þ þ 0:75pCStern pCV
Lpð1 CP1 Þ 0:95 CP 0:95 CB
þ 0:002pCStern
B-62 1.21756 - -
L CV
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
þ 0:002pCStern
B-63 0.24558 - -
L CV
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
þ 0:002pCStern

B-64 L CV 0.09726 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
Lpð1 CP1 Þ 0:95 CP 0:95 CB
þ 0:002pCStern

B-65 L CV 0.11434 - -
TA ð1 CP1 Þ
B 0:09726 0:11434
þ 0:24558p þ
Lpð1 CP1 Þ 0:95 CP 0:95 CB
þ 0:75pCStern pCV þ 0:002pCStern
3
TA
B-66 c11 ¼ 0:0833333p þ 1:33333 0.08333333 - -
D
3
TA
B-67 c11 ¼ 0:0833333p þ 1:33333 1.3333 - -
D
1:6 16
B-68 wSCHNEEKLUTH ¼ 0:5pCP p p 0.5 0.1 0.9
D L
1þ 10 þ
T B
1:6 16
B-69 wSCHNEEKLUTH ¼ 0:5pCP p p 1.6 0.5 2.5
D L
1 þ 10 þ
T B
1:6 16
B-70 wSCHNEEKLUTH ¼ 0:5pCP p p 16 6 25
D L
1þ 10 þ
T B
B-71 wKRUGER ¼ 0:75pCB 0:24 0.75 0.4 0.9
B-72 wKRUGER ¼ 0:75pCB 0:24 0.24 0.05 0.5
B-73 wHECKSCHER ¼ 0:7pCP 0:18 0.7 0.2 0.9
B-74 wHECKSCHER ¼ 0:7pCP 0:18 0.18 0.40 0.01
B-75 wTROOST ¼ 0:25 þ 2:5pðCB 0:6Þ2 0.25 0.1 0.4
B-76 wTROOST ¼ 0:25 þ 2:5pðCB 0:6Þ2 2.5 0.1 4

effective power, Stage 2 of this calibration study includes the study 5) ηs is the shafting system efficiency, as per model test results
of the self-propulsion problem, which can be reduced to the fol- for each case.
lowing equation (1):
Because the EHP prediction is calibrated throughout Stage 1
runs, the calibration in Stage 2 focuses on the independent com-
1t ponents of equation (1) in respective calibration runs through which
SHP ¼ EHP p p ηR p η0 p ηs ; (1)
1w the shaft power is calculated. It should be noted that at this point of
the study, it is important to have a clear decomposition and database
where
with values for each component.
1) w is the wake fraction; For the prediction of the wake fraction and thrust deduction, in
2) t is the thrust deduction; the herein proposed methodology the average of the formulas in
3) ηR is the relative rotative efficiency; Tables 6 and 7 is considered. The reason for such an approach is that
4) η0 is the propeller open water efficiency, as per model test the Holtrop proposed formula exhibits high instability and infidelity
results for each case; and primarily for the wake fraction coefficient. This according to the
Fig. 37–47 EHP prediction error (%) distribution over different speeds before and after Stage 1 calibration—laden condition

authors’ opinion can be attributed to the different stern geometries rotative efficiency, respectively, based on the available experi-
and viscous phenomena of the database selected versus the original mental data. For this, the gradient-based TSEARCH (Hilleary 1966)
database of Holtrop and Mennen. algorithm is used which is also available in CAESES.
In the approach proposed herein, thrust deduction, relative ro- Based on the aforementioned approach and the range of pa-
tative efficiency, and wake fraction are estimated using the same rameters used as depicted in Tables 2 and 3, the results are presented
database as in Stage 1. Whereas in Stage 1 the solver used for the as follows.
optimization problem (NSGA II) was of an evolutionary nature and
on a multiobjective basis, here a single-objective optimization
problem is used and each of the coefficients (t, w, and ηR) is checked 4.2.1. Relative rotative efficiency. The relative rotative effi-
separately. The optimization target is the minimization of the av- ciency prediction by the Holtrop and Mennen empirical formula
erage deviation of wake fraction, thrust deduction, and relative illustrates an impressive accuracy for all vessel cases, both laden

Fig. 48–54 EHP prediction error (%) distribution over different speeds before and after Stage 1 calibration—ballast condition
and ballast conditions, and throughout the entire speed range when average being an overestimation of 0.89%, whereas for the
compared with the respective model test predictions. As can be ballast condition the deviation ranged from 4.17% overesti-
T8 observed from Table 8, the deviation for the laden condition ranged mation to 3.75% underestimation with the average being 0.75%
from 3.5% overestimation to 3% underestimation with the overestimation.

Table 4 EHP deviation from model tests before and after Stage 1 calibration—laden conditions
Stage 1 calibration results—laden conditions
Average error (%) EHP Average absolute error (%) Average error (%) Average absolute error (%)
Vessel after calibration EHP after calibration EHP original Holtrop coefficients EHP original Holtrop coefficients
VSL01—design condition 11.206 11.272 6.593 7.234
VSL02—scantling—low speed 0.584 0.835 3.826 3.826
VSL02—design—low speed 0.695 1.353 4.583 4.583
VSL03—scantling condition 5.025 5.183 1.514 2.992
VSL05—scantling Condition 2.361 4.678 2.553 3.852
Entire database 1.930 4.775 2.763 5.742
Table 5 EHP deviation from model tests before and after Stage 1 calibration—ballast conditions
Stage 1 calibration results—ballast conditions
Average error (%) Average absolute error (%) Average error (%)
Vessel EHP—NSGA11 to 885 EHP—NSGA11 to 885 EHP—original Holtrop coefficients
VSL02—ballast condition 4.269 9.604 28.725
VSL03—heavy ballast condition 7.765 9.863 27.989
VSL03—light ballast condition 2.298 6.086 31.310
Entire database 0.690 9.788 32.620
Table 6 Formulas used for the prediction of the wake fraction in the Table 8 Deviation of the relative rotative efficiency (%)
self-propulsion problem
EtaR average EtaR average
1:6 16
Schneekluth formula w ¼ 0:5pCP p p prediction absolute prediction
D L
1 þ 10 þ Case error (%) error (%)
T B
Kruger formula w ¼ 0:75pCB 0:24 VSL02—scantling 1.589 1.589
Heckscher formula w ¼ 0:7pCP 0:18 VSL02—design 2.215 2.215
Troost formula w ¼ 0:25 þ 2:5pðCB 0:6Þ2 VSL03—design condition 2.209 2.209
VSL03—scantling condition 2.645 2.645
VSL04—design condition 1.630 1.630
Table 7 Formulas used for the prediction of the thrust deduction in VSL05—design condition 0.334 0.407
the self-propulsion problem VSL05—scantling condition 0.926 0.926
Holtrop and Mennen formula
t ¼ 0:001979p
L VSL07—design condition 0.870 0.870
Bpð1 CP Þ
B
þ 1:0585p 0:00524
L
D2
0:1418p
BpT
Heckscher formula t ¼ 0:5pCP 0:12 Given this very good accuracy, it was not deemed necessary to
Danckwardt t ¼ 0:5pCB 0:15 proceed with any further calibration but to focus on the wake
fraction and thrust deduction coefficient calibration.

Table 9 Average error (%) for wake fraction and thrust deduction—laden conditions
T average error (%) T average error (%) W average error (%) W average error (%)
Vessel after Stage 2 calibration before calibration after Stage 2 calibration before calibration
VSL02—scantling 2.798 29.067 13.767 13.446
VSL02—design 1.587 25.615 16.775 9.485
VSL03—design 8.600 46.163 6.593 39.991
VSL03—scantling 5.775 52.220 15.076 50.596
VSL04—design 10.084 52.034 0.188 31.043
VSL04—scantling 1.164 44.271 14.337 49.499
VSL05—design 41.604 146.104 11.462 15.751
VSL05—scantling 44.794 159.600 20.648 3.727
VSL06—scantling 19.719 95.884 10.142 17.622
VSL07—design 1.545 70.011 5.772 38.136
Entire database 7.688 72.097 3.082 26.930
Table 10 Average error (%) for wake fraction and thrust deduction—ballast conditions
T average error (%) T average error (%) W average error (%) W average error (%)
Vessel after Stage 2 calibration before calibration after Stage 2 calibration before calibration
VSL02—ballast 13.495 3.487 18.019 26.774
VSL03—heavy ballast 7.275 12.200 3.165 2.488
VSL03—light ballast 9.193 7.217 0.483 7.538
VSL04—ballast 2.148 22.587 3.954 3.041
VSL05—ballast 50.219 89.055 5.442 7.208
VSL06—ballast 21.478 46.438 6.924 11.808
VSL07—ballast 1.839 22.773 9.181 7.234
Entire database 6.0058 28.112 2.081 6.506
phenomena and hence is very sensitive to each different hull form.

The calibration converged also here yielding a significant im-
provement as shown in Table 9 for the laden conditions and
Table 10 for the ballast conditions. The average error reduced rather
Fig. 55 The generic optimization problem in Stage 2 drastically from 72.1% to 7.69% for the laden conditions and from
26.93% to 3.08% for the ballast conditions.
4.2.2. Wake fraction. Given the sensitivity of the wake fraction
on scale effects, it should be noted that its prediction is challenging
and cannot be accurately achieved with empirical methods. Also, 4.2.4. Final SHP prediction. After each of the components of
for the case of model test prediction, the extrapolation and scaling to the self-propulsion equation is, respectively, calibrated as described
full-scale wake results are something of not high fidelity, whereas earlier, one can assess their effect (combined with the results of the
the full-scale CFD viscous calculation is a task requiring time and Stage 1/EHP calibration) on SHP.
computational resources. The effect is assessed herein in terms of both the prediction error
This is also reflected when we observe the deviation of the herein at different speeds and the final speed power curves before and after
predicted wake fraction (averaged from five different formulas) calibration for each vessel and loading condition.
when compared with the respective model test results (for full For the error distribution, the translation phenomenon of the
scale). The results of the single-objective optimization problem are curves toward lower percentages and overestimation areas observed
T9 presented in Table 9 for the laden conditions and Table 10 for the in the EHP calibration is also evident here both in laden and ballast
T10 ballast conditions (Fig. 55). It is evident that the prediction im- conditions. This results in overestimations of the required power in
proved significantly from an average absolute deviation of 26.93% many cases (especially at low speed areas), leading to a more
(overestimation) before the calibration to a 3.08% (un- conservative approach of the power prediction.
derestimation) equivalent after the calibration. For the ballast Another interesting observation for both laden and ballast
conditions, the average reduced from 6.51% to 2.08% un- conditions is that the trend of SHP difference distribution of the
derestimation after the calibration. original Holtrop coefficients is retained in most of the cases and is
very similar (with some minor distortions) to the respective EHP
distributions, indicating the strong dependency of the SHP to EHP.
4.2.3. Thrust deduction. Similar to the wake fraction, thrust This trend is increasing tend to underestimation by increasing
deduction is also a coefficient highly dependent on the local speed.

Fig. 56–65 SHP prediction error (%) distribution over different speeds before and after Stage 2 calibration—laden condition
F10 4.2.5. Laden condition. From Figs. 56–65, the SHP deviation constant at 5%, then drops rapidly, and transits to the
distribution over different speeds is depicted for the laden condi- underestimating region, increasing linearly and steeply by
tions. The following can be observed: increasing speed with a maximum underestimation of 30%.
4) For 1 out of 10 cases, there is only underestimation for the
1) For 5 of 10 cases, an overestimation of the SHP for the entire
all Fn having the same trend as with the original case
range of Fn is observed from a maximum 20% at low Fn range
however translated to a much lower level of error. This can
to up to 4% at the high range, indicating the same trend of
be seen from an underestimation of 1% at low Fn and up
increasing underestimation by increasing the Fn. The reason
to a maximum of 3% at higher. Interestingly, for this case
for the high overestimation at low speeds in this case is the
(VSL04 scantling condition), the difference from 12 to 14
translation of the original curve to lower overestimating re-
knots is almost constant at 1% and changes rapidly from
gions to attain an average close to zero deviation.
0.135 to 0.15 Fn with an almost linear increase by in-
2) For 4 of 10 cases, whereas at the lower Fn range there is an
creasing speed.
overestimation of up to 15%, at higher Fn this transcends to
the underestimation region at a transition Fn ranging from
0.125 to 0.175 depending on each vessel case, and the un-
derestimation at the highest speeds is maximum 15%. 4.2.6. Ballast condition. In Figs. 66–72, the SHP deviation F11
3) An interesting case out of the 5 cases mentioned earlier is distribution over different speeds is depicted for the ballast con-
VSL07 where up to 0.16 Fn the overestimation is practically ditions. The following can be observed:

1) For 1 of 7 cases (VSL05 ballast condition), the methodology In addition to the aforementioned, it should also be noted that
underestimates the required power for the entire Fn range, when looking at the average deviation (not absolute), this has
with the latter descending by increasing speed and having the been an average underestimation of 1.74% before the calibration
same trend as before the calibration. which changed to an average overestimation of 4.65%, which
2) For the remaining 6 of 7 cases, there is an overestimation leads to a safer margin for predictions at the preliminary ship
of the required power at low speeds of maximum 25% design stages.
which linearly decreases with increasing speeds, and at a Last, all the individual components have an improved accuracy;
Fn at the region 0.15 (depending on each vessel), there is a thus, the herein proposed calibrated methodology depicts in a more
transition to the underestimation region. The underestimation accurate way the sensitivities of the methodology in main di-
also increases linearly by increasing speed up to a maximum mensions and design characteristics for all the resistance sub-
of 15%. components, and is very useful for preliminary ship design studies.
Table 13 summarizes the finalized values chosen for the Holtrop T13
T11 T12 In Tables 11 (for the laden conditions) and 12 (for the ballast
constants in terms of resistance and the propulsion power prediction
conditions), the deviation of the predicted required delivered
formula following the two-stage optimization and calibration
horsepower (SHP) before the calibration (original Holtrop co-
process.
efficients) and after the calibration, when compared with the model
test prediction, is depicted. The average absolute error (%) sig-
nificantly reduced from 28.68% to 10.3% for the ballast conditions. 5. Uncertainty analysis for ship design applications
For the laden conditions, however, the average absolute error (%)
Following the systematic calibration and for a given final pre-
increased marginally from 4.59% to 7.49%. The reason for this is
diction error (%) per vessel per speed, the error was examined to be
the already small prediction error. By trying to improve such a small
modeled for corrections in new prediction applications. The error
error, the sensitivity is at a level that can improve the prediction
was decided to be modeled by means of a nonlinear regression
accuracy for some cases and deteriorate for others. A typical ex-
formula generated in the IBM SPSS software, for the database used
ample is VSL02 for which the prediction error decree increased
herein for calibration and for the coefficients deriving from the
from 0.95% to 5% while in the meantime and same combination of
Stage 2 calibration. This has been modeled in equations (2) and (3).
variables the prediction error of VSL02, for VSL04 decreased from
6.57% to 1.37%. Therefore, the authors advise not to attempt such Errorð%ÞLADEN ¼ 0:008 p LBP 8:096 þ 0:009 p B16:346
calibration studies in applications where the prediction error is
already very small because of inherent sensitivity and volatility þ 0:006 p Tm 34:137 þ 0:039 p CB 1100:918
issues of the method. 0:014 p WS4:877 (2)

Fig. 66–72 SHP prediction error (%) distribution over different speeds before and after Stage 2 calibration—ballast condition

Table 11 Average deviation of the delivered horse power (%) per vessel—laden conditions
SHP average error (%) SHP average absolute error (%) SHP average error (%) SHP average absolute error (%)
Vessel after Stage 2 calibration after Stage 2 calibration before Stage 2 calibration before Stage 2 calibration
Entire database 5.894 7.487 2.382 4.596
Table 12 Average deviation of the delivered horse power (%) per vessel—ballast conditions
SHP average error (%) SHP average absolute error (%) SHP average error (%) SHP average absolute error (%)
Vessel after Stage 2 calibration after Stage 2 calibration before Stage 2 calibration before Stage 2 calibration
VSL02—ballast—low speed 11.467 12.240 16.954 16.954
VSL03—heavy ballast condition 3.0225 7.154 25.178 25.178
VSL03—light ballast condition 0.3793 5.228 27.354 27.354
VSL04—ballast condition 8.577 8.843 30.424 30.424
Entire database 0.207 10.311 -28.689 28.689
Errorð%ÞBALLAST ¼ 1132:36 p LBP 0:055 95:054 p B0:275 4.7% for the laden and from 32.6% to 9.7% for the ballast condition.
þ 911:906 p Tm 0:080
þ 158:633 p CB 2:854 The SHP prediction through the self-propulsion equation used the
same model test database used in the EHP calibration. Finally, the
þ 0:256 p WS 0:457
; accuracy of the SHP improved from an error of 28.7% to 10.3% for
(3) the ballast condition, whereas for the laden conditions the error
marginally increased from 4.59% to 7.49% despite the 1% im-
where provement during the EHP calibration stage. This slight de-
LBP : Length between perpendiculars; terioration is attributed to the sensitivity of the method and the
B: Breadth (moulded); already very high level of accuracy (error level of about 5% is very
Tm : Midship draft; small for an empirical method).
CB : Block coefficient; and The uncertainty of the original and reproduced methodology has
WS: Wetted surface. been examined and statistically modeled by a nonlinear regression
analysis, for future use in power prediction during the preliminary
6. Summary, conclusions, and next steps stages of various ship design studies. Next steps in this study will be
verification with CFD of the final results, calibration based on
In this study, an extensive calibration of the constants and co- databases generated from automated hull variation, and subsequent
efficients of the Holtrop and Mennen empirical power prediction CFD calculation.
method is carried out based on a database of model test results of What the reader should bear in mind as a conclusion is that the
modern commercial vessels, and a useful analysis on its accuracy study presented herein does not propose an updated view of the
over different speeds and hull forms but only focusing on full hull established Holtrop and Mennen methodology, but a rationale of
forms of a low Fn. In general, the accuracy of the original method calibrating and adapting this method to specific applications
can be considered adequate; however, it decreasing constantly depending on the available data from the user. When having this
while the Fn increases. After the application of the herein de- mentality in mind during ship design, the naval architect can adapt
veloped, multistaged optimization approach for calibrating the the resistance prediction methodology on parent and similar vessels
methodologies the EHP prediction has been improved from 5.7% to and achieve a higher level of accuracy in the early stages.

Table 13 Values of Holtrop coefficients after calibration for bare hull resistance
Value in original Holtrop Value after final Value in original Holtrop Value after final
No. publication—laden calibration—laden No. publication—ballast calibration—ballast
L-1 2,223,105 2,242,064.546 B-1 2,223,105 2,207,306.02
L-2 3.78613 2.965053788 B-2 3.78613 2.369306477
L-3 1.07961 0.987251087 B-3 1.07961 0.965964752
L-4 1.37565 1.248029297 B-4 1.37565 1.426852827
L-5 1.89 1.585484092 B-5 1.89 1.865680934
L-6 0.56 0.888603037 B-6 0.56 0.310946822
L-7 0.31 0.679055924 B-7 0.31 0.599398947
L-8 1.5 0.679055924 B-8 1.5 2.660425727
L-9 0.8 0.311482414 B-9 0.8 0.866790265
L-10 1.69835 1.968221561 B-10 1.69835 1.492373541
L-11 c15new ¼ 0 1.215701534 B-11 c15new ¼ 0 0.311146715
L-12 0.0140407 0.022718547 B-12 0.0140407 0.01282916
L-13 1.75254 1.420567636 B-13 1.75254 1.28395819
L-14 4.79323 5.681261921 B-14 4.79323 4.854474708
L-15 2 2.580880446 B-15 2 2.202572671
L-16 0.1 0.147266651 B-16 0.1 0.405693141
L-17 2 2.988265812 B-17 2 1.654390784
L-18 1.73014 1.465613794 B-18 8.07981 7.892027161
L-19 0.7067 0.6426 B-19 13.8673 12.1206
L-20 c16Cnew ¼ 0 1.516899 B-20 6.984388 9.3225
L-21 c16Dnew ¼ 0 0.340063 B-21 1 1.3920
L-22 1.446 1.815469596 B-22 2 2.8830
L-23 0.03 0.023458457 B-23 3 3.1013
L-24 0 0.318760967 B-24 1.446 2.238521401
L-25 0.9 1.202760357 B-25 0.03 0.4431
L-26 2 2.374135958 B-26 0 0.0587
L-27 0.2228446 0.377695125 B-27 0.9 0.3324
L-28 0.92497 0.668810559 B-28 2 1.8767
L-29 0.521448 0.623591974 B-29 48.2 28.87945373
L-30 0.6906 0.766962692 B-30 2.078 3.323125048
L-31 0 2.825314717 B-31 0.479948 0.446070497
L-32 0.56 0.843195239 B-32 0.92497 0.613664454
L-33 0.11 0.095151446 B-33 0.521448 0.565916533
L-34 3 3.153574426 B-34 0.6906 0.532242313
L-35 2 3.251325246 B-35 0.003 0.01538851
L-36 3 2.339742123 B-36 0 1.952800793
L-37 1.5 1.458429847 B-37 0.56 0.354276341
L-38 2 2.337720302 B-38 0.11 0.067300832
L-39 2 3.939871824 B-39 3 3.752483406
L-40 0.2 0.223463188 B-40 2 1.800297551
L-41 0.006 0.069377752 B-41 3 3.084248112
L-42 0.16 0.417208057 B-42 1.5 1.358590066
L-43 0.00205 0.034246056 B-43 2 1.663599603
L-44 0.003 0.125382971 B-44 2 1.456360723
L-45 7.5 3.736263066 B-45 0.2 0.885375296
L-46 4 4.217013809 B-46 0.006 0.223290471
L-47 0.036289769 B-47 0.16 0.092530251
L-48 cAnew ¼ 0 2.948729686 B-48 0.00205 0.082350805
L-49 cBnew ¼ 0 0.013092241 B-49 0.003 0.095168963
L-50 0.9922 0.9922 B-50 7.5 6.146944381
L-51 0.05908 0.05908 B-51 0.04 0.478261311
L-52 0.07424 0.07424 B-52 cAnew ¼ 0 0.36520943
L-53 0.0225 0.0225 B-53 cBnew ¼ 0 0.673182269
L-54 0.001979 0.0001 B-54 c4new ¼ 1 3.186513619
L-55 1.0585 3 B-53 0.9922 0.9922
L-56 0.00524 0.107307217 B-54 0.05908 0.05908
L-57 0.1418 0.001 B-55 0.07424 0.07424
L-58 0.0661875 0.0661875 B-56 0.0225 0.0225
L-59 1.21756 1.21756 B-57 0.001979 0.00001
L-60 0.24558 0.24558 B-58 1.0585 1.403225943
(continued )

Table 13 Continued
Value in original Holtrop Value after final Value in original Holtrop Value after final
No. publication—laden calibration—laden No. publication—ballast calibration—ballast
L-61 0.09726 0.09726 B-59 0.00524 0.032286653
L-62 0.11434 0.11434 B-60 0.1418 0.01
L-63 0.08333333 0.08333333 B-61 0.0661875 0.0661875
L-64 1.3333 1.3333 B-62 1.21756 1.21756
L-65 0.5 0.432109375 B-63 0.24558 0.24558
L-66 1.6 1.325859375 B-64 0.09726 0.09726
L-67 16 15.72532144 B-65 0.11434 0.11434
L-68 0.75 0.662024983 B-66 0.08333333 0.08333333
L-69 0.24 0.176972156 B-67 1.3333 1.3333
L-70 0.7 0.316622043 B-68 0.5 0.5
L-71 0.18 0.101944784 B-69 1.6 1.6
L-72 0.25 0.25 B-70 16 16
L-73 2.5 2.5 B-71 0.75 0.820510639
B-72 0.24 0.102842033
B-73 0.7 0.431767617
B-74 0.18 0.048268475
B-75 0.25 0.252269833
B-76 2.5 0.1
References LU, R., TURAN, O., BOULOUGOURIS, E., BANKS, C., AND INCECIK, A. 2015 A semi-
empirical ship operational performance prediction model for voyage opti-
CICHOWICZ, J., THEOTOKATOS, G., AND VASSALOS, D. 2015 Dynamic energy mization towards energy efficient shipping, Journal of Ocean Engineering,
modelling for ship life-cycle performance assessment, Journal of Ocean 110, 18–28.
Engineering, 110, 49–61. MAO, W., RYCHLIK, I., WALLIN, J., AND STORHAUG, G. 2016 Statistical models
DEB, K., PRATAP, A., AGARWAJ, S., AND MEYARIVAN, T. 2002 A fast and elitist for the speed prediction of a container ship, Journal of Ocean Engineering,
multiobjective genetic algorithm: NSGA-II, IEEE Transactions on Evolu- 126, 152–162.
tionary Computation, 6, 2, 182–197. SPECIALIST COMMITTEE ON CFD 2014 In marine hydrodynamics—Final report
HILLEARY, R. R. 1966 The Tangent Search Method of Constrained Minimi- and recommendations to the 27th ITTC, Copenhagen, 2014.
zation, Monterey, CA: U.S. Naval Postgraduate School. SCHNEEKLUTH, H. AND BERTRAM, V. 1998 Ship Design for Efficiency and
HOLTROP, J. A 1984 Statistical reanalysis of resistance and propulsion data, Economy, 2nd ed, Butterworth-Heinemann. ISBN 0 7506 4133 9.
International Shipbuilding Progress, 31, 272–276. ULUNGU E. L, TEGHEM J., FORTEMPS P. H, AND TUYTTENDS D. 1999 MOSA
HOLTROP, J. A AND MENNEN G. G. J. 1982 An approximate power prediction method: a tool for solving multiobjective combinatorial optimization prob-
method, International Shipbuilding Progress, 29, 166–170. lems, Journal of Multi-Criteria Decision Analysis, 8, 4, 221–236.
HOLTROP-MENNEN FOUNDERS 2010 Reveal the secret of method’s long-lasting VETTOR, R. AND GUEDES SOARES, C. 2016 Development of a ship weather
success, MARIN Report, Vol 100, August 2010. routing system, Journal of Ocean Engineering, 123, 1–14.
View publication stats

jspd170034 rp1

Uploaded by

Copyright:

Available Formats

jspd170034 rp1

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

jspd170034 rp1

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

A Study on the Statistical Calibration of the Holtrop and Mennen Approximate

Article in Journal of Ship Production and Design · January 2018

Lampros Nikolopoulos Evangelos Boulougouris

SEE PROFILE SEE PROFILE

The user has requested enhancement of the downloaded file.

A Study on the Statistical Calibration of the Holtrop and Mennen

Keywords: Propulsion; model testing; powering; estimation; design (vessels); bulk

1. Introduction computational fluid dynamic solvers that solve Reynolds Averaged

MONTH 2018 2158-2866/18/nnnnnn-0001$00.00/0 JOURNAL OF SHIP PRODUCTION AND DESIGN 1

Principal particular VSL01 VSL02 VSL03 VSL04 VSL05 VSL06 VSL07

Vessel type KVLCC2 VLCC Newcastlemax Capesize Capesize Ultramax Ultramax

2 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 3

4 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 5

3.2. Ballast condition 3.3. Laden condition

6 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 7

8 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

4.1.1. Interpretation of the results. When looking at the results

4.1.2. Laden conditions. From Figs. 37–47, the distribution of F7

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 9

Calibration variables—laden (design and scantling) condition

L-16 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 0.1 0.50 0.04

L-17 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 4.0 1.20

10 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

Calibration variables—laden (design and scantling) condition

L-26 Rw ¼ c1 pc2 pc5 p∇pρpgpexpfm1 pFnd þ m2 pcosðλpFn2 Þg 2 0 1

Frictional resistance—form factor

L-27 c12 ¼ ðT =LÞ0:2228446 0.2228446 0.05 0.80

L-31 c13 ¼ 1 þ 0:003pC stern 0 3 3

Resistance due to bulbous bow

L-40 c6 ¼ 0:2pð1 0:2pFnT Þ 0.2 0.01 0.30

Model ship correlation

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 11

Calibration variables—laden (design and scantling) condition

12 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

Calibration variables—laden (design and scantling) condition

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 13

Calibration variables—ballast condition

B-16 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 0.1 0.8 0.01

B-17 m2 ¼ c15 pCp 2 pexpð0:1pFn2 Þ 2 3.5 1.5

B-18 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 8.07981 4.0 12.0

B-19 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 13.8673 25.0 10.0

B-20 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pCP 3 6.984388 3.0 11.0

B-21 c16 ¼ 8:07981pC P 13:8673pCP 2 þ 6:984388pCP 3 1 0.9 1.6

B-22 c16 ¼ 8:07981pCP 13:8673pC P 2 þ 6:984388pCP 3 2 1.8 2.9

B-23 c16 ¼ 8:07981pCP 13:8673pCP 2 þ 6:984388pC P 3 3 2.6 3.5

14 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

Calibration variables—ballast condition

Frictional resistance—form factor

B-35 c13 ¼ 1 þ 0:003pCstern 0.003 0.001 0.70

B-36 c13 ¼ 1 þ 0:003pC stern 0 0.10 2.0

Resistance due to bulbous bow

MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN 15

Calibration variables—ballast condition

16 MONTH 2018 JOURNAL OF SHIP PRODUCTION AND DESIGN

Calibration variables—ballast condition