You are on page 1of 12

TOWARDS A CFD-BASED PREDICTION OF SHIP PERFORMANCE --PROGRESS IN PREDICTING FULL-SCALE RESISTANCE AND SCALE EFFECTS

H.C. Raven, A. van der Ploeg, A.R. Starke; Maritime Research Institute Netherlands (MARIN), Netherlands L. Ea, Instituto Superior Tcnico (IST), Portugal SUMMARY This paper discusses scale effects on the viscous and wave resistance of ships, as found from CFD computations; and compares these with the usual assumptions in the extrapolation of model test results to full scale. For some different ships computations of the viscous flow and wave pattern have been made for model and full scale, using the free-surface RANS code PARNASSOS. For these cases the viscous resistance appears to be approximately proportional to flat-plate frictional resistance (constant form factor 1+k), but relative to the ITTC 57 line 1+k increases clearly from model to ship. Computed scale effects on the wave pattern are reviewed. For a containership a 20% scale effect on wave resistance is found. A standard extrapolation method applied to the model-scale resistance here underestimates the full-scale resistance by 10%, but the empirical correlation allowance approximately corrects for that difference. This work shows how methods to predict fullscale resistance could be improved using results of todays CFD methods. NOMENCLATURE 1+k form factor Ca correlation allowance Cb block coefficient Cf frictional resistance coefficient flat plate friction coefficient Cf0 Ct total resistance coefficient Cv viscous resistance coefficient Cvp viscous pressure resistance coefficient Cw wave resistance coefficient Fn Froude number ks hull roughness parameter ship length between perpendiculars Lpp Rn Reynolds number S wetted surface V ship (or model) speed density of water 1. INTRODUCTION CFD tools are being used more and more extensively in practical ship hull form design today. Viscous-flow computations are used to provide detailed predictions of the flow around the hull, the wake field, occurrence of flow separation, for appendage alignment, etc. Based on the predictions and the insight obtained the hull form design may be modified to improve the flow quality, prior to any model testing. This CFD use already has enabled significant advances in design. However, when it comes to the final prediction of the power and RPM of the ship at full scale, in most cases a model of the final or near-final form is tested in a model basin. The performance predictions based on the model tests are still considered decisive. This is a somewhat paradoxical situation, because the prediction procedure based on model test results, the socalled extrapolation procedure, relies on simplifications that are not always adequate, and which are not needed for the CFD tools now available. However, in the long experience with these extrapolation methods empirical corrections have been established, based on correlations between predictions and trial measurements for a large sample of ships. These corrections provide an added accuracy which is not necessarily easy to beat using a CFD approach. We consider that a completely CFD-based prediction of the full-scale power is still not reliable enough in view of the large interests involved and the large accuracy required. On the other hand we believe that CFD has the potential to support and improve the extrapolation procedures in a near future; e.g. by giving insight in the validity of the scaling assumptions, providing corrections for certain ship types, or replacing parts of the extrapolation procedures by CFD computations. One of the goals of Workpackage 1 of the EC-sponsored project VIRTUE is to lay the basis for this. The first objective was to increase the accuracy of the resistance prediction. The next step was to make detailed computations of scale effects in resistance and wave patterns, and to compare these with the extrapolation methods. This step is the focus of the present paper. In the next sections, for reference we will first briefly describe the usual model-ship extrapolation method, and the RANS method used. Section 4 summarises recent research by MARIN and IST on the accuracy of the resistance prediction. Sections 5 and 6 discuss the scaling of the viscous and wave resistance, respectively, and compare these with the usual assumptions. Section 7 compares a direct prediction at full scale, with an extrapolation based on the computational model scale results, indicating some important differences. Conclusions are presented in Section 8.

2.

EXTRAPOLATION OF MODEL TEST RESULTS A resistance test for a ship model follows the procedure originally proposed by William Froude. The (hydrodynamic) resistance of a ship is considered to be the sum of two independent main contributions: the viscous (or frictional) resistance, and the wave (or residual) resistance. The former depends on the Reynolds number Rn, but is supposed to be unaffected by wave making; the latter depends on the Froude number Fn, but is supposed to be unaffected by viscosity. Therefore, the simplification is: Ct(Fn,Rn) = Cw(Fn) + Cv(Rn) (1) in which all resistance coefficients nondimensionalised with V2 S. have been

coefficient, possibly from the institutes own data, that includes the effect of hull roughness and other effects. Clearly, this extrapolation procedure is fairly crude. The assumed scale effect is determined by a single hull formdependent form factor and the plate friction line; the experimental determination of the form factor is often less accurate, its equality for model and ship is an approximation, and wave/viscous interaction effects are disregarded. While overall the procedure works satisfactorily, we believe it is of interest to check and perhaps improve its different components, hoping to increase the accuracy of the power prediction. Todays computational tools should enable this. 3. THE RANS SOLVER 3.1 GENERAL The code we use is PARNASSOS [1,2], a RANS solver developed and used by MARIN and IST, dedicated to the prediction of the steady turbulent flow around ship hulls. It solves the discretised Reynolds-averaged Navier-Stokes equations for steady incompressible flow. Various eddyviscosity turbulence models are available. No wall functions are used, not even for full-scale computations. Structured multiblock body-fitted grids are used, usually of H-O topology. A finite-difference discretisation is used, with second and third-order schemes for the various terms. The momentum and continuity equations are solved in their original, fully coupled form, without resorting to e.g. a pressure-correction or artificial-compressibility method. The whole method has been developed with numerical accuracy in mind. In practice, it is often perfectly possible to converge the solution to machine accuracy if that is needed; to work with very dense grids with large aspect ratio cells near the ship hull, etc. Full-scale computations can be done routinely. 3.2 FREE-SURFACE TREATMENT: THE STEADY ITERATIVE FORMULATION The method is of a free-surface fitting type: the upper boundary of the computational domain coincides with (an approximation of) the wave surface, which is repeatedly updated in the course of the solution. Free-surface boundary conditions are imposed here: a kinematic condition that the flow is along the wave surface; and a dynamic condition that the pressure is atmospheric and no shear stress is exerted on the wave surface. Almost all other methods for computing viscous freesurface flows follow a time-dependent procedure, starting from an initial condition and continuing until a steady result is obtained. Instead, we use the steady iterative approach, first proposed in [3] and derived in more detail in [4], which solves the steady free-surface problem

The model is tested at equal Froude number as the ship. In that case the wave pattern is supposed to be geometrically similar and the wave resistance coefficient Cw equal for model and ship. However, the viscous resistance coefficient Cv differs between model and ship, and is dealt with separately. Thus the full-scale resistance coefficient is: where the subscripts s and m indicate ship and model, respectively. The essence now is how to estimate the difference in viscous resistance coefficient between model and ship. In form-factor methods, which are mostly used today, the viscous resistance coefficient is supposed to be proportional to the frictional resistance coefficient of a flat plate at equal Reynolds number: (3) Cv(Rn) = (1+k) . Cf0(Rn) The form factor 1+k is a constant that depends on the hull form but is supposed to be independent of the Reynolds and Froude numbers. Therefore,

Cts = Cws + Cvs = Ctm Cvm + Cvs

(2)

Cvm Cvs = (1+k) . [Cf0(Rnm) Cf0(Rns)]


Empirical expressions are available for the plate friction coefficient Cf0 as a function of the Reynolds number; socalled plate friction lines, as shown in Fig.2. So if the form factor is known, the viscous resistance is known for both the model and the ship, and the ship resistance can be estimated from (2). One approach to determine the form factor is to measure the total resistance at a low speed, at which the wave resistance can be neglected or is so small that it can be corrected for. 1+k then follows directly from the viscous resistance and Cf0. To the total ship resistance coefficient so found, a correlation allowance Ca is added, an empirical

directly by iteration. The iterative solution is enabled by an alternative formulation of the free-surface boundary conditions, obtained by substituting the wave elevation from the dynamic condition into the kinematic condition. The resulting combined condition gives rise to a quickly converging iterative solution, as follows: the RANS equations are solved (iteratively) subject to the combined condition and the tangential dynamic conditions, imposed at the current wave surface; the wave surface is updated using the normal component of the dynamic condition. the grid is deformed so as to fit the new wave surface. The deformations are generated using a torsionalspring analogy method. Several applications [5,6,7] have shown that this procedure is very efficient, often requiring an order of magnitude less CPU time than other methods. A theoretical accuracy analysis [6] has indicated that we have a third-order numerical damping and dispersion, explaining the good wave pattern predictions obtained. 4. ACCURACY OF RESISTANCE PREDICTION Predicting the resistance might seem the easiest part of a ship viscous flow computation. Integrating the longitudinal component of the tangential and normal forces over the wetted surface yields the frictional and pressure resistance, respectively. However, the accuracy of these results requires more care than one would expect. The frictional resistance has usually rather little grid dependence and a numerical accuracy of about 1% is achievable, or less than that for some turbulence models. On the other hand, the frictional resistance depends significantly on the turbulence model used, and differences of about 5% can occur. The pressure resistance, however, is a larger problem. It consists of a viscous pressure resistance and a wave resistance. For slender ships it can usually be determined accurately; but not for full forms like tankers at low Fn. While for those the wave resistance can be very small, the viscous pressure resistance is of interest. The numerical prediction of the viscous pressure resistance is extremely sensitive to whatever disturbance, such as minimal pressure level changes in part of the flow domain due to incomplete convergence, the implementation and location of boundary conditions etc. As an example of the consequences, at the Tokyo workshop 2005 [7] many groups presented computations for the same KVLCC2 tanker test case; and while the flow fields were quite comparable, the standard deviation between all predictions for Cvp amounted to 23% (against 6.1% for Cf).

In VIRTUE, we have now done research [8,9,10] to improve the resistance estimates from PARNASSOS. Many possible influences have been checked, and progress was made on some: The force integration has been refined to deal more precisely with bow and stern contours that do not coincide with grid lines of the H-O type grids. The implementation of symmetry conditions at the still-water surface, in double-body flow computations, appeared to have an unexpected effect. A first-order discretisation at the symmetry plane results in gross errors in Cvp, a second-order scheme still shows a strong grid-dependence even on a grid of several millions of cells, but an improved formulation was found that removes most of the grid dependence [8]. Also, these different implementations of the symmetry conditions led to different predictions of the scale effect on the viscous pressure resistance. For some, the scale effect was 50-100% smaller on coarse grids than on fine grids [8]. The numerical prediction of scale effects thus requires small numerical uncertainties in the components of the viscous resistance, which is achieved by the present implementation in reasonably sized grids. The size of the computational domain also was found to influence the pressure resistance significantly. In particular if uniform-flow outer boundary conditions are used, large domains are required to remove the effect on the resistance [10]. These and other steps have resulted in more accurate and reliable resistance and scale effect predictions from our code; and thus have enabled the studies described below. 5. SCALING OF VISCOUS RESISTANCE In a towing-tank test, usually the form factor is determined from eq.(3) by measuring the total resistance at low speed, at which the wave resistance contribution vanishes or can be estimated. However, this can be rather inaccurate. Compared with this experimental procedure, determining a corresponding viscous resistance from RANS computations is more straightforward: the deformation of the water surface is simply not taken into account, and a double-body flow computation is done. If we make such computations for model and full scale we can determine the scale effect on the viscous resistance. We are going to compare that computed scale effect with the scaling assumptions in extrapolation methods. 5.1 HAMBURG TEST CASE Our first test case is the so-called Hamburg Test Case, a containership that was the subject of a workshop held recently in the VIRTUE project. Extensive measurements for this ship, at model and full scale, have been done by HSVA [11].

The model-scale grid had 44010052 nodes (2.3106 cells) in longitudinal, wall-normal and girthwise directions. For full scale, the same grid but with 140 cells in normal direction (3.2106 cells) and increased contraction rate was used to resolve the larger velocity gradients near the wall. The grid, of HO topology, was also contracted longitudinally towards bow and stern. All y+ values were below 1 and no wall functions were used. The turbulence model used was the recent k - kL model by Menter [12]. Computations were made for Rn = 1.177107 (model scale) and 1.2109 (full scale).

line. This is, however, not a real plate friction line; it was established in 1957 as an aid to give better extrapolations of ship resistance if no form factor method was used. It is given by:

Cf 0 =

0.075 (log Rn 2)2

Evaluating this for model and full-scale Reynolds number yields: model Cf0103= 2.917 ship Cf0103 = 1.497 ratio 0.513 From the computed viscous resistance for model and ship we then get the form factors: 1+k model= 1.138 1+k ship = 1.222 Therefore 1+k increases substantially from model to ship. An extrapolation using a fixed form factor would underestimate the ship viscous resistance by 7% !

0.004
ITTC57 Schoenherr Grigson Katsui et al.

0.003 CF
Figure 1 Computed wake fields for the Hamburg Test Case, model scale (left) and full scale (right). Fig.1 compares the model and full-scale wake fields, illustrating the usual features: a reduction of the width of the viscous wake at full scale, and a weakening of the hook shape caused by the longitudinal vortex. Rather than on the flow field, here we focus on the viscous resistance coefficients shown below: model double body ship double body Cf103 Cvp103 Cv103 2.794 0.524 3.318 1.546 0.282 1.828

0.002

0.001 6

8 log(Rn)

10

Figure 2 Flat-plate friction lines and the ITTC 57 modelship correlation line. However, the form factors evidently depend on the plate friction coefficients assumed, so let us see what other choices of the plate friction coefficients would give. Fig.2 from [13] shows the friction line proposed by Schoenherr and the more modern proposals by Grigson [14] and Katsui [15]. The difference in Rn-dependence is limited but sufficient to lead to significantly different performance predictions. Clearly the ITTC 57 line is highest at lower Rn and has a larger slope, while Grigsons line is highest at high Rn and has the lowest slope.

From model to ship Cv decreases by a factor of 0.55 in this case. In extrapolation methods the viscous resistance coefficient of a ship is supposed to be proportional to the friction coefficient of a flat plate at equal Rn, eq.(3). In the usual ITTC 78 method, a common choice for the plate friction line is the ITTC 57 model-ship correlation

The form factors for model and full scale based on these lines are shown below: friction line 1+k model 1+k ship ratio ITTC 57 line 1.138 1.222 1.074 Schoenherr line 1.161 1.221 1.052 Grigson line 1.154 1.157 1.003 Katsui line 1.179 1.202 1.020 Therefore, the computed scaling of the viscous resistance for this case matches best the Grigson line or the Katsui line if the form factor is supposed to be constant. We cannot conclude yet that some of these friction lines are more adequate than others. The Reynolds-number dependence of the predicted form factor depends on at least 3 aspects: the validity of the plate friction line, the proportionality of the viscous resistance of a ship to the equivalent plate friction; probably, the turbulence model used, and we will now consider the second and third aspect. Different turbulence models appear to give slightly different scale effects for ship viscous resistance coefficients. The same appears to be true for computed flat plate friction coefficients. Ea and Hoekstra [13] have done very careful plate friction computations, using again the PARNASSOS code, and show computed plate friction lines for a variety of turbulence models. As a matter of fact these appear to differ slightly from the lines of Fig. 2, dependent on the turbulence model. Therefore the third aspect, the turbulence model used, does play a role. The second aspect, the proportionality of the ship viscous resistance to flat plate friction, can now be assessed by determining the ratio of the computed viscous resistance for the ship and the computed plate friction, using the same turbulence model. So we use the numerical friction lines for the individual turbulence models, in this case the k-kl model[12], and get the results listed below: Cf0 model 103 = 2.736 Cf0 ship 103 = 1.490 1+k model = 1+k ship = ratio 1.213 1.227 1.012

5.2 KVLCC2 TANKER The second case we consider is the KVLCC2 tanker form [16], at Rn= 4.6106 (model) and 2.03109 (ship). A very careful study, using the same PARNASSOS code, has been done for this case by Ea et al [17]. Extensive grid dependence studies have been done for model and full scale, and the numerical uncertainty has been established. The finest grid had 4.16106 cells.

0.47 0.46 0.45 0.44 0.43 0.42 0.41

SST BSL TNT Mt KSKL ITTC Katsui Grigson

Therefore, for this Hamburg Test Case, we conclude: that the proportionality of the ship viscous resistance and flat plate resistance is well reproduced, if both are computed using the same turbulence model; that the same is true if one of the modern friction lines is used, but not the ITTC 57 line; that the computed resistances indicate that a prediction of full-scale viscous resistance using a fixed form factor together with the ITTC 57 line (i.e. the usual approach), would be 7% too low.

0.4 Figure 3 Full scale / model scale viscous resistance ratio for KVLCC2 and equivalent flat plate. Scale effects on the viscous resistance were then derived for 5 different turbulence models. Fig.3 displays the results in the form of a bar chart for the ratio of ship and model viscous resistance, compared with the various plate friction lines, both experimental and computed. The first 5 bars represent the computed ratios for the KVLCC2, using 5 different turbulence models. The right 5 bars show the same but now for a flat plate. The middle 3 columns show the scaling according to the ITTC 57 line and the Katsui and Grigson flat plate friction lines, respectively. We observe: that for the ship there is some variation in resistance ratio (so in the scaling of the ship viscous resistance) between turbulence models; that the same goes for the flat plate computations, and that the differences between turbulence models have partly the same pattern as for the ship but are smaller. Therefore the use of a numerical plate friction line for the same turbulence model gives somewhat more consistent changes in the form factor from model to full scale. that the computed resistance ratios for the ship are all smaller than their counterparts for the flat plate, indicating that the form factor 1+k (if calculated from the numerical friction lines) decreases from model to full scale by 1-4%.

that the Katsui line gives a ratio closely corresponding with both the numerical plate friction lines and the ship results, and would give a fairly constant form factor (-3 to +1%); while the use of the Grigson line would yield a significant decrease (37%), the ITTC 57 line a significant increase (4-8%) of the form factor 1+k from model to full scale.

Therefore, this second case suggests that the assumption of a constant form factor is still quite reasonable but not as precise as for the more slender previous case. Again the ITTC 57 line deviates significantly from the computed scaling of both the plate friction and the ship viscous resistance. The slightly different scalings obtained for a containership and a tanker may be related to the occurrence of flow separation at the stern for the tanker, which is present at model scale but not at full scale. In any case, there is little fundamental support to the form factor concept and we must be prepared for its validity to depend on the case. Several more ships need to be done to establish the trends. It is of interest to go back to the results submitted to the Gothenburg 2000 workshop, for the same KVLCC2 case. Eight different sets of model and full-scale double-body flow results were submitted by different groups. Form factors 1+k based on the ITTC 57 line showed a significant increase from model to full scale, varying between 5 and 17%. Also the full-scale form factors were much higher than those found now. But we believe that in the 2000 results, those indications were significantly affected by numerical and modeling inaccuracies. On the other hand, in the much more accurate computations of today we begin to discern the deficiencies of extrapolation methods. 6. SCALE EFFECTS ON WAVE MAKING

hand, along the afterbody the boundary layer thickens quickly due to the decreasing girth length and the increasing pressure towards the stern. The displacement thickness of the boundary layer and wake reduce that pressure increase, and more so at model scale than at full scale. The reduced pressure increase in most cases leads to a reduced stern wave generation, again more pronounced at model than at full scale; but this depends on the stern shape. It is to be noted that, while viscous effects thus affect the generation of the waves, they are much less pronounced on the wave propagation. Viscous attenuation of ship waves is negligible, only there could be an indirect refraction effect caused by propagation of the waves over the viscous flow field.
0.03

Series 60
inviscid flow `h / L;Y/L;V3 viscous (model) `h / L;Y/L;V3 viscous (ship) `h / L;Y/L;V3 experiments (model)

0.02

y / Lpp = 0.0755

`h / L;Y/L;V3 0.01
pp

6.1 THE WAVE PATTERN The other main assumption to be scrutinised is that of the equality of the wave resistance coefficients for model and ship, which is based on the assumed similarity of the model and ship wave patterns. This is a quite reasonable approximation. In general, the boundary layer around the hull is thin over the forward part of the hull, and in that region the pressure field is hardly affected by viscous effects. Correspondingly, we expect hardly any viscous effect on the wave making of the forebody. On the other

0.01

Figure 4. Longitudinal wave cut for Series 60 Cb = 0.60. Fig. 4 illustrates an early computation of a wave pattern scale effect from [6]. The case considered is the Series 60 parent form, Cb=0.60, at Fn=0.316, Rn =3.4 106 (model) and 8.4 108 (ship). Grids of 1.8 and 2.3 M cells were used for model and full scale, respectively. The figure shows a wave cut at 0.0755 Lpp off the centreline. The domain used at that time was rather short and just shows little of the stern wave system. However, the graph exactly agrees with our expectations: the inviscid flow computation using the free-surface panel code RAPID [18], the model and full scale wave cuts from the RANS code, and the experimental cut from [19] all agree closely for most of the length; but aft of the stern the inviscidflow computation overestimates the stern wave amplitude, the model-scale RANS/FS computation is close to the model experiment, and the full-scale computation is in between.

inviscid flow z;Y/L;V3 viscous (model) z;Y/L;V3 viscous (ship) z;Y/L;V3 experiments (model) z;Y/L;V3

y / Lpp = 0.0900

5
Figure 5. Dyne tanker at Fn = 0.165. Computed longitudinal wave cut from PARNASSOS at model scale and full scale, from inviscid-flow code (RAPID), and from model experiment. computations have been done for free trim and sinkage, at Another case from [6], the Dyne tanker, has a very full Fn=0.238, Rn=1.177 107 and 1.2109. Consistently good agreement was found with the experimental wave hull form with Cb=0.87 and a cylindrical bow shape. The cuts; more on the validations for this case will be shown computations were made for Fn=0.165, Rn=8.5106 and 1.83109. Fig.5 shows that a drastic reduction of the stern in [21]. Fig.8 compares the wave patterns from the waves occurs compared with the inviscid computations; inviscid-flow code and the viscous-flow results. There is both for the full scale and, somewhat stronger, for model close agreement except for the stern wave system, of scale. The slight scale effect on the amplitude of which the amplitude is reduced by viscous effects. Fig. 9 transverse waves along the hull, from the fore shoulder aft, illustrates the model and full scale wave patterns and is unexpected, but may be connected with the large confirms that the scale effects are local and rather limited. pressure gradients at the fore shoulder. There is good agreement with the experimental data from [20]. For the KVLCC2 tanker (Cb=0.81), we have also performed RANS/FS computations. However, for the low Froude number considered, Fn=0.147, the waves are quite short and the grid used (1.8 M cells) may still have been somewhat coarse. Fig.6 compares the full scale and model-scale wave patterns, Fig.7 compares wave cuts for both with those from the inviscid-flow prediction. There is again a drastic reduction of the stern wave system compared with inviscid flow, while the model / full scale difference is relatively minor but the full-scale stern wave amplitude is somewhat larger. Slight differences in the transverse wave system along the hull are also observed, which suggest that again there is some viscous effect on these waves as well. The final, and most complete, set of results we consider is again for the Hamburg Test Case, which is much more slender (Cb=0.645) than the two preceding ones. The

Figure 6. Computed wave pattern of KVLCC2 tanker at Fn=0.147; full scale (left) and model scale (right)

Figure 9. Stern view of computed wave patterns of Hamburg Test Case at Fn=0.238, for full scale (left) and model scale(right). Wave heights multiplied by 5.

Figure 7. Longitudinal wave cuts for KVLCC2, at y=0.125L (top) and 0.20 L (bottom); from inviscid, full scale and model scale computations.

Figure 10. Hamburg Test Case, Fn=0.238. Longitudinal wave cuts at y/L=0.184 , full scale, model scale and inviscid. Fig. 10 compares wave cuts at y/L = 0.184, from the inviscid flow computation and the model and full scale RANS/FS computations. In this more precise comparison there appears to be no viscous effect on the bow wave system, and therefore no scale effect. The graph suggests some viscous effects on the short diverging waves at x/Lpp=0.1, but we believe these are due to the numerics of the panel code rather than physics. The stern wave system displays a strong reduction by viscous effects and a significant scale effect. Fig. 11 shows a wave cut at the centreline aft of the transom, illustrating the large difference with the inviscid-flow result in this area. Figure 8. Wave pattern of Hamburg Test Case, Fn=0.238; from potential-flow (top) and RANS computation for full scale (bottom). From the examples given we note that a reduction of the stern wave system by viscous effects, sometimes also

with a change of character (e.g. less transverse waves), is rather common; and this reduction is largest for model scale. For the two tanker hull forms the viscous reduction of the stern wave amplitude was comparable for model and ship and very substantial; and there seem to be also some viscous changes of the waves along the hull. In [22], viscous and scale effects of a different nature are considered for dry and wetted transom sterns.

single coarse grid computation (grid spacings doubled in longitudinal and girthwise directions) yielded a model scale resistance 1.4% higher than on the fine grid. In view of uncertainty results for other cases we consider the accuracy of the resistance comparisons sufficient for the statements made, which are based on differences between computations with exactly similar settings and grids. Table 1 shows that the wave resistance coefficient is 20% larger at full scale than at model scale. This increase, which is contrary to the common assumption in Froudes hypothesis, seems consistent with the increase of the stern wave system. In the potential-flow result, the stern waves are still much higher, and the wave resistance coefficient computed is still 47% higher, so this fits well with the present results. Table 1: Computed resistance components for Hamburg Test Case. Cf103 model free 2.852 surface model 2.794 double-body wave resistance model Cp103 0.839 0.524 Ct103 3.691 3.318 0.373 0.689 0.282 2.276 1.828 0.448 0.66

Figure 11. Hamburg Test Case at Fn = 0.238. Wave cut at centreline aft of the transom. Full scale, model scale and inviscid. 6.2 SCALE EFFECTS ON WAVE RESISTANCE As viscous and scale effects on the stern wave making are present, the same must go for the wave resistance: a lower stern wave system gives less radiated wave energy and therefore a lower wave resistance. Below we consider the limited data we have on this so far. Here we define wave resistance as the difference between the total resistance with wave making and without wave making (i.e. in double-body flow). The case with wave making has to be computed with free trim and sinkage. The wave resistance mainly comes from a change of the pressure resistance, but also contains changes of the viscous resistance caused by the wavemaking; e.g. due to the modified flow direction and pressure field along the hull. For the Hamburg Test Case, Table 1 shows the computed resistance values. For model scale, the total resistance appeared to be 10% lower than the experimental value of Ct=4.110-3. About half of this difference is due to the choice of the turbulence model; an additional calculation using the SST model gave Ct=3.82810-3. While no complete grid dependence studies have been done here, a

ship free 1.588 surface ship double 1.546 body wave resistance ship wave resistance inviscid

It is to be noted that this scale effect on the wave resistance is a rather small quantity, and it can only be determined from most accurate computations. As a matter of fact, the study for the KVLCC2 hull at Fn=0.147, shown in Fig. 6, illustrated this. The estimated wave resistance at full scale amounted to 12% of the total resistance, and again was higher than the model scale wave resistance and intermediate between the modelscale and potential-flow values. However, a minor change in the discretisation of the free-surface boundary conditions caused a change of the computed wave resistance comparable with that scale effect. To get a quantitative estimate of the scale effect for that case, denser grids will be needed. More of these studies have to be done before we can establish the trends with some certainty.

7.

A COMPARISON WITH MODEL-SHIP EXTRAPOLATION For the Hamburg Test Case, in the two previous sections we have observed some substantial differences in the resistance scale effects compared with what is assumed in extrapolation methods: while the form factor approach in itself is applicable, the ITTC 57 line is too steep and causes an underestimation of the viscous resistance at full scale. the wave resistance coefficient for full scale is some 20% larger than that for model scale in this case. To illustrate what this means, we compare with the result of the standard extrapolation method, applied to our model scale computations. From the double-body flow computation at model scale we have found before a form factor of 1+k = 1.138 relative to the ITTC 57 line. For the ship the ITTC 57 line yields Cf0=1.49710-3 (Section 5.1). Using the same form factor as at model scale we get Cvs = 1.70210-3. As shown in Table 1, the wave resistance coefficient for model scale is Cwm = 0.37310-3. The same value is assumed for the ship at full scale, Cws = Cwm. From eq. (2) the extrapolated total resistance coefficient for the ship thus becomes Cts=2.07510-3. The table below compares this with the direct computation. Table 2 Comparison of extrapolated and directly computed full-scale resistance components for Hamburg Test Case. extrapolated computed difference Cv103 1.702 1.828 7.4% Cw103 0.373 0.448 20.1% Ct103 2.075 2.276 9.7%

added. This is an empirical addition, derived from analysis of the correlation between extrapolated power predictions and trial measurements. Consistent with the use of the ITTC 78 extrapolation method used as a reference here is the use of the Bowden-Davison formula [23]:

Ca = 0.105(ks / L pp ) 3 0.64 103


which has been derived from thrust measurements during trials for single-screw ships, and should be recognized to be a correlation allowance including effects of roughness rather than a mere roughness allowance [24]. Assuming the standard value of ks=0.00015 m yields a correlation allowance for the Hamburg Test Case, including the roughness effect, of 0.4010-3. This brings the result of the extrapolation based on the model-scale resistance to: 2.47510-3. On the other hand, Townsin et al [25] have proposed a correlation for just the effects of hull roughness,
1 1 Cf = 0.044 ( AHR / Lpp ) 3 10 Rn 3 + 0.125 103

where AHR is a roughness measure that can here be considered equal to ks. Again taking a standard value of 0.00015 m then yields a roughness allowance of 0.14710-3. In other words, the difference between both allowance estimates, 0.2510-3, could be considered to represent effects not related with roughness; such as deviations in the extrapolation procedure. Adding the roughness allowance Cf to our directly computed full-scale resistance (which disregarded roughness effects so far) we obtain a total resistance coefficient of 2.42310-3. This is in close agreement with the total extrapolated resistance plus correlation allowance. This is not to say that both ways of getting a full-scale resistance are in agreement in general, but it explains how there may be significant deviations in the scaling assumptions in extrapolation methods, which still lead to a generally accurate final prediction. Even so we believe that, to increase the reliability of model-ship extrapolation, it is desirable to decompose the correlation allowance, which now amounts to 20% of the extrapolated resistance, in contributions of known physical effects and a remainder collecting any unknown effects. E.g. if we would have a more precise knowledge of the validity of the form factor approach for different classes of ships, and of wave resistance scale effects depending on fullness and stern type, corrections could be made in the extrapolation. A correlation allowance then to be added would hopefully have a smaller magnitude and smaller variability, and lead to a full-scale power estimate of greater reliability. Todays CFD tools can either provide information on this, or can perhaps replace parts

extrapolated+correlation allowance: computed + roughness allowance: difference

2.475 2.423 -2.1%

It appears that the two deviations from the assumed scale effects add up to an almost 10% difference in the resistance of the ship at full scale! 6.1% of this is due to the scaling of the viscous resistance, 3.6% due to the scale effect on wave resistance. For the KVLCC2 case, although with limited numerical certainty we got a very similar difference. However, to the extrapolated resistance usually a correlation allowance or roughness allowance Ca is

of the extrapolation procedures by computations. Evidently, there is much more work to do before we are there. One of the next steps in our research will be a study on the prediction of roughness effects in RANS computations, hoping to again obtain a more accurate estimate of one of the contributions to the correlation allowance. 8. CONCLUSIONS The present paper has addressed scale effects in ship resistance components, as found from RANS computations; and has compared these with the assumptions underlying the extrapolation methods used to predict ship resistance from model tests. The scaling of the viscous resistance has been determined from double-body flow computations. For a containership it was in agreement with a form factor method, i.e. the viscous resistance at model and full scale was nearly proportional to the frictional resistance of a flat plate; provided that the latter is found from either the Grigson or Katsui plate friction line, or is computed using the same turbulence model. If instead the ITTC 57 line is used, the form factor at full scale is significantly higher than at model scale. For a tanker, some decrease of the form factor from model to ship was found if a proper plate friction line is used, but again an increase if the ITTC 57 line is used. Recent results for viscous and scale effects on ship wave patterns have been surveyed. In all cases the stern wave system appears to be reduced by viscous effects, and more so at model scale than at full scale. Correspondingly, for the same containership the wave resistance coefficient was found to be 20% larger at full scale than at model scale. For the containership, the computed scaling of viscous and wave resistance means that the directly computed resistance for the full-scale ship is 10% higher than the resistance found from the model-scale results using the extrapolation method with ITTC 57 line. This difference is compensated by the correlation allowance, which not only corrects for roughness effects but also for deviations in the resistance scaling. A better understanding of the scaling, as obtained from CFD studies, could reduce the required correlation allowance and its variability, and therefore increase the reliability of the prediction. Many more cases need to be studied before definitive conclusions can be drawn. However, the present study demonstrates how the advanced modelling used in CFD today can contribute to increasing the reliability of performance predictions based on model tests.

9. ACKNOWLEDGEMENT The research reported in this paper formed part of MARINs work in WP1 of the VIRTUE project, an Integrated Project in the 6th Framework Programme Sustainable development, global change and ecosystems under grant 516201 from the European Commission. This support is gratefully acknowledged. 10. REFERENCES 1. HOEKSTRA, M., "Numerical simulation of ship stern flows with a space-marching Navier Stokes method", Thesis, Technical University of Delft, October 1999. 2. VAN DER PLOEG, A., EA, L. and HOEKSTRA, M., "Combining accuracy and efficiency with robustness in ship stern flow calculation", 23rd Symp. Naval Hydrodynamics, Val de Rueil, France, Sept. 2000. 3. RAVEN, H.C. and VAN BRUMMELEN, E.H., "A new approach to computing steady free-surface viscous flow problems", 1st MARNET-CFD workshop, Barcelona, Spain, 1999. 4. VAN BRUMMELEN, E.H., RAVEN, H.C. and KOREN, B., "Efficient numerical solution of steady free-surface Navier-Stokes flow", Jnl. Computational Physics, Vol. 174, 2001, pp. 120-137. 5. RAVEN, H.C. and STARKE, A.R., "Efficient methods to compute ship viscous flow with free surface", 24th Symp. Naval Hydrodynamics, Fukuoka, Japan, 2002. 6. RAVEN, H.C., VAN DER PLOEG, A., and STARKE, A.R., "Computation of free-surface viscous flows at model and full scale by a steady iterative approach", 25th Symp. Naval Hydrodynamics, St.Johns, Canada, 2004 7. HINO, T. (editor), "CFD Workshop Tokyo", National Maritime Research Institute, Tokyo, Japan, 2005. 8. RAVEN, H.C., VAN DER PLOEG, A., EA., L. `Extending the benefit of CFD tools in ship design and performance prediction, 7th International Conference On Hydrodynamics, Ischia, Italy, October 2006 9. VAN DER PLOEG, A., RAVEN, H.C., `Accuracy of viscous resistance computations for ships including free-surface effects, MARINE 2007 symposium, Barcelona, 2007. 10. EA, L. AND HOEKSTRA, M., On the Numerical Accuracy of the Prediction of Resistance Coefficients in Ship Stern Flow Calculations, to be published in Jnl Marine Science Techn. 11. LAUDAN, J., "Korrelation von Nachstromaufmessungen an Modell und Grossausfuerung" (in German), (BMFT Foerderungskenzeichen MTK 0185/2), HSVA Bericht Nr. 1529, April 1983.

12. MENTER F.R., EGOROV Y., RUSCH D. - `Steady and Unsteady Flow Modelling Using the k-kL Model, Turbulence Heat and Mass Transfer 5, 2006. 13. EA L., HOEKSTRA M., `The Numerical Friction Line, to be published in Journal of Marine Science and Technology, Springer, 2007. 14. GRIGSON, C. W. B., `A planar friction algorithm and its use in analysing hull resistance, RINA 1999. 15. KATSUI, T., ASAI, H., HIMENO, Y., TAHARA, Y., `The Proposal of a New Friction Line, Fifth Osaka Colloquium on Advanced CFD Applications to Ship Flow and Hull Form Design, Osaka, Japan, 2005. 16. LARSSON, L, STERN, F., BERTRAM, V., Gothenburg 2000: a workshop on numerical ship hydrodynamics, Chalmers Univ. Techn., Gothenburg, Sweden, 2000. 17. EA, L., HOEKSTRA, M., RAVEN, H.C., Scaling of viscous resistance coefficients based on the numerical friction line, IST/MARIN cooperative research report, January 2008. 18. RAVEN, H.C., "A solution method for the nonlinear ship wave resistance problem", Thesis, Technical University of Delft, 1996. 19. TODA, Y., STERN, F., LONGO, J., Mean-flow measurements in the boundary layer and wake of a Series 60 Cb=.6 ship model for Froude numbers .16 and .316, IIHR report 352, Iowa, 1991. 20. LUNDGREN, H. and HMAN, M., "Experimentell och numerisk bestmning av vgmotstand fr ett tankfartyg (Dynetankern)", Report No. X-94/58, Dept. Naval Architecture and Ocean Engineering, Chalmers Univ. Techn., Gothenburg, Sweden, 1994. (In swedish). 21. VAN DER PLOEG, A., RAVEN, H.C., WINDT, J., LEROYER, A., QUEUTEY, P., DENG, G. and VISONNEAU, M., `Computation of free-surface viscous flows at model and full scale --- a comparison of two different approaches, paper accepted for 27th Symp. Naval Hydrodynamics, Seoul, Korea, Oct. 2008. 22. STARKE, A.R., RAVEN, H.C., VAN DER PLOEG, A., `Computation of transom-stern flows using a steady free-surface fitting RANS method, 9th International Conference on Numerical Ship Hydrodynamics, Ann Arbor, Michigan, August 5-8, 2007 23. BOWDEN, B.S. DAVISON, N.J., Resistance Increments Due to Hull Roughness Associated with Form Factor Extrapolation Methods, NPL Ship Report TM 3800, 1974 24. 24th ITTC, Final report and recommendations of the Specialist Committee on Powering Performance Prediction, 2005. 25. TOWNSIN, R.L., MEDHURST, J.S., HAMLIN, N.A., SEDAT, B.S., Progress in Calculating the Resistance of Ships with Homogeneous or

Distributed Roughness, NECIES Centenary Conference on Marine Propulsion, 1984.

11.

AUTHORS BIOGRAPHIES

Hoyte Raven is Principal Researcher CFD at MARIN, responsible for the coordination of R&D in this field, and mainly involved in research on computational prediction of ship wave making. Auke van der Ploeg is CFD Researcher at MARIN, a specialist in numerical methods, and a main developer of the free-surface approach in PARNASSOS. Bram Starke is CFD Researcher at MARIN, and is mainly involved in the development and utilisation of PARNASSOS, in particular the free-surface approach and propeller-hull interaction. Lus Ea is Assistant Professor at the Mechanical Engineering Dept. of Instituto Superior Tecnico from the Technical University of Lisbon. He is a main developer of the PARNASSOS code, and a specialist in CFD Uncertainty Analysis.

You might also like