To verify the new SBT algorithm proposed in this study, a still water case and three typical sloshing test cases in liquid tanks are simulated using SPH. The tank’s geometries in the three violent sloshing cases are different, including a rectangular tank with uniform boundary particle distribution, a complex geometry tank with coarse boundary particle distribution and a tank with unidirectional deformable bottom. The numerical results are compared with experimental ones and also numerical results by other methods.
4.1. Still Water Case
In this section, a 2D still water case in a rectangular tank is simulated using SPH with different boundary methods. The breadth and height of the tank are 1.0 m and 1.0 m, and the initial water depth is . A pressure probe point is placed 0.2 m below the initial free surface on the right wall. In the SPH simulations, the density of fluid particles is , and the numerical sound speed is . The artificial viscosity parameter is used. The particle spacing is set to , and the SPH time step is . The MLS correction is applied to smooth the density field and approximate the physical quantities of boundary particles at every 5 time steps. The particle shifting algorithm is not applied to these SPH simulations of this section.
Figure 2 shows the pressure field and particle distribution at 15 s, which is a period of time after the initial disturbance and the pressure histories at the pressure sensor are very stable, as shown in
Figure 3. All the pressure fields in
Figure 2 are very smooth, and all the pressure values at the pressure sensor in
Figure 3 are very close to the analytical results. Besides, the vertical distribution of pressure can also show the accuracy of the hydrostatic pressure distribution. Values of depth (
) against pressure
for each fluid particle are shown in
Figure 4. Although some slight discrepancies between the analytical values and the pressure values of the fluid particles near the free surface appear in
Figure 4, all the SPH results obtained with different boundary methods are very accurate, generally.
4.2. Sloshing in Rectangular Tank with Different Filling Ratios
In order to validate the accuracy and stability of the new SPH scheme, the experimental sloshing cases used by Pakozdi and Graczyk [
33] are simulated using Finite Volume Method (FVM) and the SPH scheme with different SBT algorithms. More details about the experiments can be found in Cao et al. [
34] and Pakozdi & Graczyk [
33]. The dimensions of the 2D rectangular water tank are: breadth
, height
, and the water depth is
, as shown in
Figure 5. A pressure probe point is placed 0.09 m above the bottom on the right wall. In addition, to visualize numerically the sloshing wave profile, a wave probe is placed 0.01 m from the right wall. The tank is subject to a sinusoidal translational excitation in the x-axis direction. In the SPH simulation, the excitation motion can be imposed in a moving frame of reference through a forcing term in continuum equation,
where
is the lateral acceleration,
is the initial time smoothing function,
is the amplitude of displacement, and
is the excitation frequency. The natural frequency of the sloshing system of a rectangular tank partially filled with water is
. Hence, the first linear natural frequency is
. Two sloshing cases in the tank with different water depths are investigated by Pakozdi and Graczyk in the experiment. The frequencies and the amplitudes of the excitation motion of the tank corresponding to the sloshing cases are listed in
Table 1.
In the SPH simulation, the density of fluid particles is
, and the numerical sound speed is
. The artificial viscosity parameter
is used in this case. The particle spacing is set to
. Hence, the numbers of fluid particles these cases are 3332 and 1666 respectively, as listed in
Table 2. There are some discrepancies in the water depths between the experiment and numerical simulations, but they are very small and negligible. The SPH time step is
. The MLS correction is applied to smooth the density field and approximate the physical quantities of boundary particles at every 5 time steps. The particle shifting algorithm is not applied to these SPH simulations of this section.
For the FVM simulations, which are based on the Euler two-phase separated flow model and computed using the commercial code STAR-CCM+, the computation domain is discretized into uniform grids with
, too. Due to the restrictions of the grid size, the numerical parameters are the same as those in the SPH simulation, as shown in
Table 2. For all FVM simulations in this section, the time step is set to 0.001 s, which satisfies the Courant condition. The interface between water and air is captured with a volume of fluid (VOF) technique [
14]. STAR-CCM+ is a reliable numerical tool, which has been widely used by many researchers for sloshing simulations [
35,
36,
37]. The results of the FVM simulation are chosen as the benchmark value.
Figure 6 and
Figure 7 show the comparison of the pressure and the wave elevation on the measuring probes from the numerical results and the experimental results for the two sloshing cases, which the experimental results only present in 38 s [
34]. As seen in
Figure 6 and
Figure 7, all the numerical results of 0–40 s are in good agreement with the experiment in both the wave elevation and the pressure on the tank, and the SPH results with different boundary methods are very close to each other. However, for the numerical results of 40–80 s, the SPH results with the previous boundary methods show some obvious discrepancies with the FVM results and the SPH results with the new boundary method. It should be noted that the obvious numerical errors of the SPH results in
Figure 6a at 38–42 s are caused by the splashing fluid particles, because the wave probe is easily disturbed by these splashing fluid particles.
As shown in the pressure histories of
Figure 7, the numerical dissipation in the SPH simulations with the previous boundary methods is becoming severer as the simulation goes. In the previous coupled dynamic SBT algorithm of Shao et al. and Chen et al., one inner layer of the repulsive particles is arranged right on the solid boundary, and the repulsive force produced by the repulsive particles can improve the ability of preventing fluid particles penetration compared with the traditional fixed ghost particles method. On the other hand, however, the repulsive force can also produce severe numerical dissipation for long duration simulation. In the new coupled dynamic SBT algorithm of this study, the outer layers of boundary particles located outside the solid boundary area are the repulsive particles, and the repulsive particles only exert repulsive forces on the fluid particles which have crossed the inner layer of ghost particles. Thus, in the SPH simulations with the new boundary method, the pressure history is more stable throughout the entire simulation, and the downtrend of the pattern of the SPH results with the new SBT algorithm is smaller, as shown in
Figure 7. The numerical dissipation of the SPH simulations with the new boundary method is very small, and the numerical error is approximately within 10% of the FVM results, as shown in
Figure 7. However, the numerical error of the SPH results obtained with the previous boundary method can be more than 25% of FVM results, approximately.
For further investigating the numerical dissipation, three different values of the numerical sound speed
, namely
,
and
, are used in the SPH simulation with different SBT algorithms for the Sloshing Case 1.
Figure 8 shows the comparison of the pressure histories obtained from the FVM simulation and the SPH simulations with different SBT algorithms. As shown in
Figure 8, although there are some measurements proposed by Chen et al. [
16] to avoid the large initial errors, there is still a drawback of the previous SBT algorithms, especially for these SPH simulations with a large value of
. However, due to the innovative arrangement of the repulsive particles, the SPH simulations with the new SBT algorithm not only have small numerical dissipation, but also have small initial errors. As shown in
Figure 8c, the SPH results with
are very close to the FVM results, and the numerical dissipation is extremely small.
The numerical accuracy and stability of the new coupled dynamic SBT algorithm have been validated by these sloshing simulations in the rectangular tank. However, due to the simple geometry of the water tank and the uniform distribution of boundary particles, very few fluid particles can penetrate the solid wall and activate the repulsive forces of the repulsive particles. Thus, more violent sloshing cases with complex geometry need to be investigated.
4.3. Sloshing in Tanks with Complex Geometry
Violent sloshing simulations in tank with complex geometry are conducted with FVM and SPH with different boundary methods. The geometry of the sloshing tank is shown in
Figure 9a. The tank breadth, height and initial water depth are 1.0 m, 1.65 m and 0.45 m, respectively. The radius of the semicircles is 0.15 m. A pressure sensor is placed 0.1 m below the initial free surface on the right wall, and a wave gauge is placed 0.01 m from the right wall, as shown in
Figure 9a. For the lateral acceleration
, the amplitude of displacement is 0.01 m and the excitation frequency is 4.0 rad/s. In the SPH simulations, the numerical sound speed is
, and the other SPH parameters are as the same with those in previous sloshing cases in
Section 4.2. The particle spacing is set to
, resulting in about 2009 fluid particles generated in the computation domain. In the FVM simulation, the grid spacing is set the same as the particle spacing used in SPH simulation, and other parameters are the same as these in the FVM simulation of
Section 4.2. It should be noted that although FVM is well validated in
Section 4.2, FVM still has some problems for violent sloshing simulations with dynamically evolving free surface, and the free surface is difficult to be accurately captured.
The sketch of boundary particles distribution around the tank’s bottom is shown in
Figure 9b. It is obvious that the boundary particles distribution is a little coarse due to the restrictions of the tank’s geometry. If the boundary methods used do not have a strong ability of preventing fluid particles penetration, fluid particles can easily cross the solid wall under violent sloshing excitation.
In the SPH simulations with the previous SBT algorithms used by Shao et al. and Chen et al., at least more than 450 (22%) fluid particles penetrate the solid wall from the gap between boundary particles during the 60 s simulation. As shown in
Figure 10, the SPH results of the simulation with the previous SBT algorithms are less accurate and reliable. However, in the SPH simulation with the new SBT algorithm, no fluid particle penetration occurs throughout the entire simulation, and the SPH results obtained with the new boundary method agree well with the FVM results. The reason is that the magnitude of the repulsive forces in the previous SBT algorithm only depends on the distance between the fluid particle and the repulsive particle. In the new SBT algorithm, the magnitude of the repulsive forces, produced by the repulsive particle, not only depends on the distance between the fluid particle and the repulsive particle, but also the distance between the fluid particle and the plane to which the repulsive particle belongs. Obviously, the repulsive force in the new SBT algorithm, partially based on the plane, has better ability of preventing particle penetration than that in the previous SBT algorithm.
Besides, it should be noted here that, the criterion for determining fluid particle penetration in the SPH simulation with the new SBT algorithm is different from that in the SPH simulations with the previous SBT algorithms. For the new SPH scheme in this study, fluid particles that completely leave the outermost layer of the computation domain will be marked as penetration particles and removed from the computation domain. In the simulations with the previous SBT algorithms, fluid particles that penetrate the inner layer of boundary particles at the solid wall will be removed from the computation domain. If the same criterion as used in the new SPH scheme, the SPH simulations with the previous SBT algorithms will break, and the 60 s simulation cannot be accomplished because of too many fluid particles penetration occurrences.
Figure 11,
Figure 12 and
Figure 13 show the comparisons of the flow pattern and pressure distribution from the FVM simulation and SPH simulations at some typical instants. The flow field in SPH simulations is in general consistent with that in FVM simulation. As shown in
Figure 11b,
Figure 12b and
Figure 13b of the FVM simulation, the interface between water and air is not clear under so violent sloshing excitation. But the flow fields in SPH simulations are still stable and smooth, and the SPH results obtained using the new SBT algorithm are close to the FVM results for the pressure histories and wave elevation at the probes, as shown in
Figure 10. The advantage and robustness of the new SBT algorithm can be validated by the violent sloshing simulation with coarse boundary particles distribution.
4.4. Sloshing in Tank with Unidirectional Deformable Boundaries
Modelling deformable boundaries is quite a problem for these fixed ghost particle boundary methods. To investigate the performance of the new SBT algorithm in simulating deformable boundaries, a case of sloshing in water tank with unidirectional deformable bottom is simulated by both SPH and FVM. The dimensions of the tank are shown in
Figure 14a. The tank is 1.0 m wide and 2.0 m high, which is large enough to make sure that no water particle can reach the ends. The initial water depth is 0.4 m, and the initial position of the deformable bottom is 0.5 m above the ground. A pressure probe point is placed 0.2 m below the initial free surface on the right wall, and a wave probe is placed 0.01 m from the right wall, as shown in
Figure 14a. The tank’s motion is driven by a horizontal excitation and a vertical excitation of the deformable bottom. In the horizontal direction, the excitation pattern of the tank is as the same as the tank’s motion of the sloshing case in
Section 4.2. The amplitude of displacement
is set to 0.03 m and the excitation frequency
is 5.0 rad/s. In the vertical direction, the deformation of the tank’s bottom is controlled by the deformation velocity:
where
is the frequency,
the amplitude and
the wave number. In this study, the deformable motion parameters of the tank’s bottom are
,
and
. It should be noted that the time step of boundary deformation
in SPH simulation is set 100 times of the time step
of the Predictor-Corrector scheme. During the time step
, the deformation velocity of the bottom is approximately regarded as constant for the calculation of the physical quantities of fluid particles, as which can greatly improve the numerical stability of the SPH simulation.
Figure 14b shows the initial distribution of particles in the tank. The bottom boundary particles move with the deformable bottom, as shown in
Figure 14c.
In the SPH simulations, to examine the convergence of the SPH scheme, three different particle spacings are explored in these SPH simulations,
,
and
, resulting in about 980, 2528 and 3960 fluid particles in the computation domain. The time step
is set to
for all SPH simulations. The deformation time step
is set to 0.005 s, which is an acceptable approximation of deformation velocity. The density of water is
, the numerical sound speed is
, the artificial viscosity coefficient is
. The MLS correction of density field is applied at every time step for both fluid particles and boundary particles, which is more frequent than these sloshing simulations in
Section 4.2 and
Section 4.3. Since the deformable boundary can intensify particles clumping and numerical dissipation, the particle shifting correction is applied to improve the fluid particles distribution. The particle shifting coefficient
is a little different for different particle spacings. In the SPH simulation with
, the particle shifting coefficient
is set to 0.8. While in the other SPH simulations, the particle shifting coefficient
is 0.3.
As for the FVM simulations, the sloshing simulations are conducted with the commercial code STAR-CCM+ using the Euler two-phase separated flow model. The STAR-CCM+ contains a morphing motion model that can be used to define the deformable motion of the tank’s bottom [
37]. Three different grid spacings are explored, which are the same as the particle spacings used in the SPH simulation. For the purpose of convenience, the time step of all the three FVM simulations is set to 0.001 s, which satisfies the Courant condition. As shown in
Figure 15, the results of the three FVM simulations in 20 s converge reasonably well, although some discrepancies which are caused by jet flow near the wall boundary appear in
Figure 15a. Thus, the results of the FVM simulation with grid spacing
are chosen as the benchmark value.
In order to examine the convergence of the new SPH scheme in this study, three SPH simulations with different particle spacings are conducted.
Figure 16 shows the comparisons of SPH results and the FVM results for the wave elevation and pressure on the measuring probes in 20 s. The SPH results obtained with
agree most with the FVM results, as shown in
Figure 16.
Figure 17 and
Figure 18 show the comparisons of the typical flow pattern and pressure distribution for the FVM simulation and SPH simulation at two typical instants. The SPH results are highly consistent with the results of FVM. Some slight discrepancies are present in
Figure 17, because the air cavity only can be captured in the FVM simulation based on the Euler two-phase flow model. The pressure fields of the SPH simulation with the new SBT algorithm are smooth, and almost no particle penetration occurs. It can be concluded that, with the new SBT algorithm, the SPH scheme in this study has the ability of simulating sloshing cases with unidirectional deformable boundary.
To compare the new SBT algorithm with previous ones, the SPH simulations with different coupled dynamic SBT algorithms are conducted for 140 s. Although fluid particles penetration also appears in the SPH simulation using the new SBT algorithm, the number of fluid particles penetrating the wall boundary is much smaller than those using the previous SBT algorithms, as shown in
Table 3. A great improvement in the ability of preventing fluid particles penetration is observed. It should be noted that, at the initial state of the SPH simulation with
, there are 3960 fluid particles in the fluid domain.
Figure 19 shows the results of the liquid sloshing SPH simulations. Since the grid distortion in FVM simulation after 20 s is gradually becoming severe as the simulation goes, the FVM results are less accurate and not present in
Figure 19. As the SPH simulation goes, more and more fluid particles penetrate the wall boundary and are removed from the computation domain in the SPH simulation using the previous SBT algorithms. The discrepancy between the SPH results with the new SBT algorithm and the previous SBT algorithms is becoming more and more obvious along with the increase of the number of penetration fluid particles, as shown in
Figure 19.
In this section, the sloshing simulations in the tank with both horizontal translational excitation and deforming bottom excitation are conducted using the new SPH scheme. The numerical stability and accuracy of the new SBT algorithm are demonstrated, even for complicated and violent sloshing cases. Success in obtaining stable and accurate numerical results is attributed to the repulsive force of repulsive particles and the relatively uniform fluid particle distribution near the boundaries.