Recent Researches in Mechanics

Investigations of Boundary Treatments in Incompressible Smoothed Particle Hydrodynamics for Fluid-Structural Interactions
Fanfan Sun, Mingyi Tan, and Jing T Xing

Abstract Two boundary treatment methods were developed for

incompressible flow simulations and fluid-structural interaction problems using Smoothed Particle Hydrodynamics (SPH): 1) To apply repulsive force on the boundary particles while keeping the same particle spacing for inner fluid particles and wall boundary particles; 2) To use denser wall particles without any additional force. The dam-breaking problem and another testing example are used to demonstrate the performance of this method. Results obtained from the present approach show reasonable agreement with experimental data. The fluid pressure values obtained with SPH method is investigated. Based on the result of the study, it can be concluded that the present approach is reliable to simulate incompressible fluid and the pressure value obtained can be used to solve fluid-structural interaction problems.

each fluid particle as a constant in the simulation using Lagrange multipliers [9]. Another way to enforce incompressible fluid is to set the density variation and velocity divergence to be zero. This method is used for multi-phase fluid simulations to enforce the incompressibility [10]. All those treatment methods, either setting density variation to be zero or force the velocity divergence to be zero, require additional consideration on the fluid density variation. In fact the density of the fluid can be simply set to be a constant for the incompressibility, and the zero velocity divergence can be satisfied automatically [11]. This method provides a straightforward approach to the incompressible fluid problem and it is adopted in this paper. It is well known that SPH does not have zeroth order consistency in boundary area. On the boundaries, the failure of SPH modelling is characterized by wall penetration of fluid particles. Generally, there are three ways to prevent this from happening: 1) mirror particles [10], 2) repulsive forces [3] or 3) dummy particles [6], [12]. Usually, repulsive forces are used in WCSPH whereas mirror particles need special consideration on corners or curved surfaces. Hence, dummy particles or ghost particles are preferred in ISPH method [13]. This paper focuses on investigation of boundary treatment methods in order to improve the efficiency of SPH model for incompressible flow simulations. Ghost particles are useful to keep the symmetry configuration of the particles near the wall. Therefore, the kernel domain of the particles can remain complete and the physical properties such as density can be calculated correctly. However, when dealing with problems with complex solid boundaries the ghost boundary treatment becomes difficult. Taking compartment flooding as an example, water can fill both inside and outside of the structure and at least two layers of ghost particles need to be placed on the inside wall and outside wall respectively. These ghost particles sometimes overlap the true fluid particles, which causes inaccurate neighbouring particles counting and results in a wrong predictions. It is also difficult to treat the angled boundaries by using ghost particles. Special consideration is required to calculate the exact position of the ghost particle for the angled points since the ghost particle position is important to ensure

KeywordsBoundary condition treatments, incompressible SPH, pressure prediction. I. INTRODUCTION The Smoothed Particle Hydrodynamics (SPH) method is a fully Lagrangian mesh-free method used widely in large deformation problems such as fluid motions where the continuum hydrodynamic equations are solved with a set of interacting fluid particles [1], [2]. The original equations that are discretised are those for a compressible viscous fluid. When SPH is applied to simulate incompressible flows, there are generally two ways to impose incompressibility: one is to run the simulations in the quasi-incompressible limit by assuming a small Mach number to ensure density fluctuations within 1% [3]-[5], which is known as Weakly Compressible Smoothed Particle Hydrodynamics (WCSPH); the other one is called truly Incompressible SPH (ISPH) in which incompressibility is enforced by solving a Poisson equation at every time step. The velocity divergence is set to zero as a condition to ensure the incompressibility in this method [6][8]. It is noted that incompressible condition also means that the volume of each fluid particle should not change. Hence, the incompressibility can be enforced by setting the volume of
Manuscript received May 11, 2011. F. Sun is with the Fluid Structural Interaction research group of University of Southampton, SO171BJ UK (e-mail: [email protected]). M. Tan is with the Fluid Structural Interaction research group of University of Southampton, SO171BJ UK (e-mail: [email protected]). J. T. Xing is with the Fluid Structural Interaction research group of University of Southampton, SO171BJ UK, (e-mail: [email protected]).

ISBN: 978-1-61804-020-6

92

Recent Researches in Mechanics

that there is no fluid particle penetration. Practically, as long as the density can be kept as constant, preventing particles from penetrating the walls is the major concern of these boundaries. Therefore, repulsive force can be applied on the wall particles instead of using several lines of dummy particles which not only increases computation time but also complicates the model especially in fluid structural interaction problems. Another boundary treatment using denser wall particles is also investigated. With repulsive force, all the particles can be maintained in a uniform arrangement but the additional force may affect the pressure values obtained. This problem can be overcome by using denser wall particles, say half spacing wall particles, but it requires slightly more computation time in this case. These two boundary treatments can be chosen according to different situations. Both the present boundary treatments allow efficient simulations with complex solid boundaries and even provide a new coupling approach for fluid structural interactions in the future. II. NUMERICAL MODEL 2.1 SPH formulation The SPH formulation is based on the theory of integral interplant that uses kernel function to approximate delta function. A physical property is obtained by the interpolation between a set of points inside a certain area. These points known as particles carry all the properties the fluid has, such as mass and velocity. The basic idea of this method is to approximate a function A(r) as [14]

Where t is the time, is density, g is the gravitational acceleration, P is pressure, v is the velocity, is viscous stress tensor and D/Dt refers to the material derivative. The momentum equations include three driving force terms, i.e., body force, forces due to divergence of stress tensor and the pressure gradient. 2.3 Incompressible SPH and numerical formulation For incompressible flow the mass density is a constant. According to (4) the velocity divergence will be zero [11] .

v = 0

(6)

Splitting the momentum equation into two parts, one is with the effect of body force and viscosity introducing an intermediate velocity,

v n +1 / 2 v n 1 = g + n t

(7)

Another is from pressure influence and the new velocity can be updated based on the intermediate one obtained from previous step

v n +1 v n +1 / 2 1 = P n +1 t

(8)

Taking the divergence of (8) and substitute (6) so the pressure Poisson equation can be derived

A(ra ) = mb
b

Ab

W ( ra rb , h)

(1)

P n +1 =

v n +1 / 2 t

(9)

A model of SPH formulated gradient term in the N-S equation is employed to preserve linear and angular momentum [3] P P 1 (2) ( P) a = mb a2 + b2 aWab b a b

The Poisson equation is formulated with SPH method

n n mb Pab+1rab W'n = mb u n +1/ 2 W'n ab ab ab 2 2 2t b rab +

(10)

where

aWab =

x a x b Wab rab rab

Hence, the pressure can be updated by (10) implicitly. The viscous force is computed in SPH form as (3)

2.2 Governing equations The governing equations for incompressible continuum including the conservation of mass and momentum are presented in the following equations, respectively.

1 = mb a2 + b2 aWab b a a b

(11)

1 D +v = 0 Dt

(4) (5)

Where the stress tensor is related to the strain tensor. The suffix a and b represent different particles. u u (12) ij = ji = eff i + j x j xi The full derivative between two particles is first obtained using finite difference before decomposing it into x and y directions. Thus

Dv 1 1 = g + P Dt

ISBN: 978-1-61804-020-6

93

Recent Researches in Mechanics

u i x j

= ui a rab

rab x j

u u = r ab
i a

i b

x x r ab
j a

j b

(13)

r 1 r 2 r f ( r ) = D 0 0 2 r r r
p p

(19)

It is zero when r > r0 so that the force is purely repulsive. For Newtonian fluids such as water, the viscosity coefficient

eff

(effective viscosity) has a constant value . Hence, the

Mostly, p1 = 4 and p 2 = 2 , D=5gH according to [3]. The length scale r0 is taken to be the initial spacing between the particles. In incompressible SPH method, ghost particles which mirror the physical properties of inner fluid particles are the usual treatment of wall boundary conditions. These ghost particles make smoothing domain complete for the near wall fluid particles so that the consistence of SPH simulation near wall boundaries is ensured. The repulsive force boundary treatment as shown in Fig. 1 has not been used in incompressible SPH method. Normally ghost particles are considered to be necessary to avoid unphysically large density variation for the near boundary particles. However, when the density of all the particles is set to be a constant, ghost particles will not be necessary any more. Wall particles are involved in the Poisson equation, using denser particles on the wall boundary can produce pressure to keep the inner particles away from the boundary. A halfed spacing is set on the wall particles compared with the inner fluid particles as shown in Fig. 2. Alternatively, the simple repulsive force treatment can be applied to save computational time.

SPH formulation of viscosity term can be written as [6]:

4m r W (u u ) 2 b u = b ab a ab a 2 2 b 2 rab + a
The quantities are updated following the steps

(14)

4m r W (u u b ) u n +1 / 2 = u n +1 + t g + b ab a ab a a a 2 2 b rab + 2
(15)

Pa
n +1

2t
b

mb u n +1 / 2 W'ab ab
b n n Pab+1rab 2 rab + 2

(16)

mb

W'ab

P n +1 P n +1 u n +1 = u n +1 / 2 t mb a 2 + b 2 W 'ab a a b

(17)

III. BOUNDARY CONDITIONS 3.1 Free surface Free surface particles are tracked down to set their pressure to zero to simplify the dynamic surface boundary conditions [15]. The following quantity is calculated to identify the free surface particles Fig. 1: Boundary treatment: using halfed spacing on wall particles

r =

mb

rab w'ab eab

(18) Fig. 2: Boundary treatment: using repulsive force

This value equals to 2 in 2-D applications or 3 in 3-D cases when the smoothing domain is not truncated but it is far below these values for surface particles, a criterion used in this paper is 1.6 in 2-D cases. 3.2 Wall boundary The solid walls are simulated by particles which prevent the inner particles from penetrating the wall. Conventionally, the wall boundary conditions are modelled by fixed particles exerting a repulsive force on inner fluid particles in weakly compressible SPH method.

IV. COURANT NUMBER CONDITION Since this incompressible SPH method calculates pressure implicitly and other properties explicitly, the size of time step must be controlled in order to have stable and accurate results. The following Courant condition must be satisfied [6] h (20) t 0.1 v max where h is the initial particle spacing and

v max is the

maximum particle velocity in the computation. The factor 0.1 ensures that the particle moves only a fraction (in this case 0.1) of the particle spacing per time step. Another constraint is

ISBN: 978-1-61804-020-6

94

Recent Researches in Mechanics

based on the viscous terms [10] h2 t 0.125 eff /

(21)

where

eff

is the effective viscosity. The allowable time-step Fig. 5: testing case 1 using repulsive force on wall particles when t=0.4s, 0.6s, 1.8s V. TESTING EXAMPLES

should satisfy both of the above criteria.

Without ghost particles, model with complex wall boundaries can be simulated efficiently. An example is used to test the two boundary treatments in simulating water flooding into compartments as shown below in Fig. 3.

Only small differences can be observed from the results obtained based on these two different boundary treatments.

VI. PRSSURE INVERSITGATION The investigation of these two boundary treatments are carried out with a 2-D dam break simulation. The model is set up as shown in Fig. 6. The spacing of fluid particles is set to be 0.01m, the overall height of the water column (H) is 0.6m and its width (a) is 1.2m. The size of the solid container is 3.22m long. The water column is kept in hydrostatic state in the beginning. The flow starts when the right side gate is suddenly opened so the water column collapsed. The pressure values (P) at point 0.16m on the right wall are traced. The results are obtained using ghost particles, half spacing wall boundary particles and repulsive force boundary treatments and the results are compared with experimental data provided by [16]

Fig. 3: water flooding compartment The water column is set to be 2m high and 1m wide. It is kept in a static state in the beginning, but suddenly a hole is unblocked on the right wall and the water starts to flow out. Simulation of the water flooding was recorded at time 0.4s, 0.6s and 1.8s as shown below in Fig. 4.

Fig. 6: Dam break Fig. 4: testing case with halfed spacing for wall particle when t=0.4s, 0.6s, 1.8s Water flows violently after the unblocking of the hole. The front of water impacts the right side wall and comes back impacting the second building. Some air bubbles are generated during the process, fluid motion changes quickly. If ghost particles are used to model the wall boundary, several layers of these ghost particles must be placed on each side of the "small structures". This will affect the fluid particles when flow fills both sides of the structure. Some of the ghost particle may overlap with the true fluid particle, which makes the counting of the neighbouring particles inaccurate. Results obtained using repulsive force applied on the boundary particles are also displayed in Fig. 5 to make a comparison with the denser wall particle treatment. Water configuration with time is shown in Fig. 7

Fig. 7: Water configuration at time 0s,0.5s,0.75s,1.5s Results obtained using different boundary treatments in SPH method are compared with experimental data [16] as shown in Fig. 8.

ISBN: 978-1-61804-020-6

95

Recent Researches in Mechanics

Fig. 8: Results obtained from three different boundary treatments compared with experimental data and another numerical method All these three boundary treatments find the first pressure peak around the right time compared with experimental data. The first peak values obtained from numerical method are slightly larger than observed in the experiment. Boundary treatments with repulsive force and ghost particles gave similar results, denser wall particle boundary treatment give slightly higher peak value than the other two treatments. Except the second peak value, the overall curves agree with experimental data well. There is no obvious second peak pressure in the simulations. This is perhaps because of entrained air effects which are not predicted during the simulation. But compared with other numerical method such as Navier-Stokes Solver provided by [17], SPH gives closer values to the experimental ones.

Fig. 10: Investigation of halfed wall particle spacing treatment with different time stepping size From Fig. 10, it can be seen that similar to repulsive force treatment, using time stepping size of 0.0001s provides better results than using 0.0005s. However, the curves do not change as much as the previous boundary treatment case. Further time stepping size decreasing does not change the results noticeably. And a second peak is predicted in this case when using time stepping size of 0.00005s, which indicates that using denser wall particles is a better boundary treatment to obtain accurate results compared with repulsive force treatment. VII. CONCLUSION Simpler boundary treatments can be used instead of conventional ghost particles for incompressible SPH with constant fluid density. Simulations with complex solid boundaries can now be modelled without difficulty. Two testing examples have been used to demonstrate the application of these two boundary treatments. Model set-up can be done more efficiently using one of these two boundary treatments. They not only offer a simpler configuration for the model but also produce better simulations. It is observed from the pressure results analysis that incompressible SPH can provide more accurate pressure value. REFERENCES
[1] R. A. Gingold and J. J. Monaghan, Smoothed Particle Hydrodynamics: Theory and Application to Non-Spherical stars, Monthly Notice of the Royal Astronomical Society, vol. 181, pp. 375-389, 1977. L. B. Lucy, Numerical approach to testing the fission hypothesis, Astronomical Journal, vol. 82, pp.1013-1024, 1977. J. J. Monaghan, Simulating free surface flows with SPH, Journal of Computational Physics, vol. 110(2), pp. 399-406, 1994. J. P. Morris, P. J. Fox, and Y. Zhou, Modelling low Reynolds number incompressible flows using SPH, Journal of Computational Physics. vol. 136, pp. 214-226, 1997 X. Y. Hu and N. A. Adams, A multi-phase SPH method for macroscopic and mesoscopic flows, Journal of Computational Physics, vol. 213(2), pp. 844-861, 2006 S. Shao and E. Y. M. Lo, Incompressible SPH method for simulating Newtonian and non-Newtonian flows with a free surface, Advances in Water Resources, vol. 26(7), pp. 787-800, 2003 J. Pozorski and A. Wawrenczuk, SPH computation of incompressible viscous flows, Journal of Theoretical Applied Mechanics, vol. 40, pp. 917, 2002 S. M. Hosseini, M. T. Manzari, and S. K. Hannani, A fully explicit three-step SPH algorithm for simulation of non-Newtonian fluid flow,

Fig. 9: Investigation of repulsive force boundary treatment with different time stepping size The Fig. 9 shows large variations of the first peak value when using different time stepping sizes. The rest of the curves are almost the same. When time stepping size is 0.0001s, the results are fairly close to the experiment, which means that a time stepping size between 0.0005s and 0.0001s is sufficiently accurate for the simulation. This time stepping size value can be obtained from Courant number condition.

[2] [3] [4]

[5]

[6]

[7]

[8]

ISBN: 978-1-61804-020-6

96

Recent Researches in Mechanics

[9]

[10] [11]

[12]

[13]

[14] [15] [16]

[17]

International Journal for Numerical Methods for Heat & Fluid Flow, vol. 17, pp. 715-735, 2007 M. Ellero, M. Serrano, and P. Espaol, Incompressible smoothed particle hydrodynamics, Journal of Computational Physics, vol. 226(2), pp. 1731-1752, 2007 S. J. Cummins and M. Rudman, An SPH projection method, Journal of Computational Physics, vol. 152(2), pp. 584-607, 1999 E. S. Lee, D. Violeau, and R. Issa, Application of weakly compressible and truly incompressible SPH to 3-d water collapse in waterworks, Journal of Hydraulic Research, vol. 48, pp. 50-60, 2010 A. J. C. Crespo, M. Gomez-Gesterira, and R. A. Dalrymple, Boundary conditions generated by dynamic particles in SPH methods, Computers, Materials & Continua, vol. 5, pp. 173-184, 2007 E. S. Lee, D. Violeau, and R. Issa, Comparisons of weakly compressible and truly incompressible algorithm for the SPH meshfree particle method, Journal of Computational Physics, vol. 227(18), pp. 8417-8436, 2008 G. R. Liu, Meshfree methods: Moving beyond the finite element method, CRC Press, 712 pages, 2002 J. J. Monaghan, On the problem of penetration in particle methods, Journal of Computational Physics, vol. 82, pp. 1-15, 1989 Z. Q. Zhou, J.O. DeKat, and B. Bunchner, A nonlinear 3-d approach to simulate green water dynamics on deck, Proceedings of the 7th international conference on numerical ship hydrodynamics, Nantes, July 1999 K. Abdolmaleki, K. P. Thiagarajan, and M. Morris-Thomas, Simulation of the dam break problem and impact flows using a navierstokes solver, 15th Australasian Fluid Mechanics Conference, 2004

ISBN: 978-1-61804-020-6

97