Sunteți pe pagina 1din 18

Ocean Engineering 28 (2001) 585602

Estimating the hydrodynamic forces on a mini TLP with computational uid dynamics and design-code techniques
B.A. Younis
a

a,*

, P. Teigen b, V.P. Przulj

Department of Civil Engineering, City University, London EC1V 0HB, UK b STATOIL, Postuttak, N-7005 Trondheim, Norway Received 10 October 1999; accepted 12 January 2000

Abstract This paper reports on the prediction of the hydrodynamic forces on a full-scale mini Tension Leg Platform (TLP) of the type typically deployed for deep-sea operation. Two alternative prediction techniques were used: Computational Fluid Dynamics (CFD), which is based on the solution of the fundamental equations that govern turbulent uid ow; and engineering calculations based on force coefcients derived from a design code that is in routine use in the Offshore Industry. The results from these two techniques were compared with each other and with experimental data obtained from wind-tunnel and towing-tank tests on a 170 scale model. It was found that the two techniques, while yielding very similar predictions for the front TLP members, give substantially different predictions for the aft members a result that is consistent with the presence of signicant interference effects that are captured only by the CFD. The design code yielded the highest value for the global drag coefcient, followed very closely by the towing-tank result. The wind-tunnel tests produced the lowest value for this parameter. The CFD predictions, which were the rst to be obtained in this study, fall in the mid-range of the experimental values. These and other results are discussed in the context of the use of CFD in practical design applications. 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Hydrodynamic loads; Fluidstructure interactions; TLP; CFD

* Corresponding author. Fax: +44-020-74778570. E-mail address: b.a.younis@city.ac.uk (B.A. Younis).


0029-8018/01/$ - see front matter 2000 Elsevier Science Ltd. All rights reserved. PII: S 0 0 2 9 - 8 0 1 8 ( 0 0 ) 0 0 0 1 9 - 6

586

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

1. Introduction There is, at present, strong interest within the Offshore Industry in deep water explorations and development. Fields in the range 8002000 m water depth are being considered in such different areas as the Norwegian Sea (Vring Platau), the Gulf of Mexico, Greenland, Brazil and West Africa. To meet the new challenges, a variety of strategies and concepts for oil production have been developed which, although differing markedly in many aspects, share the common feature of being reliant on oater technology. A principal design consideration here is the maximum offset which is produced by the combination of the mean offset and the estimated extreme of the corresponding dynamic motion response. In areas with harsh environment, the total offset will usually be dominated by wind- and wave-induced responses, whereas in areas with a more benign climate, the currents tend to take precedence and may be responsible for more than 60% of the total offset. While rst- and, to a certain degree, second-order wave diffraction effects are generally well understood and can be calculated fairly accurately, the prediction of the current loads is still subject to uncertainty leading, inevitably, to conservatism in the overall design. Thus, for example, a single, global force on the TLP may be obtained from measuring this quantity on a small-scale model and then extrapolating the result to the much higher Reynolds numbers encountered in practice. This is known to involve scaling effects that are not always easy to quantify. More usually, estimates of the global force on a TLP are obtained by using engineering design codes to obtain values for the drag coefcients for the individual TLP members. These then allow for the determination of the drag forces on each member and, by summation, on the whole TLP. Cumulative experiences with using this approach suggest that the forces obtained in this way are somewhat too large, leading to over-conservatism in the nal design. This, in part, is due to the substantial interference effects that arise whenever a TLP member lies in the wake of another member located upstream of it. Corrections for the effects of interference do exist but these have to be applied separately and are, at any rate, an additional source of uncertainty in the design. An alternative to this semi-empirical approach is provided by Computational Fluid Dynamics (CFD) a technique based on the numerical solution of the fundamental equations that govern the uid motion. These equations are exact but their solution, especially for turbulent-ow conditions, involves a number of assumptions and approximations that can profoundly inuence the quality of the predictions. Amongst the principal sources of uncertainty in the CFD predictions is the choice of turbulence model used to represent the effects of turbulence in the time-averaged mean-ow equations. A number of different models are in use (e.g. Speziale, 1991), and these differ widely in their complexity as well as in their ability to account for the effects of the turbulence generated by interaction with the TLP. Thus, simple algebraic models that have the advantages of being computationally robust and economical fail badly in the massively-separated ows that occur around the TLP. Conversely, second-order closures do better in such complex ows but are generally less robust and far more demanding of computational resources. Another source of uncertainty in all CFD studies is the presence of numerical errors that arise from the transformation of the differential governing equations

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

587

into algebraic approximations. These errors can be quite substantial, especially in the three-dimensional complex ows around the TLP. There is then a clear need for a systematic study of these various sources of uncertainty as a prerequisite to the acceptance and use of CFD as a tool for routine analysis. In this paper, we put on record our experiences in using CFD to predict the forces on a full-scale mini TLP and compare the results with values obtained from an engineering design code. Comparisons are also made with experimental data obtained from wind-tunnel and towing-tank tests on a small-scale model of the same TLP.

2. Mathematical model 2.1. Governing equations The numerical simulation of the ow eld around the TLP is based on the solution of the differential equations that govern the conservation of uid mass and momentum in three directions. For the steady-state ow of a uniform property uid, these equations may be written as: Continuity: Ui 0 xi Momentum: Uj (1)

Ui Ui 1 p n uiuj xj xj xj rxi

(2)

In the above, Ui is the mean-velocity vector, p is the static pressure and r and n are, respectively, the uids density and its kinematic viscosity. Conventional Cartesian tensor notation is used wherein repeated indices imply summation. The solution of the steady-state form of the governing equations precludes the capture by the predictions of mean-ow unsteadiness due, for example, to the shedding of vortices from the various components of the TLP. This, we recognize, is an important limitation in the scope of this work and is one that we aim to address in a later study. Eq. (2) is recognizably the Reynolds-averaged form of the NavierStokes equations wherein the instantaneous values of velocity and pressure are replaced by their long-time averaged counterparts. This is done by formal averaging of the original equations over a time interval that is substantially longer than the time scale of the turbulence uctuations. This is the usual approach that is adopted in the practical simulation of complex ows at high Reynolds numbers. The alternative, which is to solve for the instantaneous variables themselves, would require computational resources that are far in excess of any that are available for the task. A consequence of the averaging process is the appearance of the Reynolds-stress correlations (uiuj ).

588

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

These are not known a priori and need to be determined in order to close Eq. (2). Determination of these correlations is done here by adopting Boussinesqs linear stressstrain relationship: uiuj nt

Ui Uj 2 dij k xj xi 3

(3)

where nt is the eddy viscosity and k is the turbulence kinetic energy. The eddy viscosity is determined in this work by the use of a two-equation model of turbulence wherein this quantity is obtained from: k2 ntCm (4)

where is the rate of turbulence dissipation by viscous action. k and are obtained from the solution of their own differential transport equations. Those are given by: Uj Uj k nt k Pk xj xj skxj nt 2 C1 PkC2 R xj xj sxj k k Ui xj

(5) (6)

where Pk is the rate of production of k: Pkuiuj (7)

The term R in Eq. (6) is absent from the standard k model (Launder and Spalding, 1974) but appears in the RNG variant of this model (Renormalization Group, Yakhot et al., 1992) where it is dened as: R Cmh3(1h/ho)2 1+bh3 k (8)

In the above, h is the ratio of the time scale of the turbulent motions to that of the mean ow and is determined from k h S S (9) 1 Ui Uj + 2 xj xi

(10)

The values assigned to the various model coefcients are given in Table 1. These values are generally accepted as being standard; they were originally arrived at

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

589

Table 1 k model coefcients Model Standard RNG Cm 0.09 0.0845 sk 1.0 0.72 s 1.3 0.72 C 1 1.45 1.42 C2 1.90 1.68 b ho

0.012

4.38

from a combination of theory, matching to experimental data for simple, homogeneous ows and from numerical optimization against data from benchmark twodimensional shear layers. Assessment of the sensitivity of the computed solutions to this choice of model coefcients was not attempted in this work on the basis that a modied set of values that produces improved results for this case is not guaranteed to produce improved predictions for different geometries and ow parameters. 2.2. Solution methodology and boundary conditions Eqs. (1)(10) were solved by using the computer code CITY3D (Younis and Przulj, 1996). This is a nite-volume method for the solution of the generalized ow equations in arbitrary domains. The method is based on a term-by-term integration of the governing equations over micro-control volumes sub-dividing the solution domain. The diffusive uxes were approximated using the second-order accurate central-differencing scheme. The convective uxes were evaluated using a variety of differencing schemes of varying orders of accuracy. Specically, we used the UPWIND scheme which is only rst-order accurate and is typically employed in commercial CFD software; the MINMOD scheme which is second-order accurate; and the SMART scheme which is third-order accurate. We report below on the sensitivity of the computations to the choice of differencing scheme. The CITY3D solution methodology is iterative and is based on the SIMPLE algorithm (Semi-Implicit Method for Pressure Linked Equations) of Patankar and Spalding (1972). This algorithm ensures that the calculated ow eld, once converged, simultaneously satises both the continuity and momentum equations. The set of linear algebraic equations that result from integration of the equations was solved using an efcient conjugategradient technique. The boundary conditions employed for the TLP simulations were as follows (refer to Fig. 1). The inlet to the computational domain was placed at a distance of 14D upstream of the TLP (D being the column diameter). The velocity of the approach ow was taken to be uniform in the vertical plane (i.e. U(y,z)=Uo) and the relative turbulence intensity level (u/Uo) was set at the low value of 0.05. The ratio of eddy to molecular viscosity was set equal to 100 and used, via Eq. (4), to determine the dissipation rate value at inlet. The ow at the exit to the computational domain (located at a distance 32D from the TLP) was assumed to have become fully reestablished in the streamwise direction and thus the gradients of all dependent variables in that direction were set equal to zero. The free surface boundary was treated here as a plane of symmetry and hence the gradients of all dependent variables across

590

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

Fig. 1.

TLP geometry and coordinates.

it were set equal to zero, except for the velocity component normal to it which was itself set equal to zero. Finally, the TLP walls were assumed to be smooth, and the ow in their immediate vicinity was assumed to follow the universal logarithmic distribution: U 1 utn ln E ut n

(11)

where ut is the friction velocity and n is the normal distance from the wall. The constants (the von Karman constant) and E were assigned their usual values of 0.41 and 9.0, respectively.

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

591

Table 2 Dimensions of TLP Parameter Column height, H Column diameter, D Column separation distance Pontoon height, B Draught Dimension (m) 22.25 8.75 28.50 6.25 28.50

3. Results and discussion Fig. 1 shows the TLP geometry under consideration and denes the coordinate system used. The TLP dimensions used in the present simulations are listed in Table 2. CITY3D is interfaced to the PATRAN and FEMSYS suites of pre- and postprocessing software packages and the latter was used to generate computational meshes that conform to the boundaries of the TLP. Three such meshes were generated, consisting of 160,000, 330,000 and 600,000 nodes, respectively. All the nodes were active in the sense that they were all located within the ow domain with none inside the TLP members. This was achieved by building the meshes within a number of separate but inter-connected blocks (29 in the present case). The resulting meshes are highly non-uniform with the greater concentration being near the TLP itself. The parameter that is of primary interest in the engineering design process is the global drag coefcient Cd, dened as: Cd Fx 1 2 rU A 2 o

where Fx is the pressure force in the x-direction and A is the TLP cross-sectional area. The sensitivity of the predicted Cd to the number of nodes used is shown Table 3. Those results were obtained with the RNG model and with the UPWIND scheme. The Reynolds number (based on Uo and D) was 7.36106. A total of 600 iterations were performed to satisfy the tight convergence criterion of the absolute sum of the
Table 3 Variation of Cd with mesh size No. of nodes 160,000 330,000 600,000 Cd 1.39 1.39 1.24 CPU time (h) 4.5 14 36

592

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

normalized eld residuals to fall below 105. Table 3 illustrates the difculty in obtaining genuinely grid-independent simulations of complex three-dimensional ows: the quoted CPU times are for reasonably fast engineering workstations (HP K460 and SGI Origin 2000) and their extrapolation would suggest that computations on a mesh of 1,200,000 nodes would take about 1 week to complete. The hazards of performing an incomplete grid-sensitivity check are also apparent from Table 3: the rst doubling in the number of nodes yielded no change in Cd while a second doubling reduced this quantity by about 11%. The inuence on the predicted global drag coefcient due to the choice of differencing scheme can be seen from Table 4. The results presented there were obtained on the 330,000 nodes mesh, using the RNG model. The Reynolds number was again taken as 7.36106. It is immediately obvious from Table 4 that the inuence of increasing the order of accuracy of the differencing scheme is analogous to that of increasing the number of nodes in the computational domain. This result would suggest that it is preferable to adopt the robust but less-accurate UPWIND scheme and use it in conjunction with the ner grid only because the higher-order schemes tend to be less robust and sometimes harder to converge. We turn next to consideration of the predicted ow eld around the TLP. Figs 2(a,b) show the predicted velocity vectors and contours of uid static pressure at a horizontal cross section half way through the front and aft pontoons (z/B=0.5). The ow accelerates as it passes around the rst circular base and, again, over the second base, leading to the formation of a large region of reversed ow behind the downstream pontoon. The pressure distribution around the upstream pontoon is uniform and positive on the ow side, and negative on the reverse side. This combination suggests that the x-direction force coefcient on this member is large. In contrast, a low value of this coefcient is expected to pertain for the downstream pontoon as the pressures on either side of it are roughly equal. The rapid ow acceleration gives rise to substantial pressure gradients from a location of about 45 from the stagnation point. Within the void bounded by the four pontoons, the ow is dominated by a large recirculation zone with low mean velocities observed throughout. Fig. 3(a) shows the predicted ow eld at a horizontal plane located at half-column length (z=B+H/2). The ow around the upstream column is different from that observed for a single, isolated, cylinder: there is uneven uid acceleration on either side of the column leading to a shift in the location of the separation bubble away from the TLP core and in a marked skewness of the ow directed towards the downTable 4 Variation of Cd with differencing scheme Scheme UPWIND MINMOD SMART Order of accuracy 1 2 3 Cd 1.39 1.35 1.24

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

593

Fig. 2.

Computed velocity (a) and pressure (b) at z/B=0.5.

594

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

Fig. 3. (a,b) Computed velocity and pressure at columns mid-height. (c,d) Computed turbulence kinetic energy and grid.

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

595

stream column. A second recirculation zone is captured behind the downstream column, much smaller in extent than the rst. These changes to the ow eld suggest that the interference effects are quite signicant for this conguration. This is evident from Fig. 3(b) where the pressure contours are presented. The pressure levels at the stagnation point on the forward column are almost twice as high as their levels on the downstream column, while the base pressures are roughly equal. This suggests values of the x-direction force coefcient that are different by a factor of about two. Moreover, close inspection of the contours around the q=90 and 270 points suggest that the y-component force coefcient on the forward column is higher than its value on the downstream column. Fig. 3(c) shows the contours of turbulence kinetic energy at the same horizontal plane. There is almost no turbulence in the incident stream, a consequence of the low turbulence levels specied at inlet to the computational domain and the absence of mean shear. That notwithstanding, it is clear that turbulence is generated within and around the TLP structure due to the shear layers produced around the columns. Note, for example, the high levels of k observed in the shear layer downstream of the separation point on member 9 and in the wake of the TLP though the values there are somewhat inuenced by the relatively coarse grid employed there (see Fig. 3(d)). Fig. 4(a) presents the computed ow eld in a vertical plane passing through the middle of the columns and one of the otation pontoons. Regions of reversed ow are apparent at the upstream corner of the pontoon and, more substantially, in the region in-between the two columns. The latter is centred closer to the pontoon with, possibly, a smaller counterpart near the junction with the free surface. The pressures on both upstream sides of the columns are positive though, as before, much greater on the upstream column. Further indication of the ability of the internal ow to generate and sustain turbulence is evident from Fig. 4(c). Figures 5(a,c) show the computed results in a vertical plane passing through the centres of the front and aft otation pontoons. Flow reversal is obtained in the wakes of both pontoons. The resulting ow is skewed towards the free surface. The turbulence kinetic energy is high everywhere, as might be expected in separated wakes. High levels of k are usually observed as small-scale, high-frequency uctuations, though their intensity may not be sufcient to have adverse implications on the TLP. Note also that the contours of k are seen to rise towards the free surface. In an operational context, an unmanned mini TLP would have to be assisted by a tender vessel during drilling and workover operations. This vessel would normally be moored adjacent to the TLP with the gap between the two being typically of the order of 10 m. The presence of massive ow disturbance within such a small gap is expected to have a signicant impact on the mooring forces. The picture that emerges from the whole-eld predictions presented above is that of a ow eld that is fairly complex in detail but one which remains predominantly aligned with the x direction. It is therefore in the force coefcients appropriate to this direction that the interference effects are expected to be most pronounced. We dene the streamwise force coefcient (Cx) for each member thus: Fx Cx 1 2 rU A 2 o s

596

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

Fig. 4.

Computed velocity, pressure and turbulence kinetic energy at pontoon half-width.

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

597

Fig. 5.

Computed velocity, pressure and turbulence kinetic energy at y=0.

598

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

where Fx is the pressure force on the individual member and As is the area, dened as: Member Pontoons Circular bases Columns As [1,2,4]B(LD) [5,6]DD [9,10]HD

The predicted streamwise force coefcients are presented in Table 5 which also shows their variation with Reynolds number. The most remarkable result seen there is the reduction, by almost a factor of 10, of the loading on the aft pontoon [2] relative to its level on the front pontoon [4] (the number in the square brackets refers to the designation in Fig. 1). Support for this unexpectedly large reduction in Cx is provided by the experiments of Sakamoto and Haniu (1988) on two square-sectioned prisms placed in tandem. Their measurement of Cx for the upstream prism was 1.025 while, for the downstream prism, this quantity was reported to be zero. The ratio of the prism height to width in their tests was 3; that of the centre-to-centre distance to width was 3.25. In the present calculations, the equivalent ratios were 3.5 and 4.5, respectively. This difference in geometry explains, in part, the slight differences between the predicted and measured reductions. In contrast to the large shielding observed for the square prisms, the value of Cx on the downstream circular column is just over half its value on the upstream column. Intuition suggests, and the eld predictions conrm, that the disturbance associated with a circular cylinder is far less than that produced by the square members, especially with respect to the size of the separated ow region behind each member. It is interesting to note that the value of Cx on the upstream column is, at 0.741, quite close to the value of 0.69 reported by Sakamoto and Oiwake (1984) for an isolated cylinder of aspect ratio in the range 16 (the aspect ratio here is 3.26). It is nally worth noting that the present predictions indicate that the ow around the front pontoon [4] remains largely symmetric and that the force coefcient for this member in the vertical (z) direction is of the order of 103. Around the aft pontoon, however, the predictions show that symmetry is destroyed due to the upward deection of the ow giving a vertical force coefcient of 0.113. The upward forces on the remaining members are predicted as negligible. Taken together, these results would imply a denite tendency for the TLP to tilt. We turn next to obtaining comparative estimates of the loads on individual memTable 5 Predicted loading on individual members Re Front column [9] Aft column [10] Front pontoon [4] Aft pontoon [2] 5104 0.74 0.49 1.50 0.15 105 0.68 0.49 1.45 0.14 5.2105 0.63 0.51 1.37 0.14 2107 0.79 0.49 1.39 0.18 6107 0.70 0.44 1.36 0.17 108 0.64 0.51 1.64 0.23

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

599

bers from the use of an engineering design code. For this purpose, we adopt the recommendations put forward in the DNV Design Code (DNV Classication Note, 1993) on the basis that it is a particularly detailed code, and one which provides explicit and quantitative guidance on the force coefcients. The drag force Fx on each member is obtained from Morisons equation: 1 Fx rACxUo|Uo| 2 Here, we take Uo to be 0.95 m/s (which can be taken as the 100 year surface current for West African conditions) and obtain a Reynolds number1 of 8.84106 for the columns and 6.32106 for the pontoons. For smooth circular cylinders, the quoted value of Cx in the DNV guidelines is appropriate to a two-dimensional ow and, for the present Reynolds number, is obtained as approximately 0.7. The guidelines further state that (6.1.13): For several cylinders close together, group effects may be taken into account. If no adequate documentation of group effects for the specic case is available, the drag coefcients for the individual cylinders may be used. For the present purposes the individual drag coefcients have been used. The guidelines also discuss the effect of nite length on the drag coefcient, and for a circular cylinder with a length to diameter ratio of approximately 6, as in the present case,2 the recommended reduction factor is around 0.8 for supercritical ow (DNV Tab. 5.3). This then gives the drag coefcient for the columns as 0.56. In practical calculations, this gure would usually be considered to be somewhat unconservative and a value closer to 0.7 would normally be used for design purposes. For square-shaped pontoons, the DNV guidelines (Tab. 5.2) recommend a Cx value of 2.0 for the case of innite length and zero incidence. For a nite-length pontoon, Section 5.3.4 of the guidelines states that For members with one end abuting on to another member or a wall in such a way that free ow around that end of the member is prevented, the ratio L/D (the aspect ratio) should be doubled for the purpose of determining the reduction factor. When both ends are abuted as mentioned, the shape coefcient C should be taken equal to that for an innite long member. For the present case, a full reduction factor would give a drag coefcient for the pontoons of Cx=1.6 while strict interpretation of Section 5.3.4 would give no reduction at all. However, some modication for nite length should be introduced, as the ow has some freedom to pass around. Here, we adopt a pragmatic approach and assign to Cx the average value of 1.8. With the above assumptions in place, it is now possible to compare the DNV recommended values for Cx with those obtained from CFD. As the CFD values of Cx for the pontoons and columns differ substantially depending on whether or not those members are subject to interference effects, an average value is adopted for
The kinematic viscosity is taken to be 0.94106 m2/s, assuming a sea temperature of 25C. The mean free surface effectively doubles the length of the cylinder.

1 2

600

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

Table 6 CFD and DNV values for Cx Member Column Pontoon CFD 0.56 0.75 DNV 0.56 1.8

the purposes of this comparison. This is shown in Table 6. It is immediately clear from there that the CFD and DNV values for Cx for the circular columns are exactly matched but that there is a large discrepancy between the values pertaining to the square pontoons. This result is attributed to the larger extent of the ow disturbance around the square pontoon, leading to the substantial interference effects discussed earlier. By summing the forces on the individual members it is possible to deduce the global drag coefcient implied by the DNV Code. Table 7 compares the CFD and DNV predictions for this parameter. The 30% difference between the two results is indicative of the extent of conservatism inherent in the use of engineering design codes. Finally, a series of experiments were commissioned to obtain the global drag coefcient on a small-scale model of the same TLP. Details of these experiments can be found in Chaplin et al. (2000). Briey, however, two 170 scale models were built: one made of wood for use in wind-tunnel tests and another made of rigid PVC sheet and pipe for use in towing-tank tests. The global drag force was obtained from the wind-tunnel tests by direct measurement using a calibrated six-component straingauge balance. In the towing-tank tests, the global force was estimated from measurement of the model offset when towed at constant speed. Both tests were conducted in conditions approximating uniform approach ow with low turbulence intensity. The measured values of the global drag coefcient are presented in Table 7. The range of values quoted there for the wind-tunnel tests corresponds to measurements obtained at four orientations of the model, 90 apart. The spread in values is indicative of departures from exact symmetry in the geometry of either the wind-tunnel or the model (or, more likely, in both). The spread in values of the towing-tank tests corresponds to repeat runs conducted over a period of days. The global drag coefcients obtained in the towing-tank tests are expected to be higher than those
Table 7 Predicted and measured global drag coefcient Technique CFD (k) CFD (RNG) DNV Wind-tunnel Towing-tank Re 6107 6107 8.84106 1053.5105 2.51049104 Cd 1.32 1.39 1.61 1.020.93 1.61.4

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

601

obtained in the wind-tunnel and by CFD due to blockage effects, increased drag due to upwelling and the presence of a surface boundary layer in the wind tunnel that is absent in the towing tank tests. The CFD results are quoted for both the standard and for the RNG forms of the k model. The difference between the two amounts to about 5%. The difference between the values obtained with two established experimental techniques is of the order of 50%.

4. Conclusions The CITY3D CFD package was used to predict the ow eld around a full-scale mini TLP in a steady, turbulent current at a range of Reynolds numbers. The sensitivity of the predictions was quantied with respect to the total number of grid nodes used and the choice of scheme employed for discretizing the governing equations. The difculty of obtaining truly grid-independent results for the present complex three-dimensional ow, with acceptable computer run times, was noted. Inspection of the computed ow eld suggests the presence of substantial interference effects associated with interactions with wakes produced by upstream TLP members. Quantitatively, it was found that the drag coefcient on the downstream pontoon was only one tenth of its value for the upstream member. A reduction is also obtained for the shielded circular columns but at a much reduced level. A comparison was made with the values recommended by the DNV design code. It was found that the CFD and DNV values for the drag coefcient were exactly matched for the circular columns but quite different for the square pontoon with the DNV value approximately 23 times higher than the CFD result. Consequently, the DNV result for the TLPs global drag coefcient was about 30% higher than the CFD value. Comparisons with experimental data obtained from wind-tunnel and towing-tank tests lend support to the judicious use of CFD as a tool for practical design applications.

Acknowledgements The development of CITY3D was sponsored by EPSRC (Grant GR/J24003) and jointly funded with: Aker Engineering, Amoco (UK) Exploration Company, BP Exploration Operating Co. Ltd, Brown and Root Ltd, Exxon Production Research Co., HSE, Shell UK exploration and Production, STATOIL and Texaco Britain Ltd. The present investigation was funded by STATOIL.

References
Chaplin, J.R., Younis, B.A., Teigen, P., 2000. Wind-tunnel, towing-tank and CFD estimates of the current loading on a model TLP. In preparation. DNV Classication Note, 1993. Environmental conditions and environmental loads, No. 30.5.

602

B.A. Younis et al. / Ocean Engineering 28 (2001) 585602

Launder, B.E., Spalding, D.B., 1974. The numerical computation of turbulent ows. Computer Methods in Applied Mechanics and Engineering 3, 269. Patankar, S.V., Spalding, D.B., 1972. A calculation procedure for heat, mass and momentum transfer in three-dimensional parabolic ows. International Journal of Heat and Mass Transfer 15, 1787. Sakamoto, H., Haniu, H., 1988. Aerodynamic forces acting on two square prisms placed vertically in a turbulent boundary layer. Journal of Wind Engineering and Industrial Aerodynamics 31, 4166. Sakamoto, H., Oiwake, S., 1984. Fluctuating forces on a rectangular prism and a circular cylinder placed vertically in a turbulent boundary layer. ASME Journal of Fluids Engineering 106, 160166. Speziale, C.G., 1991. Analytical methods for the development of Reynolds-stress closures in turbulence. Annual Review of Fluid Mechanics 23, 107157. Yakhot, V., Orszag, S.A., Thangam, S., Gatski, T.B., Speziale, C.G., 1992. Development of turbulence models for shear ows by a double expansion technique. Physics of Fluids A4, 1510. Younis, B.A., Przulj, V., 1996. CITY3D: A Users Manual. Final Report on Project: Prediction of mean and uctuating wind loads on offshore stuctures, EPSRC Research Grant No. GR/J24003.

S-ar putea să vă placă și