Вы находитесь на странице: 1из 15

J Mar Sci Technol (2004) 9:143157 DOI 10.

1007/s00773-004-0180-z

Original articles A CIP-based method for numerical simulations of violent free-surface ows
Changhong Hu and Masashi Kashiwagi
Research Institute for Applied Mechanics, Kyushu University, 6-1 Kasuga-koen, Kasuga 816-8580, Japan

Abstract A CFD model is proposed for numerical simulations of extremely nonlinear free-surface ows such as wave impact phenomena and violent wavebody interactions. The constrained interpolation prole (CIP) method is adopted as the base scheme for the model. The wavebody interaction is treated as a multiphase problem, which has liquid (water), gas (air), and solid (wave-maker and oating body) phases. The ow is represented by one set of governing equations, which are solved numerically on a nonuniform, staggered Cartesian grid by a nite-difference method. The free surface as well as the body boundary are immersed in the computation domain and captured by different methods. In this article, the proposed numerical model is rst described. Then to validate the accuracy and demonstrate the capability, several twodimensional numerical simulations are presented, and compared with experiments and with computations by other numerical methods. The numerical results show that the present computation model is both robust and accurate for violent free-surface ows. Key words CIP method Violent free-surface ow Multiphase computation

1 Introduction In rough seas, ships or ocean structures may experience highly nonlinear phenomena such as slamming, water on the deck, wave impact by green water, or capsizing due to large-amplitude waves. Such extremely nonlinear wavebody interactions sometimes cause localized structural damage, and a quantitatively precise estimation method for wave loads is therefore necessary. To date, experiments have been the most practical way to study these problems, and numerical simulations have been restricted to a few very simple cases. The

Address correspondence to: C. Hu (hu@riam.kyushu-u.ac.jp) Received: October 27, 2003 / Accepted: May 11, 2004

major difculty in computing such wavebody interaction problems arises from their extremely complicated hydrodynamic phenomena. For example, it is necessary to treat highly distorted or broken free surfaces, and to consider the effect of the compressibility of water or the elasticity of the body for the case of impacting water. Conventional numerical analysis methods, such as the boundary element method for potential ows or the CFD method using curvilinear grids adapted to both a free surface and a body surface, are not applicable to extremely nonlinear problems, although some efforts are seen to take into account wave breaking by suitable modeling (see Muscari and Di Mascio1 and Olivieri et al.2). Recently there were some papers in which free surface was captured as a part of solution, thus being capable of computing much more complex free surfaces than conventional surface-tted methods. Park et al.3 used the MAC method to compute wave-current-body interactions. Andrillon and Alessandrini4 used the VOF method to compute bow breaking waves. But it is still true that complex free surface phenomena still remain as a challenge to CFD. A new numerical method was needed. This new method should not only be able to handle complicated phenomena associated with wave body interactions, but should also be a relatively simple scheme which can perform three-dimensional simulations in an acceptable spatial and temporal resolution at reasonable cost. Here, we propose a new CFD simulation approach for extremely nonlinear free-surface problems. This approach is a nite-difference method based on the constrained interpolation prole (CIP) algorithm.5,6 The key points of the CIP method can be summarized as (1) a compact upwind scheme with subcell resolution for the advection calculation, and (2) a pressure-based algorithm that can treat liquid, gas, and solid phases, irrespective of whether the ow is compressible or incompressible, by solving one set of governing equations. The method for hydrodynamic problems using the

144

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

pressure-based algorithm is called the CIP combined and unied procedure (CCUP) method. The CIP method is a Eulerian approach on a regular, stationary Cartesian grid with multiphase computation. This has the multiple capabilities which we need, e.g., handling the complicated free-surface geometry and treating violently moving oating bodies. Furthermore, since no remeshing calculation is required, the computation time can usually be shortened for timedependent problems. The Eulerian approach was rst proposed in the marker and cell (MAC) method7 nearly 40 years ago. Since then, this approach has been developed into a number of methods with different interface/ front-capturing techniques, such as the volume of uid (VOF) method8 and the level set method.9 The motivation for applying the CIP method as the base scheme of our model originates from the two key features that have just been described. In particular, the unied procedure coupled with the CIP scheme (CCUP), which is applied in our CFD model, has the advantage of simplicity of coding because only one set of equations is used for the whole ow eld, including the region inside solid bodies. Although extra computation time is required for the region inside solid bodies, Xiao10 has shown by a couple of 3-D computations that this unied procedure is relatively easy to implement numerically with parallelization techniques which take advantage of the rapid progress in the computational power of modern computers. The current model with the CIP method is quite different from many other CFD methods for wave body interaction problems in marine engineering. For example, a numerical wave tank (NWT) problem is treated as a multiphase problem in our model, which includes liquid (water), gas (air), and solid (wave-maker and oating body) phases. The motion of all materials is numerically solved by one set of hydrodynamic equations in a xed Cartesian grid. Then the free surface and the body boundary are treated as two types of phase interface: the interface between a liquid and a gas, and the interface between a solid and a uid, respectively. Is the CIP method capable of quantitative prediction of the solution to extreme wavebody interaction problems? To answer this question is the main purpose of this article. Some new implementation details for applications of the CIP method will be described, and the focus will be placed on investigating several features of the CIP method that can be important for the problems of interest, but which have not been studied before. These features include the inuence of a term of compressible effect in the Poisson equation for incompressible computations, and the precision in the calculation of hydrodynamic forces acting on a oating body. This article is organized as follows. Section 2 introduces the governing equations and numerical imple-

mentation of the method. The CIP scheme and the CCUP method are briey described. The interface capturing method, the calculation method for hydrodynamic forces acting on a oating body, and the method for constructing a numerical wave tank are outlined. In Sect. 3, the current numerical method is applied to three free-surface problems: propagation and reection of a solitary wave, a dam-breaking test in a rectangular tank, and a forced oscillation test with a wedge-type oat in a two-dimensional numerical wave tank. In each computation, a grid renement test is rst carried out, and then a validation test compares this with theoretical or measured results. Some special features associated with the CIP method are discussed using the numerical results. The article ends with some conclusions.

2 CIP-based nite-difference method 2.1 Governing equations Assuming that there is no temperature variation in the problem, we start from the following equations for compressible uid:
r r u + ui = -r i t xi xi

(1)

ui u 1 s ij + uj i = + fi t x j r x j

(2)

where r is the density, ui (i = 1, 2, 3) is the velocity component, and sij is the total stress. For a Newtonian uid, the total stress can be written as sij = -pdij + 2mSij - 2mdijSkk/3, where Sij = (ui/xj + uj/xi)/2 and dij denotes Kroneckers delta. The second term on the right-hand side of Eq. 2 represents the body force, such as the gravity force, etc. As there is no temperature variation, the equation of state (EOS) of the problem can be written as p = f(r). Theoretically, after the density r is solved by Eq. 1, the pressure p can be determined. For incompressible or nearly incompressible ow, however, as the sound speed Cs = p r becomes very large, a small numerical noise from the density will produce a large pressure variation. This means that it is numerically difcult to obtain an accurate calculation of pressure using the EOS. However, this difculty can be overcome by calculating the pressure independently. Applying the EOS to Eq. 1, the pressure equation can be obtained as
p p ui 2 (3) + ui = - rCS t xi xi Equations 13 are the governing equations for hydrodynamic problems, and will be solved numerically by a fractional step method.

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

145

2.2 The fractional step approach By applying the fractional step approach, the numerical solution of the governing equations can be divided into the following three steps. 1. Advection phase:
r * - rn r n + uin =0 xi Dt

By differentiating Eq. 11 with respect to the spatial coordinates, we obtain the transportation equation of ji. u j j i j H + uj i = - jj t x j xi xi (12)

(4)

ui* - uin uin + un =0 j Dt x j


p * - pn pn + uin =0 Dt xi

(5)

As shown in the previous subsection, computation of Eq. 12 can also be divided into two steps, an advection phase and a nonadvection phase. The nonadvection phase calculation will be included in nonadvection phase (i). For the advection calculation of Eqs. 11 and 12, the following semi-Lagrangian procedures are used. ) c * (x) = c n (x - uDt ) ) j i* (x) = j in (x - uDt ) (13) (14)

(6)

2. Nonadvection phase (i):


1 ui** - ui* 2 m * - d ij Skk * + fi = * Sij Dt 3 r x j

(7)

3. Nonadvection phase (ii):


r n +1 - r * u n +1 = -r * i Dt xi

(8)

uin +1 - ui* * 1 pn +1 =r * xi Dt
pn +1 - p* u n +1 = - r *Cs2 i xi Dt

(9)

(10)

The advection phase computation of Eqs. 46 is conducted by the CIP method. The nonadvection phase is divided into a state-related part to reect the uid compressibility, denoted by nonadvection phase (ii), and the remaining part denoted by nonadvection phase (i), which includes a viscous term and a source term. The fractional steps in the present numerical method are arranged in the order advection phase, nonadvection phase (i), and nonadvection phase (ii). Xiao10 has shown that this procedure yields rst-order accuracy in time-integration of the governing equations.

) where c n is an interpolation approximation to cn, and ) ) j in = c n /xi. For each computational cell, the interpo) lation function c n can be constructed using a cubic polynomial in which the coefcients of the polynomial are determined from the continuity condition imposed ) on cn and j in at the grid points. For a multidimensional case, several forms have been developed for the cubic polynomial,5 and details of the two-dimensional CIP scheme that is used in the present numerical model are described in Appendix A. As the advection of spatial gradients in each computation cell can be solved, and only the information (value and its spatial gradients) at the grid points of one cell is needed for the interpolation function, the CIP scheme has both a subcell resolution feature and a compact structure. Therefore, for a multiphase computation in which there are discontinuities or large gradients for physical quantities at the interfaces, the CIP scheme can maintain sharpness better than other upwind schemes.
2.4 The nonadvection phase calculation For nonadvection phase (i), the Euler explicit scheme is used for the time-integration. The physical nature involved in nonadvection phase (ii) is uid compressibility. Since resolving the acoustic wave propagation requires extremely small time steps and is beyond our interest, the implicit scheme is used for the timeintegration. Taking the divergence of Eq. 9 and substituting uin+1/xi using Eq. 10, we obtain the pressure equation

2.3 CIP method The basic idea of the CIP method is that for advection computation of a variable c, not only the transportation equation of c, but also the transportation equation of its spatial gradient, ji = c/xi, are used. Here, c represents each of r, ui, and p in Eqs. 1, 2, and 3, respectively. The transportation equation of c can be written as

c c + ui =H t xi

(11)

xi

1 r n +1 pn +1 - p* 1 u ** i + = 2 2 * * x t x D r r C t D i i s

(15)

146

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

This is a Poisson-type equation for the pressure calculation. The rst term on the right-hand side is the contribution of compressibility. As Eq. 15 is valid for the liquid, gas, and solid phases, we can obtain the pressure eld in the whole computation domain by solving this equation. Then the boundary conditions for the pressure at the interface between different phases are not necessary, and a fast solver or parallel computing technique can easily be applied. This is a very important feature, because the calculation of Eq. 15 is generally the most time-consuming part of this type of computation. Another advantage of Eq. 15 is that it provides a very simple and robust way to calculate hydrodynamic forces on a moving body in a xed Cartesian grid, as will be described in Sect. 2.6. For a perfect incompressible uid, we can assume that Cs = . Then Eq. 15 takes the form xi 1 r n +1 1 u ** i = * x t x D r i i (16)

This is a conventional Poisson equation for incompressible ows. We will use Eq. 16 for most of the computations shown here because under the assumption of an incompressible uid, instead of solving Eq. 1, the density can be obtained directly by Eq. 19, as shown in the next subsection. A comparison will be made in Sect. 3.2 to check the difference between computed results using Eqs. 15 and 16. 2.5 Interface capturing method The moving body boundary and the free-surface boundary are distinguished by a density function fm, which is solved by the equation
fm f + ui m = 0 (17) t xi For the numerical wave tank problem shown in Fig. 1, m = 1, 2, and 3 denote liquid, gas, and solid phases, respectively. There are two types of interface that need to be captured in the numerical simulation, i.e., the interface besolid phase x3 x1 wave maker gas phase damping zone

tween gas and liquid, the so-called free surface, and the interface between solid and uid, such as a oating body boundary. As the behaviors of the two types of interface are quite different, we will use different capturing methods for them. The free surface is determined by solving Eq. 17 with the CIP method. Like most of the Eulerian methods, the original sharp phase interface may become a layer with nite thickness due to the numerical diffusivity. However, owing to the subcell resolution feature of the CIP method, the thickness grows very slowly as the computation proceeds. Therefore, in many cases it is considered that this degree of interface diffusion is acceptable for computations when the time corresponds to actual experiments. A continuum surface force (CSF) model11 is also incorporated in the present computation code to approximate the effect of surface tension, in which the surface tension is considered as a continuous three-dimensional effect across the interface. The surface tension force included in the body force term fi of Eq. 2 has the expression fS = -

s S f1 f1 r f1

(18)

Here, ss is the surface tension coefcient. By applying this model, no extra treatment is needed even when the interface is topologically distorted. For the oating body boundary, we only consider the rigid body case. Instead of the computation using Eq. 17, a direct computation method has been developed to determine the density function for the solid phase f3. The basic idea of this method is to map the geometry information of a moving body to a xed Cartesian grid. A detailed description of the method for the twodimensional case is shown in Appendix B. The advantage of this method is that the solid body boundary position can be obtained accurately without any numerical diffusion. After the density function for all phases is determined, the physical properties for each computation cell can be calculated by the equation
3

l = fm l m
m =1

(19)

floating body

where l denotes the viscosity, sound speed, etc. Under the incompressible uid assumption, the density can also be determined by Eq. 19. 2.6 Hydrodynamic forces on a oating body The hydrodynamic force acting on a oating body, Fi, can be calculated by integrating the pressure and skin

d liquid phase

Fig. 1. Schematic view of a numerical wave tank

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

147

friction along the body surface. For the rigid body case, the formulation is written as
Fi = Fi
( p)

+ Fi

(v)

s e b t stream boundary (x1 < x1 < x1 , x3 < x3 < x3 ), and an articial damping force is added to the body-force term of Eq. 2, which is expressed as

= (- pd ik )nkdA + 2 mS ik nkdA
A A

(20)

where Fi(p) represents the force due to the pressure, and Fi(v) represents the force due to the friction. A denotes the surface of the oating body, and ni is the i-th component of the outward unit normal vector. In this numerical model, the forces can also be calculated easily by integration over the whole computation domain. Applying Gauss theorem to Eq. 20, we obtain
Fi = Fi
(p)

s f x - x1 x3 - x3 fdi = a 1 1 ui d i 3 s e t b - x3 - x1 x3 x1

(22)

s e b t Here, x1 , x1 and x3 , x3 denote the positions of the damping-zone boundaries, and xf3 is the average free-surface position. The constants in Eq. 22 are taken as a = 0.5/Dt, m = 4, and n = 1 for the computations shown here according to several previous test calculations.

+ Fi

(v)

= -
V

(2 mSik ) p dV + dV xk xi V

3 Numerical results (21) Several two-dimensional numerical simulations were carried out by the proposed numerical method. These examples were chosen both for validation and to illustrate the numerical properties of the method. In the computations, except where otherwise noted, the liquid phase is treated as water and the gas phase is treated as air. Their density and viscosity are r1 = 103 kg m-3, m1 = 10-3 kgs-1 m-1, and r2 = 1 kg m-3, m2 = 10-5 kgs-1 m-1, respectively. 3.1 Propagation and reection of a solitary wave The propagation of a solitary wave in a shallow channel and the reection of such a wave from a vertical wall are considered as good test problems to validate a numerical method on free-surface ows, and are taken as our rst numerical example. The denitions of the problem are the same as in Tang et al.12 The computation conditions are shown in Table 1, where H0 and d denote the wave-height of a solitary wave and the reference depth, respectively. As the present computation is for twophase ow, which is different from most previous calculations for a single liquid phase, we dene the Reynolds number for a solitary wave as

= -
W

(2 mSik ) p f dW + f 3 dW xk xi 3 W

where V and W denote the space occupied by the oating body and the whole computation domain, respectively. As the pressure, the velocities, and the density function for all computation cells are obtained by the procedure described in the previous subsections, we can directly calculate the hydrodynamic forces on a moving body in the xed Cartesian grid using Eq. 21. The advantage of using Eq. 21 is that we do not need to know the exact position and orientation of the boundary surface in order to calculate the unit normal vector. However, it is necessary to check the accuracy of the force computation by Eqs. 20 and 21, and this will be done through a numerical example in Sect. 3.3. 2.7 Absorbing boundary condition for NWT In order to perform simulations in a numerical wave tank with a nite computation domain over a long period of time, a nonreection boundary condition is required at the downstream boundary (see Fig. 1). In this study, an articial damping zone is placed at the down-

Table 1. Computation condition for a solitary wave Solitary wave propagation H0/d Re m1 (kg-1 m-1) Grid number Time-step Solitary wave reection H0/d Re Grid number Time-step 0.2 500 50 000 0.198 0.00198 280 (horizontal) 88 (vertical)

50 1.98

Dt g d = 5 10 -4
0.1, 0.2, 0.3, 0.4, 0.5 50 000 300 (horizontal) 136 (vertical)

Dt g d = 5 10 -4

148

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

10

10

20

Fig. 2. Grid convergence test for an inviscid solitary wave propagation computation

10

10

20

Re =

r1dU r1d gd = m1 m1

(23)

where r1 = 103 kg m-3, d = 0.1 m, and U = gd is the phase speed. We can then obtain different values of Re by changing m1, as shown in Table 1. Our computations show that the physical properties of the gas phase may also affect the results. In order to compare these with other relative single-phase results, the density and viscosity of the gas were chosen as r2 = 1 kg m-3 and m2 = 0, respectively. The computation grid shown in Table 1 is determined by a grid renement test, in which an inviscid solitary wave is simulated. Such an inviscid case is convenient for code verication because the solitary wave can be known exactly by theoretical analysis, and the wave prole does not change during propagation. In the computation, we used the slip condition for the bottom boundary and let m1 = 0. Figure 2 shows the result of the grid renement test. Six types of grid were used. The computed wave heights by grids 4, 5, and 6 are almost the same, i.e., grid convergence is achieved. It is also clear from the gure that the variable grid (grid 6), in which the grid points are concentrated near the free surface and the bottom boundary in the vertical direction and in the region containing the wave crest in the horizontal direction, gives a converged result although the grid number is much less than the uniform grid (grid 5). This also conrms that our variable mesh treatment is successful, and thus grid 6 was used for the following computations. On the other hand, even for the converged results, a slight attenuation of the wave height is found in the computation. This can be explained by the numerical diffusion of the present method. The computations were then carried out to investigate the effect of viscosity on the propagation of a solitary wave by using the different values of m1 shown in

Fig. 3. Solitary wave propagation for different Reynolds numbers

13

Fig. 4. Damping of wave height with distance for different Reynolds numbers. CIP, constrained interpolation prole

Table 1. A no-slip condition was applied on the bottom boundary. Figure 3 shows computed free-surface elevations between t/t0 = 0 and t/t0 = 12. The computed free surface is denoted by the line of f1 = 0.5. It can be seen that the amplitude of wave decays due to viscous damping and the decay in amplitude decreases with increasing Reynolds number. In Fig. 4, the calculated attenuation of the wave height with distance traveled by the wave is compared with the theoretical results of Mei.13 The result obtained by the computation, including the surface tension effect,

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

149

14

15 16

14

Fig. 5. Time variation of run-up at the wall

Fig. 6. Maximum run-up versus incident wave amplitude

which is modeled by Eq. 18, is also shown in Fig. 4. It can be seen that all computed wave heights are lower than the corresponding theoretical results. The main reason for this is that the theoretical analysis only considers the viscous effect at the bottom of the channel, whereas the present computations are for the viscous ow in the whole uid region. It can also be seen that the effect of surface tension slightly increases the attenuation of the wave height, but we note that the effect of surface tension depends on the spatial scale of the problem. Another computation on a solitary wave with a larger reference depth has shown a much smaller difference between the results with and without the surface-tension effect. As the surface-tension effect is not the most important issue of the present research, the other computations in this paper were performed without the surface-tension model. The second computation on a solitary wave is a reection of a wave from a vertical wall. The computation grid shown in Table 1 is determined by a similar check to the one shown in Fig. 2. The computed time variation of wave run-up at the wall is shown in Fig. 5, and good comparison is obtained for the run-up stage with the inviscid theory of Power and Chwang.14 The difference in the run-down stage is due to the viscous effect at the vertical wall that reduces the run-down speed in the present fully viscous computation. Figure 6 shows a comparison of the computed maximum run-up versus the incident wave amplitude with some available experimental and theoretical results. By comparing these with the experiment, we can see that the prediction by the present numerical method is better than that of other theoretical models. 3.2 Dam-breaking problem The water ow after the sudden break of a dam is widely used as a test problem for checking the computation accuracy for a largely distorted free surface because of its simple initial and boundary conditions.

12cm

12cm z y x 68cm 50cm

A 1cm

Fig. 7. Schematic view of the dam-breaking experiment

The development of water ow along the oor after the sudden break of the dam has been a conventional target for numerical studies, but our research interest is on the second stage of the ow development, i.e., the ow after impact onto the vertical wall. In the second stage, overturning and breaking of the free surface as well as air entrapment are observed, and the computation of these complicated phenomena is a more challenging subject. 3.2.1 Experiment A laboratory experiment was conducted at the Research Institute for Applied Mechanics, Kyushu University. The geometry of the experimental set-up is shown in Fig. 7. As this is a small-scale tank, the viscous effect at the tank walls might be an important factor. The water in the tank is colored, and the temperature of the water is adjusted to be the same as that of the surrounding air in order to reduce the temperature effect on the pressure sensor. A pressure sensor with diameter of 0.8 cm is installed on the right-hand vertical wall at point A in Fig. 7. In the experiment, the pressure is measured and the development of the free surface is recorded by a high-speed digital video camera. The experiment was repeated eight times. All pressure data measured at point A are shown in Fig. 8 by circular symbols, and the mean value is shown by a solid line. Although a certain scatter is seen in the measured

150

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

grid 1 grid 2 grid 3

0.2

0.15

0.1

0.05

Fig. 8. Pressure measured in the experiment shows the scatter of all measured data. Solid line, mean value; circular symbols, the measured data in all tests

0 0.1 1 1.1

x (m)
Fig. 10. Free-surface prole at t = 0.7 s. Computations using different grids

Fig. 9. Time variation of the pressure at point A. Computations using different grids. The pressure eld was solved using Eq. 16

data, the repeatability of the experiment seems good, and the measurements are considered to be reliable for the purpose of validation. The mean value is used for a comparison with the numerical simulation. 3.2.2 Grid renement test The computation conditions were set as in the experiment. A variable grid was used for the computation, in which the grid points are concentrated near the oor and the right-hand wall. To perform a grid renement test, we use three grids, 1, 2, and 3, as indicated in Fig. 9. The minimum grid spacing varied from 0.8 mm to 0.2 mm. Figure 9 shows a comparison of the time series of the pressure at point A, for which the computations were performed using three different grids. The pressure eld was solved by the incompressible Eq. 16. Good general agreement was found for all the computations. However, for computations with grids 2 and 3, the position of the rst peak and its value agree well with the experiment, whereas with grid 1, the coarse grid, a delay for the rst peak was observed and the peak value was

lower than the measured value. In the result with the ne grid (grid 3), a strong oscillation at the rst peak and oscillations around the second peak were also found. The reason for this is not very clear at present. From Fig. 9, it can also be seen that the prediction of the pressure in the latter stage (t > 0.75 s) is poor for all three grids. In the experiment, the ow at this stage is three-dimensional and severely broken. Therefore, we consider that this disagreement is due partly to the 2-D computation, and partly to the smearing of the computed free surface in this stage. Figure 10 shows free-surface proles near the corner at t = 0.7 s obtained using different grids. The computed free-surface prole with grid 1 is quite different from the other two results. Since the difference between the results with grids 2 and 3 is small for both the pressure and the surface prole, it was considered that grid 2 is sufcient, and that was used for the following computations. Figures 11 and 12 are the computed time variation of the free surface and the pressure distribution, respectively. Grid 2 is used for the computation. In Fig. 11, the complicated phenomena of the second stage of ow development can be seen; the ow hits the vertical wall, overturns, and hits the free surface again with a breaking of the free surface as well as air entrapment. In Fig. 12, the contour ranges from p - patm = 100 Pa to 1500 Pa at increments of 100 Pa. Here, patm means the initial pressure of the atmosphere. It was found that a very large spatial variation appears near the lower part of the vertical wall at the time of water impact. 3.2.3 Effect of compressibility The CIP method was originally developed for compressible ows. Later improvements made it computationally applicable to incompressible problems as well.

z (m)

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows


t = 0.0 sec t = 0.5 sec

151

t = 0.1 sec

t = 0.6 sec

t = 0.2 sec

t = 0.7 sec

Fig. 13. Time variation of the pressure at point A. Pressure computed using different types of Poisson equation
t = 0.3 sec t = 0.8 sec

t = 0.4 sec

t = 0.9 sec

Fig. 11. Time-series of computed free-surface development


t = 0.0 sec t = 0.5 sec

Fig. 14. Time variation of the pressure at point A. Computations using different velocity boundary conditions at the tank walls. BC, boundary condition

t = 0.1 sec

t = 0.6 sec

t = 0.2 sec

t = 0.7 sec

t = 0.3 sec

t = 0.8 sec

t = 0.4 sec

t = 0.9 sec

Fig. 12. Time-series of computed pressure distribution

By using Eq. 15, both compressible and incompressible ows can be solved by the same scheme. However, under the assumption of an incompressible uid, we tend to use Eq. 16 instead of Eq. 15 to calculate the pressure eld in order to reduce the computation time. Then a natural question is how different the results of the impact pressure are by using Eq. 15 or 16. Figure 13 shows a comparison of the computed pressures at point A obtained with Eqs. 15 and 16. The sound speed used for the computation with Eq. 15 was Cs = 330 m/s for gas and Cs = 1400 m/s for liquid. Differences were found in both the position of the rst peak and the peak value. A better prediction was obtained with Eq. 16, which is the Poisson equation for incompressible ow. On the other hand, the computation with Eq. 16, the Poisson equation with a compressible term, gives the result without oscillation. We may conclude that the compressible term in the Poisson equation stabilizes the pressure calculation for surface-impact problems. Since using Eq. 15 requires extra effort in calculating the density, the computations shown here were carried out using Eq. 16, the incompressible version of the Poisson equation, when there is no other explanation.

152
0.04

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows


0.04

no slip wall BC
0.03

no slip wall BC

10

40

600

0 50 0

0.03

z (m)

30

0.02

0.02

30 0

0.01

2 00

0.01

3 00
5 00
600
0 1.17 1.18

500
0 1.14 1.15 1.16 1.17 1.18

1.14

1.15

1.16

x (m)

x (m)

0.04

0.04

slip wall BC
0.03

slip wall BC

20

400
0.03

500

z (m)

0.02

10
20 0

0 30

0.02

0 40 0 50

0 70

80

0
0.01

0.01

60

0 1.14 1.15 1.16 1.17 1.18

90

0 1.14 1.15 1.16 1.17 1.18

x (m)

x (m)

Fig. 15. Velocity elds at t = 0.4 s. Computations using different velocity boundary conditions at the tank walls

Fig. 16. Pressure elds at t = 0.4 s. Computations using different velocity boundary conditions at the tank walls. The unit of the contour labels is Pa.
z 0.5a a a x

3.2.4 Effect of wall boundary condition For a small-scale problem such as the current dambreaking experiment, the viscous effect on the wall must be important. A numerical simulation of this problem would be affected by using a slip or no-slip boundary condition at the wall. The results shown in Sect. 3.2.2 and 3.2.3 were obtained using the no-slip condition, and we shall illustrate that the slip condition is not good for this example. Figure 14 shows the time variation of the pressure at point A obtained by using different wall boundary conditions. The computation with a no-slip boundary condition gives a much better result than that with a slip boundary condition. The reason can be found in Figs. 15, 16, in which the velocity elds and pressure elds, respectively, are shown. The vortex near the corner is clearly seen in the no-slip result, but not in the slip result. Correspondingly, the pressure elds are quite different in the two computations. 3.3 Hydrodynamic force on a heaving wedge The third example tested is the computation of wave body interactions. It is well known that a successful

b=2.5a

Fig. 17. Schematic view of the wedge for the forced heaving computation

prediction of a oating body in waves requires an accurate calculation of the hydrodynamic forces. Here, we will check the numerical accuracy of the two force calculation methods described in Sect. 2.6. 3.3.1 Three-phase computation The computation example is a forced oscillation in heave with a wedge-type oat, as shown in Fig. 17.

z (m)

600

z (m)

20 0

4 00

20 0

500
40 0

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows Table 2. Computation condition for the forced oscillation test Wave maker Geometry Motion Amplitude Wave number Water depth Grid and time-step Grid number Min grid spacing Time-step a = 0.0792 m z(t) = Z sin(2pt/T) Z/a = 0.6 Ka = w 2a/g = 0.2, 0.4, 0.6, 0.8, 1.0 d/a = 7.6 460 (horizontal) 165 (vertical) Dx/a = 0.02, Dz/a = 0.02 Dt/T = 5 10-4

153

velocity

pressure
40

40

80
120 160

80

200 240

12 16
200 240

Fig. 19. Computed ow eld at t/T = 10 near the wedge. The unit of the contour labels is Pa
(v) force Fz(p) and a friction-related force Fz , which are obtained by the volume integration method (Eq. 21). For the pressure related force, computations using grids 2, 3, and 4 resulted in a converged time variation. On the other hand, the friction-related force increases with increasing grid density. However, as the pressurerelated force is 1000 times larger than the frictionrelated force by the present computation, grid 3, which is considered to be sufcient for an accurate calculation of the pressure-related force, is used for the following computations. In order to show the concept of our numerical model, a snapshot of the computed ow eld (velocity and pressure) near the wedge is shown in Fig. 19. The computed free surface is represented by the line f1 = 0.5. The thin viscous boundary layer near the boundary of the wedge is unclear owing to the boundary-embedding treatment. It should be pointed out that using the present grid, the boundary layer cannot be fully resolved, i.e., the viscous effect is only approximately considered by the numerical model. We will check whether such an approximation of the boundary layer affects the calculation accuracy of the hydrodynamic force on the body in the following section.

Fig. 18. Comparison of the time variation of the hydrodynamic forces acting on the wedge for computations using different grids. Ka = 0.4

For this problem, a laboratory experiment was conducted by Yamashita,17 and a numerical simulation by the potential theory was performed by Kashiwagi.18 The characteristic length for the oat is a = 0.0792 m, and the other computation details are shown in Table 2. The wavenumber is dened as Ka = w2a/g, where w is the frequency of the forced oscillation. We consider that this problem is a good example of a three-phase computation to check the present numerical model, because the computation domain includes a liquid phase (water), a gas phase (air), and a solid phase (oat). A grid convergence test was carried out rst. Figure 18 shows a comparison of the time variation of the computed vertical force acting on the wedge using four types of variable grid. The wavenumber is Ka = 0.4. The vertical force is divided into a pressure-related

154

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

present CIP

BEM (Kashiwagi17) Exp. (Yamashita18)


Linear Z = 0.2 a Z = 0.6 a Z = 0.2 a Z = 0.4 a Z = 0.6 a

Z = 0.6 a

Ka

Fig. 20. First-order hydrodynamic forces on a wedge

Fig. 22. Comparison of the time variation of the hydrodynamic forces acting on the wedge for computations using different force calculation methods. Ka = 0.4

Exp. (Yamashita ) BEM (Kashiwagi ) present CIP

17

18

F2 / 2 gZ 2

Z = 0.2 a Z = 0.4 a Z = 0.6 a

Z = 0.2 a Z = 0.6 a

Z = 0.6 a

The experiments were carried out for three different oscillation amplitudes, i.e., Z/a = 0.2, 0.4, and 0.6, and the nonlinear computations by BEM were for Z/a = 0.2, 0.6, and a linear one for the case of innite water depth. The present computation was conducted for the largest amplitude of Z/a = 0.6 only. Fourier analysis of the computed results is carried out in the same way as that used in the experiments. As the rst-order hydrodynamic forces, we show the heave-added mass (A33) and the damping coefcient (B33), and as the second-order hydrodynamic forces, we show the amplitude (F2) and the phase lead (d2) of the time-varying force with the second harmonic of the oscillation frequency and the time-averaged steady force (F0). The denition of these quantities can be found in the reference paper.18 Both the rst-order (Fig. 20) and second-order (Fig. 21) hydrodynamic forces obtained by the present numerical model compare well with the experiments and the nonlinear BEM results. However, we should note that the predicted damping force coefcient B33 is slightly lower than in the experiment. This is due partly to the incorrect prediction of the skin friction because of the insufcient resolution of the boundary layer. 3.3.3 Two force calculation methods As described in Sect. 2.6, the hydrodynamic forces acting on a oating body can be obtained by two methods: the surface-integration method represented by Eq. 20 and the volume-integration method represented by Eq. 21. By the surface-integration method, we mean integration along the body surface, and by the volume-

-F0 / 2 gZ 2

2 ( deg. )

Ka

Fig. 21. Second-order hydrodynamic forces on a wedge

3.3.2 Hydrodynamic forces Figures 20 and 21 show comparisons of the hydrodynamic force coefcients acting on the wedge-type oat with a half-beam over draft ratio (a/b) equal to 0.4.

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

155

(v) whereas Fz obtained by the surface-integration method is larger than that obtained by the volume-integration method. As discussed in relation to Fig. 18, grid convergence cannot be achieved for the friction force calculation. Therefore, this discrepancy between the two methods is mainly due to the inadequate grid resolution. The dependence of the difference in integration method on the computed hydrodynamic forces is shown by Figs. 23 and 24. The volume-integration method shows a better performance for the added-mass prediction, while the surface-integration method gives a better prediction of the damping coefcient. For the secondorder hydrodynamic forces, the two methods give the same results except for the steady component F0(2).

4 Conclusions
Ka

Fig. 23. First-order hydrodynamic force coefcients on the wedge obtained by different force calculation methods

Fig. 24. Second-order hydrodynamic force coefcients on a wedge obtained by different force calculation methods

integration method, we mean integration over the whole computation domain. Figure 22 shows the time variation of the hydrodynamic forces acting on the wedge obtained by the two force calculation methods. The wave-number is (p) Ka = 0.4. Only a slight difference was found for Fz ,

We have introduced a new numerical approach for extremely nonlinear free-surface problems in the eld of marine engineering. The wavebody interaction is treated as a multiphase problem in which the free surface as well as the body boundary is immersed in the computation domain. One set of governing equations is used to represent the motion of all materials, and the governing equations are solved numerically on a nonuniform, staggered Cartesian grid by a CIP-based nite-difference method. Several 2-D numerical simulations were carried out. The numerical results provide an assessment of the proposed model and a partial answer to the question: Is the CIP method capable of quantitative prediction of extreme wavebody interaction problems? In particular, what we most want to know, but have not investigated before, is how accurate the CIP calculation is for the hydrodynamic force acting on the body. Therefore, as well as proposing some new developments for the implementation of the CIP method, the originality of this paper lies rather in the quantitative assessment of the precision of the hydrodynamic force calculation. The main ndings by the numerical results presented are given below. For the solitary wave propagation computation, grid convergence can be achieved, but damping of the wave height by the numerical diffusion is still found. The dependence of the damping on the Reynolds number compares well with the theoretical results. The result of the dam-breaking computation shows that the computed peak value of the impact pressure and the measurements are in good agreement, and are not very sensitive to grid density if a sufciently large grid number is used. As numerical predictions of the impact pressure are still very difcult in this type of computation, the present CIP method is shown to be quite satisfac-

156

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

tory. The last example we presented, the forced heaving oscillation test with a wedge-type oat in a 2-D numerical wave tank, is another challenging subject for a Cartesian grid approach. This is a three-phase computation, and a new method was developed to calculate the motion of a solid body. The accuracy of the two forceintegration methods, the surface-integration method and the volume-integration method, were also investigated. We found that the relatively simple method, the volume-integration method, is nearly as accurate as the surface-integration method. The precision of hydrodynamic force predictions using the volume-integration method is comparable to that using the nonlinear BEM when compared with the experiments. The results imply that for a wavebody interaction problem, if the pressure-related hydrodynamic force acting on the body is dominant, a quantitatively correct prediction can be achieved by the current CIP-based method. Appendix A Equations for the 2-D CIP method Rewriting Eqs. 11 and 12 in the two-dimensional form, we have
c c c +u +v =0 t x y c x c c + u x + v x =0 t x y

polynomial can be constructed to approximate the spatial distribution of the value c in the upwind cell, as follows:
X(x , h) = C30x 3 + C 21x 2h + C12xh 2 + C03h 3 + C 20x 2 + C11xh + C02h 2 + C10x + C01h + C00

(A4)

There are ten unknown coefcients in Eq. A4, Cmn, n n which will be determined by the values of cn, cx , and cy n at grid points (i, j), (iw, j), and (i, jw), and the value of c at the grid point (iw, jw). We obtain
n C30 = x1 c x (iw, j ) + c xn (i, j ) n n

{[

] - 2[ c (iw, j ) - c (i, j )]}


+ 3 c n (iw, j ) - c n (i, j )

x13

(A5)

n C 20 = -x1 c x (iw, j ) + 2 c xn (i, j )

{ [
[

]
x12
(A6)

]}

n C03 = h1 c y (i, jw) + c yn (i, j ) n n

{[

] - 2[ c (i, jw) - c (i, j )]}


+ 3 c n (i, jw) - c n (i, j )

h13

(A7)

(A1) (A2)

n C02 = -h1 c y (i, jw) + 2 c yn (i, j )

{ [
[

]
h12
(A8)

]}

n C 21 = c x (iw, jw) - c xn (iw, j ) - c xn (i, jw) n + cx (i, j )

(A3) +u +v =0 t x y The terms on the right-hand side of Eqs. A1A3 are not shown because they can be included in the nonadvection phase calculation. Considering a grid point (i, j), we can nd an upwind cell with four grid points, (i, j), (iw, j), (i, jw), and (iw, jw), as shown in Fig. 25. Here, iw = i-sign(u), jw = j-sign(v). Then a cubic

c y

c y

c y

] (2x h )
1 1 1 1

(A9)

n C12 = c y (iw, jw) - c yn (iw, j ) - c yn (i, jw) n + cy (i, j )

] (2x h )
1 1 21 1

(A10)

C11 = c n (iw, jw) - c n (iw, j ) - c n (i, jw) + c n (i, j )


n C10 = c x (i, j) n C01 = c y (i, j)

] (x h ) - C

x - C12 h1

(A11) (A12) (A13) (A14)

C00 = c n (i, j)

where z1 = -sign(u)Dx and h1 = -sign(n)Dy. Then the values after the advection calculation, c*, c* x, and c* y, can be obtained directly as follows:

c * (i, j) = C(x , h )
* (i, j) = cx
Fig. 25. Interpolation concept for the CIP method

(A15) (A16)
x =x h =m

C(x , h) x

C. Hu and M. Kashiwagi: CIP method for violent free-surface ows

157

* (i, j) = cy

C(x , h) h
x =x h =m

(A17)

where z = -sign(u)Dt and h = -sign(n)Dt.

Appendix B Calculation method for the density function for a 2-D rigid body For a rigid body, since the geometry does not change with time, a Lagrangian method was developed to calculate f3 to obtain the solid-body boundary positions very accurately without any numerical diffusion. For the two-dimensional case, the main procedure is as described below. 1. The two-dimensional body boundary is approximated by a series of straight line segments (pk, pk+1), k = 1 ~ N, as shown in Fig. 26a. 2. The coordinates for the end points (xpk, zpk) are calculated by the equations
0 0 0 x pk = xc + x 0 pk - xc cos a - z pk - zc sin a

where (xc, zc) is the mass center of the oating body, a is the roll angle, the superscript 0 denotes the initial value (xc, zc), and a is calculated in a Lagrangian way because the hydrodynamic forces on the body can be obtained by the method described in Sect. 2.6. 3. All of the intersection points (nodes) of line segments and grid lines are then calculated. For each computation cell, if there are more than two nodes, the cell is considered as a cell which includes the solid-body boundary, and the area of the solid body in this cell is computed to determine f3, as shown in Fig. 26b.

References
1. Muscari R, Di Mascio A (2003) A local model for the simulation of two-dimensional spilling breaking waves. J Mar Sci Technol 8:6167 2. Olivieri A, Pistani F, Di Mascio A (2003) Breaking wave at the bow of a fast displacement ship model. J Mar Sci Technol 8:6875 3. Park J-C, Kim M-H, Miyata H (2001) Three-dimensional numerical wave tank simulations on fully nonlinear wave-current-body interactions. J Mar Sci Technol 6:7082 4. Andrillon Y, Alessandrini B (2004) A 2D+T VOF fully coupled formulation for the calculation of breaking free-surface ow. J Mar Sci Technol 8:159168 5. Yabe T, Wang PY (1991) Unied numerical procedure for compressible and incompressible uid. J Phys Soc Jpn 60:2105 2108 6. Yabe T, Xiao F, Utsumi T (2001) The constrained interpolation prole method for multiphase analysis. J Comput Phys 169:556 593 7. Harlow FH, Welch JE (1965) Numerical calculation of timedependent viscous incompressible ow of uid with a free surface. Phys Fluids 8:21822189 8. Hirt CW, Nichols BD (1981) Volume of uid (VOF) method for the dynamics of free boundaries. J Comput Phys 39:201 225 9. Sussman M, Smereka P, Osher SJ (1994) A level set method for computing solutions to incompressible two-phase ow. J Comput Phys 114:146159 10. Xiao F (1999) A computational model for suspended large rigid bodies in 3D unsteady viscous ows. J Comput Phys 155:348 379 11. Brackbill JU, Kothe DB, Zemach C (1992) A continuum method for modeling surface tension. J Comput Phys 100:335354 12. Tang CJ, Patel VC, Landweber L (1990) Viscous effects on propagation and reection of solitary waves in shallow channels. J Comput Phys 88:86113 13. Mei CC (1983) The applied dynamics of ocean surface waves. Wiley, New York 14. Power P, Chwang AT (1984) On reection of a planar solitary wave at a vertical wall. Wave Motion 6:183195 15. Maxworthy T (1976) Experiments on collisions between solitary waves. J Fluid Mech 76:177185 16. Chan RKC, Street RL (1970) A computer study of nite amplitude water waves. J Comput Phy 6:6894 17. Yamashita S (1977) Calculations of the hydrodynamic forces acting upon cylinders oscillating vertically with large amplitude (in Japanese). J Soc Nav Archit Jpn 141:6170 18. Kashiwagi M (1996) Full nonlinear simulations of hydrodynamic forces on a heaving two-dimensional body. J Soc Nav Archit Jpn 180:373381

zpk = zc

( + (x

0 pk

) ( - x ) sin a + (z
0 c

0 pk

) - z ) cos a
0 c

(B1) (B2)

Fig. 26. Method for calculating the density function of the solid phase for the 2-D case. a 2-D body represented by the enclosing line segments. b Density function for a boundary cell

Вам также может понравиться