As for drifting motion shown in Figs. 16 and 17, calculated Y' and N' show a good agreement with experimental value. Although measured value is slightly shifted when drift angle is zero, CFD calculation well estimates the inclination against drift angles.

　

　

 However, among comparisons in turning motion shown in Figs. 18 and 19, discrepancy can be found especially between the estimated and measured lateral force for pure turning motion ( β=0deg.) in Fig. 18.

　

　

 Hydrodynamic derivatives derived from computed and measured hydrodynamic force and moment are shown in Table 1 . MMG-type 3rd order expression model[6] is adopted here for the regression analysis of each derivatives that is determined for r' and β independently.

 As is the same conclusion by the comparison of hydrodynamic forces and moment, linear derivatives Y β, Nβ, Nr are estimated with high accuracy. However calculated Yr is about half of experimental value which means a lower estimation accuracy than other derivatives.

 Referring to 3^rd order nonlinear derivatives, calculated and measured Yβββ well agree to each other. On the other hand, obvious discrepancy can be found on other non-linear-derivatives. However their estimation error won't have a large influence on the estimation of maneuvering motion practically, since absolute values of non-linear hydrodynamic force and moment are small compared with the total force and moment.

　

　

 At initial design stage, evaluation of inherent directional stability based on a stability index is quite important. Accordingly, sway damping lever lv', yaw damping lever lr' and stability index C are calculated from linear derivatives and shown in Fig.20 together with the tank test data. CFD computation estimates sway and yaw damping lever with good accuracy, although slightly smaller than experimental value. Practically, it should be noted that the low estimation accuracy for Yr' does not have much effect on the whole because yaw damping lever shown in the following formula is greatly affected by non-dimensional mass coefficient: m' which is equal to 2^*Cb/(L/B) and much greater than Yr' especially for full ship.

　

　

　

 To establish the CFD code as a design tool, it is also important to confirm the estimation accuracy of stability index for ships with different fullness and aft frame shape.

　

　

 Fig. 21 shows a comparison between computed and measured stability index for several full ships.

 In this case, calculated stability index is determined by linear-derivatives which is derived from the calculated lateral force and yaw moment e within a range of β<10deg and r'<0.4.

 Although the result is scattering in some extent and CFD calculation can't estimate the stability index quantitatively in some case, present estimation code can grasp the tendency of inherent directional stability when ship's fullness and aft frame shape is changed. This result indicates that practical evaluation of directional stability at initial design stage is possible by use of CFD code.

 Although the CFD code can be used as a practical tool, their accuracy should be improved in the future to estimate all the derivatives as closely as possible.

 Hydrodynamic forces acting on hull during maneuvering motion are strongly influenced by viscous vortices generated on hull surface, i.e. vortex generated at bow bottom, the bilge vortex and resultant longitudinal vortex, the vortices generated at both end of the side flat and so on. These vortices flow into downstream after the separation. The definition of downstream direction is simple in drifting motion, however, possibly that the once separated flow returns to hull surface again in turning motion. To simulate such a flow field, adequate resolution of computation must be maintained up to far field. However, according to the restriction for computer resource, computing grid for CFD code is usually concentrated inside the viscous boundary layer, so that the grid resolution in far field becomes poor. Consequently the separated vortices disperse rapidly by numerical dissipation, which comes from the shortness of grid resolution. CFD methods acquired high estimation accuracy by focusing exact simulation of flow field in the vicinity of hull surface rather than vortices in far field. However, it is thought that, in pursuit of higher estimation accuracy, more attention should be paid to the precise simulation of vortices in far field.

　

3.3 Accuracy of estimation methods for sea-keeping performance

　

 To evaluate the sea margin, which consist of wind and wave effect, resistance increase in waves must be estimated accurately. Further, aft fullness is determined together with the restriction of fore fullness, correlation between fore fullness and resistance increase in waves should be carefully estimated at the initial design stage. In case of large full ship like VLCC, resistance increase is mainly consisted of the diffraction component around bow which is governed by bluntness factor(βf)

 As a practical method for the estimation of resistance increase in waves, Faltinsen's method[7] is normally applied. Fig.22 shows a comparison between estimation and tank test data for ships with various kinds of fore bluntness. Estimation and measurement shows a good agreement to each other, which imply this method can be applied as a design tool.