Professional Documents
Culture Documents
OMAE2010-20509
ABSTRACT INTRODUCTION
The objective of this paper is to present Computational Fluid The flexible metal pipe has been used in smaller diameters for
Dynamics (CFD) modeling of fully developed turbulent flow more than 30 years for all kind of cryogenic Liquid Natural
through a flexible corrugated pipe and to investigate the Gas (LNG) transfer applications (Refs. [1,2]). Today these
pressure drop reduction potential of liners. This work also LNG loading systems have evolved towards large, complex
aims to establish a framework to be used in large scale industrial systems, which have to respect increasingly
numerical simulations of the offshore transfer of cryogenic stringent rules and standards while continuing to maintain
fluids. A 3-D CFD approach is considered more appropriate high levels of safety and availability. For both design and
than 2-D axisymmetric one, since the wavy corrugation operational standpoint, the LNG from ship to ship loading is a
profiles lead to a great deal of internal turbulent structures for new application of this well known technology (Ref. [2]).
high Reynolds number over Re > 106. Such flexible LNG pipes are usually provided with corrugated
walls. The basic design of a Nexans vacuum insulated LNG
Three geometries of the bellows’ (corrugation) depth are transfer hose is illustrated in Figure 1.
considered to determine the potential value of a cryogenic
liner, corrugation filler or geometric variations for the 16” Relatively minor deviations in corrugation geometry can affect
pipe. The reduction in cost and complexity of developing a the flow/pressure drop characteristics that are important from
robust cryogenic liner or corrugation filler, plus eventual the design and operational standpoint. The relative pressure
certifications, would be significant and needs to be worth the drop per unit length in a pipe differs according to the distance
improvement (decrease) in pressure drop. We conduct a from the inlet, normally related to a distance of some 50 x pipe
straight pipe corrugation depth study for pressure drop (deep inner diameter to achieve a fully developed flow profile. After
corrugation, shallow corrugation and liner), and include this length the flow is normally fully turbulent and the relative
suitable mesh convergence and unsteady simulations. We also pressure drop per unit length is constant (and lowers than in
attempt to validate the friction coefficient data with the the inlet section).
empirical formulas and recent experimental tests. Operational
cryogenic transfer flow rates ranging from Q=1000 m3/h to Armoured metallic corrugated pipes are well known structures
Q=5000 m3/h are considered. which can withstand tensile and internal pressure loads, as
well as perform better from a fatigue and heat transfer
1
standpoints. However, series of corrugations can induce To begin, it is useful to summarize briefly the essential
complex and undesirable flow behavior in the pipes. The elements of wall turbulence, mainly to establish notation and
wavy configuration of the corrugations promotes turbulence, define some basic terms. From simple observations, the effect
which is a problem because of the extra work required to of turbulence on the mean flow is to flatten the profile relative
surmount the pressure drop. In a broad sense, corrugations to the parabolic profile that occurs in pipe flows. The total
may lead to the following flow physics phenomenon: shear stress is sum of the Reynolds stress − ρ uv and the
Large pressure head-loss viscous stress µ ( du / dy ) and it is defined as
Flow induced pulsations τ = µ ( du / dy) − ρuv (2)
Multiphase with bubbles and cavitations
Increase heat transfer
where µ denotes the dynamic viscosity of the fluid.
2
catenary with varying curvatures) using unstructured meshes. simulation, the time integration is automatically set to 1st
The followings are the basic components of CFD modeling. order accuracy. In this configuration, the timestep size taken
by AcuSolve is set to infinity (1.0x1010 s) to convect the errors
Turbulence Modeling through the domain and arrive at the steady solution. The mass
continuity and momentum differential equations are then
converged to 4 decades of the solution accuracy. This residual
The turbulence level is typically high due to the corrugations
reduction ensured iterative convergence in all cases.
and turbulence modeling is critical to get the accurate
predictions. To model the steady effects of the turbulence on
For unsteady cases, we try to minimize the additional effects
the mean flow field, we employ the Spalart-Allmaras
of finite element stabilization in both space and time
Reynolds Averaged Navier Stokes (RANS) model. This model
integration so that numerics do not suppress the small scales
is a general purpose model that provides reasonable results for
of turbulent eddy motions. For the spatial integration, we
a broad array of industrial applications. This model relies on
switch off the discontinuity capturing operator for the DES
the fact that the complete turbulence behavior has to be
model and we set the lumped mass fraction to zero. In time
enclosed within an appropriate eddy viscosity variable which
domain, we utilize an optimal high frequency damping factor
takes into account all turbulence scales (from the largest
to be 0.5 for the generalized second-order time integration
eddies to the Kolmogorov scale). This model solves a single
scheme. A global time step of 1 x 10-2 sec is used to capture
transport partial differential equation (PDE) for the eddy
the unsteady turbulent motions.
viscosity with the appropriate boundary conditions and solver
settings.
Initial Conditions
For unsteady simulations, we employ Delayed Detached Eddy
Simulation (DDES), a hybrid RANS model with Large Eddy For steady state simulations, the flow solution for the velocity
Simulation (LES). In the LES based on dynamic subgrid scale vector is initialized by the entrance averaged flow speed and
estimation, an attempt is made to capture the large scale pressure field is set to be zero. For unsteady calculations, the
unsteady motions which carry the bulk of the mass and flow solution field is initialized by the steady state solution of
momentum in a flow, but the near wall turbulence behavior is the corresponding flow rate. The flow solution is then
treated with a wall function. In the DDES model (Refs. [6,9]), advanced in time to obtain the unsteady turbulent motions
we resolve the large eddies that have the biggest effect on the with fixed and small time-increment to resolve the unsteady
wall shear stress and use the RANS equations to describe the local motions.
flow near the wall. This was done not only to economize on
mesh size, but also because most pipes have relatively rough Boundary Conditions
walls. Wall functions reduce mesh size by providing an
integrated relationship between the wall and the logarithmic
In various industrial applications, the flow physics and
region of the boundary layer.
geometry can be modeled as repetitive in nature by applying
periodicity. Thereby, a representative building block can be
Material Model
considered for computational efficiency and simplicity, while
maintaining the desired accuracy of flow physics
In all simulations, the working fluid LNG is viscous and in corresponding to the experiments. In this purpose, the flow
liquid phase with constant density, that is to say profiles at exit are iteratively reported to the entrance (Ref.
incompressible flow. The flow is assumed to be isothermal, [7]). In AcuSolve, to simulate the large length of corrugated
i.e., the energy equation is not solved. The flow rate and hose with fully developed flow, periodic conditions are
Reynolds number based on the averaged velocity are the input applied between the outlet (exit) and inlet (entrance) of the
for the models. Table 1 summarizes the material properties of domain.
LNG.
Table 1: Summary of the properties of LNG The quantity imposed at the entrance is the flow rate Q or
equivalently the averaged velocity as the surface integrated
Property Value boundary conditions (bulk BC). This is the only user input to
define the fully developed flow and Reynolds number based
Density 450 kg m -3 on the averaged velocity. Instead of outlet condition at the exit
Dynamic viscosity 1.4 x 10-4 Pa sec plane, the classical periodicity on the mean velocity and the
eddy viscosity are used to couple the entrance and the exit
Kinematic viscosity 3.11x 10-7 kg m-1 s-1 boundaries. By this way, we attempt to achieve the similar
stabilized profiles of mean velocity and eddy viscosity at the
Solver Settings entrance and the exit. For the pressure, the condition based on
the constant offset is imposed to obtain the fully developed
The standard solver settings were specified in AcuSolve for like condition. In other words, the pressure can change along
steady RANS calculations. When performing a steady the stream-wise direction and the pressure can be decomposed
3
into a variable term and a linear varying term in the resolve turbulent boundary layer and to find the optimized
streamwise direction as volumetric mesh distribution.
dp
p = p − x (4)
dx
NUMERICAL MODELING
Model Description
Mesh Generation
4
discretization parameters and to perform sensitivity analyses. Table 4: Mesh statistics for the liner #1 design model,
For CFD, this typically entails a mesh refinement study as A*=0.01583
well as a near wall modeling sensitivity analysis. A reasonable
mesh sensitivity analysis was performed for this work. Mesh Fine mesh (Ref.) Medium mesh
(wall function)
An important issue in the accurate prediction of industrial Total no of elements 35,368,685 19,882,402
turbulent flows is the formulation and the numerical treatment
of equations in regions close to solid walls. The near-wall Number of nodes 6,113,454 3,429,323
formulation determines the accuracy of the wall shear stress
(friction factor) and has an important influence on the
development of boundary layers and near wall turbulence Table 5: Mesh statistics for the liner #2 design model,
structures. Typically the two following approaches are used to A*=0.00798
model the flow in the near-wall region: (a) the wall function
method (b) the wall integration (low Reynolds number) Mesh Fine mesh (Ref.) Medium mesh
method. (wall function)
Because of the linear variation of total stress, the wall shear Total no of elements 33,107,421 18,780,090
stress and the Reynolds (turbulent) stresses are related, Number of nodes 5,728,804 3,240,773
justifying the introduction of the friction velocity
uτ =
τw SIMULATION RESULTS AND VISUALIZATIONS
ρ as a scale representative of the turbulent
fluctuations. The viscous length scale alluded to the viscous We present full-scale 3D simulations using the steady RANS
boundary layer thickness is given by δv = ν u and the
and transient DDES models for the three configurations for the
τ range of flow rates.
distance above the wall scaled in wall units is denoted by
y Base Corrugated Model
y+ = (5)
δv We first perform 3D RANS simulations using the wall-
ν =µρ
function approach with the fine mesh. As mentioned earlier,
where y is the distance from the wall and is the length of modeled domain is 6D as shown Fig. 3.
kinematic viscosity.
5
Figures 4 (top) shows the contours of velocity magnitude (top)
using the RANS model at the Reynolds number of Re=9.38 x
106 for the base model of corrugated pipe. The fully developed
and time averaged steady flow behavior can be observed from
the figure. As expected from the RANS model, there are no
physical unsteady motions in the velocity field. Figures 4
(bottom) shows the contours of streamwise velocity at the
cross section of the corrugated pipe with the DDES model.
The 3D turbulence structures and unsteadiness in the flow can
clearly be inferred in the image
3.00E-01
v'/U
w'/U
2.00E-01
Normalized velocity fluctuations
u'/U
1.00E-01
The interactions between these structures are essential In this section, we conduct a simple straight pipe corrugation
ingredients of wall-bounded turbulence. These fluctuating depth study with the DES model. The meshing guidelines and
vorticity structures are somewhat organized, often called flow conditions are similar to those used for the base model.
vortical flow structures. On further flow visualizations, Figs. 8-9 show comparison of
velocity magnitude at the cross-sectional and streamwise
planes. In the images, we can see a greater degree of
turbulence structures (red color zone) in the liner 1 model as
compared to the liner 2 model.
6
This is equivalent to obtaining the turbulence statistics from
the single-point probe. The velocity fluctuations are
normalized by the mean velocity. The liner 2 model decreases
the velocity fluctuations (i.e., turbulence level) by a factor of 5
compared to the liner 1 model.
1.5E-01
5.0E-02
0.0E+00
-5.0E-02
v'/U
-1.0E-01
w'/U
u'/U
-1.5E-01
1.00E+00 1.10E+00 1.20E+00 1.30E+00 1.40E+00 1.50E+00
Normalized time
1.5E-01
5.0E-02
0.0E+00
Figure 8. Streamwise variation of velocity magnitude
contours for the liner 1 (top) and liner 2 (bottom) models -5.0E-02
at Q=3333 m3/h v'/U
-1.0E-01 w'/U
u'/U
-1.5E-01
1.00E+00 1.10E+00 1.20E+00 1.30E+00 1.40E+00 1.50E+00
Normalized time
7
Completely rough regime: In the present configuration, 1.25E-01
the corrugations may be considered equivalent to periodic
roughness (i.e., 2k/Dh) of the diameter of pipe. A
1.00E-01
systematic investigation on the effects of Reynolds
Coefficient of Friction
number and relative roughness k on the friction factor was
performed by Nikuradse Ref. [5]). For the higher values 7.50E-02
of Reynolds number (ranging from 104 to 107), the Baseline
Smooth
factor for the completely rough regime was estimated as
Base Rough. Theory
2.50E-02
Liner1 Rough. Theory
1
f= 2
(8) Smooth Pipe Theory
Dh 0.00E+00
1.74 + 2 log 2k 1.0E+05 1.0E+06 1.0E+07 1.0E+08
Reynolds number, Re
Figure 11 shows the variation of coefficient of friction for the Figure 12 shows a cross plot of friction factor vs. the depth of
range of Reynolds number for the three configurations and the corrugation A*, which provides another view of the impact of
smooth pipe. The friction factor was determined by evaluating reducing the corrugation depth. For the range of flow rates,
the pressure gradient along the pipe from the integrated the friction factor increases as we increase the depth of
pressure values. For the baseline case, the friction coefficient corrugation. This suggests the potential value of liners for
is consistently larger than the liner 1 & liner 2 models. The reducing the pressure loss. The friction factor values for the
wall shear stress of the liner 2 model is converging towards liner 1 are converging to that of the smooth pipe. The amount
the stress values corresponding to the smooth pipe. This decrease doesn’t appear to be simply a linear function.
implies that, by introducing liner materials, the coefficient of
friction can be reduced by 80% with respect to the deeper 0.125
metallic hose configuration. Due to complex flow behavior
and recirculation in the base & liner 1 models, the friction
factor changes significantly with the Reynolds numbers. 0.1
Coefficient of Friction
8
improved match between the RANS and DDES may be reasonable agreement between the CFD and experimental
obtained. values can be seen.
1.20E-01
For the base and liner 1 geometry at Re~10M, an inflectional Base CFD Model
behavior in the pressure drop and wall shear stress have been
Liner 1 CFD Model
observed in the RANS and DDES results. As mentioned
earlier, this dip in the frictional drag may be attributed to the 1.00E-01 Water Test (Ref. [1])
Coefficient of Friction
sudden shift in the point of separation for the base and liner 1
geometries. In this range, the laminar viscous sub-layer
portion of boundary layer becomes unstable and undergoes 8.00E-02
transition to turbulence. As seen in the flow over a cylinder,
the turbulent boundary layer, because of its greater energy, is
able to overcome a large adverse pressure gradient. The
turbulent boundary layer separates at a further downstream 6.00E-02
location along the corrugation profile, resulting in a thinner
wake and a pressure distribution more similar to that of
potential flow. For values of Re >10M, the separation point
4.00E-02
slowly moves upstream as the Reynolds number is increased,
1.0E+05 1.0E+06 1.0E+07 1.0E+08
resulting in an increase of the friction factor. For the liner 2 Reynolds number, Re
and smooth pipe, the geometry is streamlined and the point of
separation and the transition of boundary layer remain Figure 14. Comparison of the CFD results for A*=0.0604
somewhat unchanged. of 16” ID pipe with the water test with A*=0.0513 in 10.5”
ID pipe
Next, we would like to establish the relevance of friction
1.25E-01
factor in actual pipe configurations as function of pipe lengths
and corrugation depths. This may be important to estimate the
1.00E-01 capacity and limitations of LNG pumps for the corrugated
pipe flow. Typical LNG pumps can develop 6.6 bar
C o e fficien t o f F rictio n
Base
Liner 1
Figure 13. Variation of friction coefficients for the range of 40 Liner 2
Reynolds number for the RANS and DDES models
30
We also present a comparison with the experimental test, and
perform further analysis of the results to facilitate 20
interpretation in the commercial settings. The experiment tests
were performed on a 268 mm (10.5”) ID pipe for a range of 10
flow rates with fresh water as working fluid. All the pressure
drop readings were average values taken over a period of 1 0
minute ‐ 2,000 4,000 6,000 8,000 10,000 12,000
Flow Rate Q [m^3/hr]
Figure 14 shows the comparison of CFD values with the
experimental test done with water in 10.5” ID pipe (Ref. [1]).
Figure 15. Pressure drop as a function of flow rates for the
The friction factors are compared with respect the non-
16” pipe with three corrugation geometries
dimensional dynamic similarity parameter, Reynolds number.
The depth and shape of the corrugation profiles are marginally
different between the 16” ID pipe and 10.5” pipe. A
9
periodic variations of cross-sectional area”, ASME, J. Heat
Transfer, Vol. 99, pp. 180-186
CONCLUSIONS
[8] Piomelli U. and Balaras, E. “Annual Review of Fluid
Cryogenic flexible pipe based LNG transfer system seems to Mechanics”, Vol 34, pp 349-374, 2002
be a good candidate for CFD modeling, and to qualify the pipe
system for the LNG industry requirements. [9] Shur, M.K., Spalart, P.R., Strelets, M.K. and Travin,
A.K.,”A hybrid RANS-LES approach with delayed-DES and
Significant 3D turbulence effects were found for the pipe wall-modeled LES capabilities”, International Journal of Heat
geometry with circular corrugations suggested by both and Fluid Flow, Vol 29, pp 1638-1640, 2008.
qualitative features and quantitative information. The 3D
steady RANS and DDES models provided a consistent [10] Allen, J.J, Shockling, M.A., Kunkel, G.J. and Smits,
estimate of the pressure drop and friction factor for varying “Turbulent flow in smooth and rough pipes”, Philosophical
flow rates. We hoped that we have done a reasonable job in Transactions of the Royal Society, vol. 365, No. 1852, 2007
predicting complex internal turbulent structures with the
unsteady DDES simulations.
REFERENCES
10