Вы находитесь на странице: 1из 9
Computers & Fluids 64 (2012) 108–116 Contents lists available at SciVerse ScienceDirect Computers & Fluids

Contents lists available at SciVerse ScienceDirect

Computers & Fluids

journal homepage: www.elsevi er.com/locate/compfluid CFD analysis with fluid–structure interaction of opening

CFD analysis with fluid–structure interaction of opening high-pressure safety valves

A. Beune, J.G.M. Kuerten , M.P.C. van Heumen

Department of Mechanical Engineering, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands

article info

Article history:

Received 11 March 2011 Received in revised form 26 October 2011 Accepted 16 May 2012 Available online 24 May 2012

Keywords:

Computation

Fluid mechanics

Safety

Simulation

abstract

A multi-mesh numerical valve model has been developed to analyze the opening characteristic of high-

pressure safety valves. Newton’s law and the CFD result for the flow force are used to model the move-

ment of the valve. In incompressible transient flow simulations a large force rise and collapse is caused by

a redirection of the bulk flow. This flow-history effect cannot be incorporated in a quasi-steady approach.

For real-gases at a set pressure of 40 bar oscillations have been observed during closing of the valve. They

are caused by the interaction between the flow in the cavity of the valve disk and the flow towards the valve outlet. At a higher set pressure the flow force continually decreases, which indicates that only a suf- ficiently fast inlet pressure rise forces the valve to open. With this tool the operation characteristics of safety valves can be assessed to optimize the valve design.

2012 Elsevier Ltd. All rights reserved.

1. Introduction

Worldwide there are many industrial applications of high- pressure safety devices, such as safety valves or rupture disks, in the range between 250 and 3600 bar set pressure. Examples are the production of synthesis gas or low-density polyethylene. Exist- ing standards, such as EN ISO 4126-1 [6] , do not cover this pressure range and test facilities for an experimental determination of mass flow capacities and opening characteristics are not available. The aim of this paper is to study the influence of valve dynamics on steady flow performance by means of numerical methods. The approach is to include fluid–structure interaction (FSI) in computa- tional fluid dynamics (CFD) calculations of flow through a safety valve. Then the opening characteristic of a high-pressure safety valve and possible valve instabilities can be distinguished, which may not become apparent in simulations with steady flows in non-moving meshes. As a result, the design of a safety valve can be evaluated and improved to avoid flow instabilities during valve operation as much as possible. A spring-loaded safety relief valve consists of a compression spring, which presses the valve spindle with disk on the valve seat in order to seal the pressurized system in case of operating condi- tions below the valve set pressure. Such a safety valve is usually in- stalled on top of the pressurized system, like a vessel, to be protected and directly connected to the system through a short pipe. Fig. 1 shows on the left the set up of such a vessel with safety valve and on the right a construction drawing of the safety valve.

Corresponding author. E-mail address: j.g.m.kuerten@tue.nl (J.G.M. Kuerten).

0045-7930/$ - see front matter 2012 Elsevier Ltd. All rights reserved.

During valve opening a complex flow pattern is formed be- tween the valve seat and spindle with disk. In this region with the smallest flow cross-sectional area the geometry forces the flow to be accelerated up to the smallest cross-sectional area and to be deflected. The numerical method should sufficiently accurately represent challenging flow phenomena, such as real-gas effects, choking below the valve disk and flow separation. The opening characteristics in valves have previously been investigated with CFD in preliminary studies of Domagała [5] and Srikanth and Bashker [14] . These studies showed that it is possible to define a single deformable grid that can cover the whole operat- ing range with sufficient accuracy. FSI has already been used in practical engineering problems of a vacuum relief valve [10] , where the design was improved to avoid the tendency of the valve to flut- ter under expected subcritical gas flow conditions. The problem was solved with two grids that interact with each other by an overset mesh module. In another design optimization study [7] the closing characteristics of a subsurface safety valve operating in productive gas wells have been qualitatively analyzed in combination with field tests. Third, the performance of a stirling cycle machine [8] has been more accurately predicted than in the traditional approach with analytical models on a single moving grid. The problem of reduced grid quality has been tackled in an FSI study of multiple grids for high-speed flows in a pneumatic valve [4] . It has been shown that dynamic flow effects become significant for a predefined piston movement in a critical gas flow. Due to large deformations of the computational domain intermediate meshes are necessary to keep the mesh quality appropriate. In simulations with a single moving grid in each iteration of the solution process the positions of the mesh nodes are adjusted to the new geometry while preserving the mesh quality parameters

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

109

safety valve vessel
safety valve
vessel

Fig. 1. Process diagram (left) and construction drawing (right) of a high-pressure safety valve.

orthogonality, expansion and aspect ratio as much as possible. In this process the mesh topology remains the same, because no nodes can be added or connections can be changed. Simulations with a single deforming grid are common practice for small mesh

deformations. In case of large mesh deformations the application of

a single deforming grid would lead to too large stretching of some

cells in the computational domain. The resulting poor mesh quality would result in large discretization errors or failure of the numer- ical solver. Therefore, a multiple grid approach is necessary in such

a situation. In the case of a high-pressure safety valve the deformation of the grid will not only be large, but also the geometry at the small- est cross-section is complex so that multiple grids are necessary. Furthermore, it is desirable to model the valve opening starting from a closed position, which needs special attention to the grid quality as well. The second challenge is to minimize discretization errors that occur in the transfer of the solution variables between two grids with different topology. In order to reduce geometrical complexity and to limit calculation time the research described here is restricted to axially symmetric flow simulations. The inten- tion of this work is to show that the principle of the method pro- posed here works well and that it enables an increase in the understanding of the physical processes that play a role during the opening and closing of a safety valve. The geometry used in the simulations is based on a commercial proportional spring- loaded high-pressure safety valve manufactured by BASF SE. In the next section the numerical model parameters are pre- sented. In Section 3 results of an opening valve are given for liquid and for gas flow. The results are analyzed with the focus on pres- sure and velocity distributions. In the last section conclusions are presented.

2. Numerical approach

First the mathematical models to model the valve movement and flow dynamics, such as the real-gas equation of state are intro- duced. Then, the discretization method is presented followed by the solution strategy of the valve model with FSI.

2.1. Mathematical models

When a closed safety valve with seat area A 0 is pressurized at a certain set pressure p set for which the valve just starts to open, the spring force F spring and gravity force F gravity are in equilibrium with the flow force F ow that is the set pressure multiplied with the seat area. Fig. 2 shows the forces that act on a moving spindle with a disk of a safety valve. The initial displacement h 0 of the compression spring with stiff- ness k spring equals

h 0 ¼ p set A 0

spring

k

:

ð 1Þ

The acceleration h of the spindle with disk during valve move-

ment is given by Newton’s law as

h ¼ F flow k spring ð h þ h 0 Þ m spindle g

m spindle

;

ð 2Þ

with m spindle the equivalent mass of the moving parts of the valve and g the gravitational acceleration when the valve is operated in vertical orientation. Friction forces due to small possible misalign- ment of the valve spindle are expected to be small. Apart from the moving spindle that deforms the computational domain, the

110

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

A. Beune et al. / Computers & Fluids 64 (2012) 108–116 Fig. 2. Force balance of

Fig. 2. Force balance of moving valve spindle with disk.

thick steel walls of the high-pressure safety valve are considered rigid. To describe the three-dimensional flow phenomena in a safety valve, the conservation laws of mass, momentum and energy are numerically solved in the commercial software package ANSYS CFX [1] . In order to model the effects of turbulence, the shear stress transport (SST) model [9] has been adopted, since of the available two-equation Reynolds-Averaged Navier–Stokes models in CFX, it gives the best results [3,2] . The good performance of this turbu- lence model has been verified with benchmark simulations, such as a 1D shock tube, a two-dimensional supersonic ramp, axially symmetric real gas nozzle flows and two-dimensional and three- dimensional valve models, that represent relevant physical effects occurring in high-pressure safety valves. The conservation equations have to be completed with defini- tions of the fluid properties in the form of an equation of state (EoS). For gas flows at high pressures (or low temperatures) the gas cannot be considered to behave as a perfect gas anymore, so that the stagnation properties deviate from the calorically perfect gas approximation and have to be calculated from a real-gas EoS. For the gas flows described in this paper nitrogen has been ap- plied. In order to calculate real-gas flows at pressures up to 3600 bar with the CFD code the fluid properties have to be known at temperatures between 100 and 6000 K and pressures between 0.01 and 10000 bar. These ranges are so large, since during the iter- ation process large fluctuations are possible. Therefore, look-up ta- bles with the thermodynamic properties specific heat at constant pressure c p , specific volume t , specific heat at constant volume c v , pressure-specific volume derivative at constant temperature

T , speed of sound a , specific enthalpy h , specific entropy s , dy- namic viscosity l and thermal conductivity k have to be generated. In the look-up tables the temperature is linearly divided and the pressure logarithmically into 400 intervals each. The cubic Redlich–Kwong (RK) EoS relates pressure to temper- ature and specific volume of a supercritical gas. This equation was extended by Soave [13] for improved accuracy for larger and polar molecules. For many gases the coefficients of this EoS are well tab- ulated and with the help of mixing rules it can also be applied to gas mixtures, which is beneficial for practical applicability of the valve sizing models and the numerical tool. The equations to calcu- late the thermodynamic property tables are given in [2,3] . In order

@

p

@

t

to calculate the thermodynamic properties as functions of temper- ature and pressure the EoS with the parameters for nitrogen is combined with the specific heat capacity at constant pressure as a function of temperature at one reference value of the pressure. This is the minimum set of information from which all thermody- namic quantities in the whole pressure and temperature range can be derived in a thermodynamically consistent way, i.e. by obeying the Maxwell relations. The dynamic viscosity of nitrogen is defined according to the ri- gid, non-interacting sphere model [11] . Its thermal conductivity is defined according to the modified Euken model [11] .

2.2. Discretization

To reduce computational time the safety valve model is reduced to a 2 slice of the full three-dimensional safety valve model. The horizontal outlet of this quasi axially-symmetrical model is re- duced to obtain a similar outflow area as in the three-dimensional safety valve. Moreover, upstream influences on the force on the disk are likely to be small since the flow near the disk is supersonic. The valve spindle consists of the truncated cone with an angle of 40 that presses on the sharp edge of the valve seat. Fig. 3 illus- trates a mesh deformation dh , where the mesh cells at this edge experience nearly only shear so that the mesh orthogonality be- comes poor. This effect becomes even larger for more dense meshes with at least 10 cells in the smallest cross-section and at very small disk lifts in the order of the minimum required bound- ary layer thickness. The numerical method can only handle con- nected computational domains so that a closed valve has to be approximated by a valve with small disk lift. This disk lift is chosen to be 0.01 mm, which is only 1% of the nominal disk lift. Forces on the valve disk do not only lead to displacement of the disk, but also to deformation of disk and spindle. Just before the valve opens the spindle is deformed in the axial direction, since the force exerted upon the disk by the gas or liquid is equal to the spring force, but in opposite direction. Since the valve opens proportionally, the change in deformation during the opening of the valve is relatively small. A characteristic additional flow force of 200 N and Young’s modulus of steel lead to an extra deformation of the spindle of approximately 0.001 mm, which is small com- pared to the nominal disk lift. Therefore, in this work deformation and the associated aerodynamic noise have not been taken into account. In order to cover the whole disk lift range of 3 mm, 17 grids are generated, which are structured in the region close to the smallest cross-section where the largest flow gradients are located. An unstructured hexahedral grid is used in the rest of the computa- tional domain. Unstructured hexahedral grids are used to have the best compromise between geometrical flexibility and the pos- sibility to refine the mesh locally without deteriorating the mesh quality.

the mesh locally without deteriorating the mesh quality. Fig. 3. Mesh deformation around safety valve seat
the mesh locally without deteriorating the mesh quality. Fig. 3. Mesh deformation around safety valve seat

Fig. 3. Mesh deformation around safety valve seat of a grid with reduced mesh density.

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

111

Fig. 4 shows an example of a grid at h = 0.89 mm. From the smallest disk lift of 0.01 mm each grid is predefined with a factor 1.5 larger disk lift compared to the previous one resulting in the 17 predefined meshes in total. The smallest nodal distance of the mesh with disk lift h = 0.01 mm is 0.001 mm which increases up to 0.025 mm for the larger meshes. The meshes have approxi- mately 42,000 nodes. For a small part of a simulation with only two subsequent predefined meshes a grid convergence study has been performed, in which four sets of grids with approximately 8000, 16,000, 42,000 and 100,000 nodes. The results for flow force and valve displacement on the coarsest mesh differ considerably from the finer meshes, whereas the results on the two finer grids almost collapse. Therefore, the grid with approximately 42,000 points has been selected for the full analysis of the valve movement.

2.3. Solution strategy

In the CFD software package ANSYS CFX the numerical discret- ization is node based and it uses shape functions to evaluate the derivatives for the pressure gradient term and the diffusion terms in the momentum, continuity and turbulence equations. The Na- vier–Stokes equations are discretized in a collocated way and solved by an algebraic multigrid solver. To avoid pressure–velocity decoupling, a robust interpolation scheme similar to Rhie–Chow interpolation [12] is used. CFX solves the conservation equations of mass and momentum in one system of equations, with all equa- tions being fully coupled [1] . The turbulence equations are solved coupled as well and the energy equation is solved separately. CFX uses advection schemes such as first-order upwind differ- ences and numerical advection with a specified blend factor. This blend factor can be varied between 0 and 1 to vary between a first- and second-order differencing scheme and to control numerical diffusion. The high-resolution scheme option will be chosen, which maintains the blend factor as closely as possible to 1 without vio- lating the boundedness principle that could result in non-physical oscillations in the solution. For the turbulence equations the first- order accurate scheme is sufficient. The grid convergence study mentioned in the previous subsection indicated that the order of accuracy of the discretization method is sufficient for the purpose of this research. In unsteady simulations a second-order accurate linear multi step method is applied for the time integration of

the mass, momentum and energy equation, while a first-order backward Euler scheme is applied for the turbulence equations. All inlet and outlet boundaries have subsonic conditions. Hence, at an inlet boundary the total pressure and total temperature are

prescribed, while at an outlet boundary the static pressure is pre- scribed. The walls of the high-speed flow are considered adiabatic.

A simulation with multiple moving meshes starts with the def-

inition file with basic solver parameters and one predefined non- deformed mesh that is consistent with the valve disk displacement defined in the definition file. The first simulation is initialized with the solution of a steady flow simulation at the same initial disk lift with velocity zero. When during a simulation a second mesh is loaded additional parameters such as the spindle velocity and flow force are updated from the previous simulation to the CFX Com-

mand Language (CCL).

A series of simulations with different predefined meshes, the

communication of the variables necessary for the calculation of the valve displacement and the export of the solution variables for postprocessing are controlled with Perl scripts.

As part of each time step first the mesh movement during the

next time step is evaluated, where the acceleration of the disk h n

is determined from the valve displacement h n and flow force F by

flow

n

F

h n ¼

n flow k spring ðh n þ h 0 Þ m spindle g

m spindle

:

ð 3Þ

Then the displacement of the disk h n +1 at the next time step becomes

_

h n þ1 ¼ h n þ h n Dt þ

1

2

h n Dt 2 ;

_

ð 4Þ

where D t is the time step and h n is the velocity of the disk.

Moreover,

_

h n þ1 ¼

_

h n þ h n Dt :

ð 5Þ

After the mesh movement the simulation time step size is eval- uated. In order to minimize errors in the interpolation of the numerical solution to the next predefined mesh, the disk lift on the previous mesh should be as close as possible to the predefined

mesh should be as close as possible to the predefined Fig. 4. Front surface of axially

Fig. 4. Front surface of axially symmetric computational domain of safety valve for h = 0.89 mm.

112

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

disk lift of the next mesh. In order to achieve equality of these two disk lifts within 1 nm, the time step size is changed with not more than 1% between a minimum and maximum value. The time step in the simulations has been optimized in order to have a good balance between calculation time and accuracy. It turned out that a time step size of 2 10 5 s for incompressible flow and of 2 10 6 s for compressible flow lead to solutions in which the disk lift and flow force on the disk are practically inde- pendent of the time step size. These values of the time step result in up to eight coefficient loops of the linear solver per time step after a few time steps after interpolation of the solution variables from a deformed mesh onto a new predefined mesh. The inner coefficient loop terminates when the maximum residuals of the mass, momentum and energy transport equations become below 10 3 or after 50 iterations. All dynamic valve simulations have been performed on a single core of an Intel Xeon 5130 dual-core processor at a clock rate of 2 GHz. For the incompressible flow simulations the computational time is 8 days for a simulation time of 0.1 s. For the compressible flow simulations a time span of 0.01 s could be simulated within 8 days.

3. Results

In this section several typical simulation results will be shown and analyzed. In the first subsection the results of the numerical model of an opening valve with liquid flow are presented. In partic- ular, velocity and pressure distributions will be analyzed. The re- sults are also compared with results of a quasi-static approach. In the next subsection the results of an FSI simulation of an open- ing valve with gas flow at 40 and 400 bar set pressure are pre- sented and discussed.

3.1. Liquid valve flow

Fig. 5 shows results of transient simulations with multiple meshes of the axially symmetric safety valve with liquid flow. The simulation started on the predefined mesh at disk lift h = 0.01 mm, with an initial spring force based on a set pressure of p set = 40 bar and spring stiffness k spring = 25328 N/m, a constant 10% overpressure at the inlet p inlet = 44 bar, outlet pressure 1 bar, equivalent mass of the moving components m disk = 0.7662 kg and simulation time step size D t = 2 10 5 s. The valve opens to approximately its maximum disk lift of 3 mm.

valve opens to approximately its maximum disk lift of 3 mm. Fig. 5. Disk lift, flow

Fig. 5. Disk lift, flow force and mass flow rate versus time of an FSI simulation with liquid water at p inlet = 44 bar.

At time t = 0 s the mass flow rate equals m ¼ 0 : 013 kg = s, which is comparable to a leaking safety valve. In the first 3 ms the flow force remains constant although the disk lift increases to 0.2 mm. In the first part of the valve opening the acceleration forces domi- nate the valve movement. At t = 8 ms and disk lift h = 1.13 mm the flow force drops relatively fast to half of its maximum value in only 0.3 ms. This results in decelerating of the valve spindle. The oscillations of the flow force during the first few iterations of a new simulation with a predefined mesh are small, as can be seen in Fig. 6 . These oscillations, if present, reduce within 10 time steps to zero and, therefore, do not affect the total behavior of the flow force. Especially for incompressible flows the flow force is sensitive to small interpolation errors and inaccuracies that occur during initialization of the simulation. The change in direction of the flow in this time period is visual- ized in four vector plots of the velocity field around the lifting-aid in Fig. 7 . It can be seen that at t = 7.2 ms and h = 0.89 mm the flow remains attached to the valve cone and directly impinges on the

_

attached to the valve cone and directly impinges on the _ Fig. 6. Flow force as

Fig. 6. Flow force as function of time of simulation with liquid water at p inlet = 44. The dashed vertical lines correspond to the times of a new mesh.

u [m/s] 100 75 t = 0.0072 s t = 0.0076 s 50 25 0
u [m/s]
100
75
t
= 0.0072 s
t = 0.0076 s
50
25
0
t
= 0.0089 s
t = 0.0106 s

Fig. 7. Vector plots of velocity field of water flow at four different simulation times at p inlet = 44 bar.

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

113

bottom of the cavity. Then at t = 7.6 ms and h = 1 mm the flow starts to detach from the valve cone and the circulating bulk flow moves away from the bottom of the cavity. At t = 8.9 ms and h = 1.39 mm the flow has completely detached and impinges on the tip of the lifting-aid with three vortices in the cavity. Finally, at t = 10.6 ms and h = 1.75 mm, two vortices remain in the cavity and the flow has stabilized again. After the jump in force the disk lift almost reaches the mechanical stop at 3 mm, but the flow force recovers only slowly to its initial value. Up to t = 0.1 s no further force discontinuities have been observed. Fig. 8 shows the forces on different parts of the wall of the spin- dle that significantly contribute to the total force on the valve as a function of time. Only the parts of the wall with forces larger than 1 N are plotted. This figure clearly shows the fast change of the force distribution over the wall as a function of time. At t = 0 s only the truncated cone (part 1 and 2) contributes to the flow force. Di- rectly after opening, the cavity of the lifting-aid (part 4) starts to compensate for the reduced force on the truncated cone. At t = 8 ms the tip of the cavity (part 6) only partly recovers the col- lapse of the force on the lifting-aid. In order to determine the influence of the flow dynamics on the valve characteristic, the previously described transient simulation with constant operating pressure p inlet = 44 bar is compared with a so-called quasi-steady simulation. In this simulation steady-state solutions are calculated on all predefined meshes, i.e. for 17 fixed values of the disk lift. The results are converted to the time domain by solving the valve dynamics Eqs. (3) and (4) with values of the disk force interpolated between the steady-state solutions starting

_

at t = 0 s, initial velocity h ¼ 0 m = s and initial spring displacement

h 0 that is exactly the same as used in the transient simulation with multiple meshes. For the whole time domain the disk lifts are cal- culated and the corresponding mass flow rates and flow forces are interpolated. In Fig. 9 the results of both approaches are compared. Up to the highest disk lift, the disk lift and the mass flow rate can be well approximated by the steady-state solutions. However, the large rise and collapse of the flow force found in the transient simulation is not obtained in the quasi-steady simulation. Fortunately, this hardly influences the mass flow rate and flow force at later times. Furthermore, when the valve closes again the valve oscillations are damped in the transient simulation, whereas the quasi-steady approach results in an undamped oscillatory motion of the spindle. This different behavior can be explained in the following way. In

equation of motion (2) a damping term would have the form b h with b > 0. From a physical point of view, the damping accounts for flow-history effects and the velocity of the valve, which is equal

effects and the velocity of the valve, which is equal Fig. 9. Comparison between transient and

Fig. 9. Comparison between transient and quasi-steady simulation results for liquid flow at p set = 40 bar and p inlet = 44 bar. Solid line: transient simulation, dashed line: quasi-steady simulation.

to zero in the calculation of the flow force in the quasi-steady sim- ulation. The results of both simulation shows that the difference be- tween the flow force in the two approaches, D F is correlated with

_

the velocity of the disk h . For example for 0.017 s < t < 0.033 s the

flow force in the transient simulation is larger than in the quasi- steady approach, while the disk velocity is negative. This damping would be even larger when friction effects of the spindle pressing on the guide box are taken into account. In a comparison between the two approaches where the set pressure and overpressure are in- creased by a factor of five, the differences in results are similar. At a valve disk lift of around 1 mm the differences between the quasi-steady solution and the transient solution are large. As pre- viously observed in the analysis of the velocity distributions around the spindle in Fig. 7 , a change in the orientation of the flow results in a sudden change in distribution of the flow forces acting on the individual parts of the spindle wall. The differences in the velocity field of the flow become more apparent when comparing a solution of the quasi-steady approach with the transient one at the same disk lift h = 0.89 mm, which is shown in Fig. 10 .

In the results of the steady simulation in the left of Fig. 10 two

_ clear strong vortices are present and the bulk flow directly im- pinges on the edge of the lifting-aid. In the right figure, which shows the transient simulation results at the same disk lift, the

the transient simulation results at the same disk lift, the Fig. 8. Contributions of the significant

Fig. 8. Contributions of the significant parts of the walls of the disk to the total flow force versus time of an FSI simulation with liquid water at p inlet = 44 bar.

114

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

u [m/s] 100 75 50 25 0
u [m/s]
100
75
50
25
0

Fig. 10. Vector plots of quasi-steady (left) and transient (right) velocity field for water flow at p inlet = 44 bar and h = 0.89 mm.

flow remains attached to the truncated cone and completely flows into the cavity of the lifting-aid, inducing a larger total flow force. It is clear that in the transient approach the flow does not have time to relax so that the redirection of the flow pattern is delayed. This has a large impact on the flow force. This flow-history effect cannot be incorporated in the quasi-steady approach.

3.2. Gas valve flow

Fig. 11 shows results of transient simulations with multiple meshes of the axially symmetric safety valve with nitrogen gas. The simulation started on a predefined mesh with disk lift h = 0.01 mm, initial spring force based on a set pressure p set = 40 - bar and spring stiffness k spring = 25328 N/m, a constant 10% over- pressure at the inlet p inlet = 44 bar, inlet temperature T inlet = 300 K, outlet pressure 1 bar, equivalent mass of the moving components m disk = 0.7662 kg and time step size D t = 2 10 6 s. Due to higher flow velocities the time step size is smaller by a factor of 10 than in the simulations presented in the previous subsection. The force jump previously observed in the liquid simulations al- ready occurs at t = 1 ms and h = 0.035 mm. This detachment of the flow from the valve cone does not lead to such large flow force

from the valve cone does not lead to such large flow force Fig. 11. Disk lift,

Fig. 11. Disk lift, flow force and mass flow rate versus time of an FSI simulation with nitrogen gas at p set = 40 bar. Solid line: transient simulation, dashed line:

quasi-steady simulation.

changes because the disk velocity is low. Nevertheless, the initial increase of the flow force provokes rapid and stable opening of the valve. At t = 22 ms the valve is almost closed and the flow starts to behave irregularly with high-frequent valve oscillations around 25 ms. This oscillatory behavior is illustrated in Fig. 12 . It shows two vector plots generated at two opposite extrema of the flow force. In the left plot the flow in the cavity has a low velocity, resulting in almost no contribution of the cavity to the flow force. In the right plot the flow has changed its direction and the supersonic area with bow shock is smaller, because the flow impinges on the cavity edge with less space to expand. Two strong vortices are present in the cavity leading to a significant contribution to the flow force. When the valve opens again the flow oscillations damp out. It is noted that this unstable flow behavior cannot always be avoided in valve design. It is important that the traveling time of pressure waves moving from the outlet to the valve should not be- come close to a multiple value of the eigenfrequency of the valve. This interaction could induce oscillatory behavior of the valve disk which is called valve chatter. In this simulation this cannot be shown, because the computational domain is too small resulting in too small traveling times. Finally, the damping in this simulation is very low. This shows that flow-history effects are unimportant and implies that this low-pressure safety valve can be well simu- lated in a quasi-steady approach. The quasi-steady simulation re- sults included in Fig. 11 are indeed in good agreement with the transient results. It is also important that the frequency of the oscillatory behav- ior observed in Fig. 12 is not close to the frequency of longitudinal waves in the spindle. The eigen frequency of these waves can be in- ferred from the stiffness of the spindle and its mass, resulting in a frequency of 45,000 Hz. This is larger than the frequency observed in Fig. 12 by a factor of 4. This also shows that spindle deformation can be ignored in the present simulations. Fig. 13 shows the results of FSI simulations of nitrogen gas flow at a 10 times higher set pressure of p set = 400 bar, 10 times higher spring stiffness of k spring = 253280 N/m, constant 10% overpressure at the inlet of p inlet = 440 bar and the same temperature, outlet pressure, equivalent mass of the moving components and time step size as the previously presented gas flow simulation. In this simulation the force always decreases as the valve opens, which results in a permanent detachment of the flow from the

_

valve cone. Furthermore, due to a high disk velocity h the damping is larger and the flow history effects have a larger contribution to

the flow force. Indeed, the transient simulation results differ more

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

115

u [m/s] 700 600 500 400 300 200 100 0
u [m/s]
700
600
500
400
300
200
100
0

Fig. 12. Vector plots at the time of minimum (left) and maximum (right) of oscillatory force of an FSI simulation with nitrogen gas at p set = 40 bar.

FSI simulation with nitrogen gas at p s e t = 40 bar. Fig. 13. Disk

Fig. 13. Disk lift, flow force and mass flow rate versus time of an FSI simulation with nitrogen gas at p set = 400 bar. Solid line: transient simulation, dashed line:

quasi-steady simulation.
quasi-steady simulation.

Fig. 14. Comparison of dimensionless flow forces of steady valve flow calculations of liquid and gas valve flow to 1 mm disk lift (left) and zoomed to 0.1 mm disk lift (right).

from the quasi-steady results, which are also included in Fig. 13 , than in the previous case. Since the flow force continually de- creases during opening, this valve will respond fast only when the pressure increase rate is sufficiently high, so that the inertia of the spindle moves the valve to an open position while the pres- sure at the inlet increases. In Fig. 14 a comparison of the dimensionless flow force of the steady flow calculations of liquid flow at p set = 40 bar and gas flow

at p set = 40 and 400 bar is made. The dimensionless force is defined as the actual flow force divided by the product of inlet pressure p in- let and valve seat area A 0 . The first measurement point is at h = 0.01 mm. It can be seen that for liquid flow the force increases during a longer time span than for gas flows. Furthermore, this ap- proach reveals the differences between the forces just when the valve opens. In the first 0.1 mm the flow force determines the opening characteristic of a safety valve. As a result, this comparison can be used as an indication for determination of the opening char- acteristic of a safety valve.

4. Conclusion

A CFD tool with fluid–structure interaction has been designed for analysis of the opening characteristics of high-pressure safety valves. To model these complex geometries from closing position to nominal disk lift several predefined meshes are necessary to cover the mesh deformation without deteriorating the mesh qual- ity. Especially the attention to the transfer of the solution variables to the next predefined mesh has resulted in a feasible solution method for both incompressible and real-gas flows. The inclusion of FSI in incompressible valve flows has led to new insights in valve dynamics that cannot be observed by a stea- dy-state approach. It is emphasized that in the case of incompress- ible fluids the absence of damping or the ability to absorb pressure waves by the flow poses higher demands on accurate transfer of the solution variables to a new mesh. In addition, the mass flow rate and the disk lift are less affected by large variations of the disk force. However, a redirection of the flow could be induced by trav- eling pressure waves and subsequently lead to unstable valve operation when occurring close to a multiple value of the valve eigenfrequency. The implementation of real-gas dynamics described by the SRK EoS in the CFD model with FSI has led to simulation results which exhibit flow oscillations when the valve is almost closed. Analysis has shown that the flow interacts with the vortices in the cavity which change direction. Furthermore, also the size of the super- sonic area and the position of the bow shock show oscillatory behavior. This behavior can lead to valve chatter when pressure waves travel at a multiple value of the valve eigenfrequency. At high-pressure valve flows, real-gas effects become apparent and lead to a decreasing force when the valve opens. This implies that the rate of pressure increase at the valve inlet becomes more important, because it is expected that this valve will respond fast only when the pressure increase rate is sufficiently high. The open- ing characteristics of a safety valve is determined in the first 0.1 mm. Therefore, it is recommended to use the presented FSI

116

A. Beune et al. / Computers & Fluids 64 (2012) 108–116

multimesh method to investigate the sensitivity to valve dynamics and the steady flow method at small disk lifts to determine the opening characteristics. The outcome of such investigation can be used for the optimization of valve design. Moreover, Beune et al. [3] have shown that dynamic simulations provide insight into the cause of deviations in the flow force between experimental re- sults and numerical results based on steady simulations.

Acknowledgements

The departments Safety & Fluid Flow Technology and Valves and Fittings (Metal Processing) of company BASF SE in Ludwigsha- fen (Germany) are acknowledged for funding this research.

References

[1] ANSYS. Manual ANSYS CFX, Release 11.0. Canonsburg; 2006. [2] Beune A. Analysis of high-pressure safety valves. Ph.D. thesis, TU Eindhoven, the Netherlands; 2009. [3] Beune A, Kuerten J, Schmidt J. Numerical calculation and experimental validation of safety valve flows at pressures up to 600 bar. AIChE J 2011;22. http://dx.doi.org/10.1002/aic.12534 .

[4] Bürk E. Numerische Strömungsberechnung ergänzt durch experimentelle Methoden zur Untersuchung pneumatischer Strömungen. Ph.D. thesis, TU Dresden, Germany; 2006. [5] Domagała M. Cfd analysis of pilot operating relief valve. Czasopismo Techniczne, wydawnictwo politechniki Krakowshiej z. 3-M; 2008. [6] ISO. EN ISO 4126-1:2004. Safety devices for protection against excessive pressure. Part 1: safety valves. Berlin: Beuth Verlag GmbH; 2004. [7] Li L, Zang H, Davis G, Kalb F, Hamid S. Improving the closing characteristics of subsurface safety valve with combined FEA and CFD modeling/numerical analysis. In: SPE western regional meeting; 2005. [8] Mahkamov K. An axisymmetric computational fluid dynamics approach to the analysis of the working process of a solar stirling engine. J Sol Energy Eng

2006;128:45–53.

[9] Menter F. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA J 1994;32:1598–605. [10] Reich A, Parthasarathy V, DiMeo A. Coupled fluid-structure interaction simulation of the opening of the target rock vacuum relief valve. ASME Publ PVP 2001;424:23–8. [11] Reid RC, Sherwood TK. The properties of gases and liquids. 2nd ed. New York: McGraw-Hill; 1966. [12] Rhie C, Chow W. Numerical study of the turbulent flow past an airfoil with trailing edge separation. AIAA 1983;21:1525–32. [13] Soave G. Equilibrium constants from a modified Redklich–Kwong equation of state. Chem Eng Sci 1972;27:1197–203. [14] Srikanth C, Bashker C. Flow analysis in valve using moving grids through CFD techniques. Adv Eng Softw 2009;40:193–201.