Beruflich Dokumente
Kultur Dokumente
a r t i c l e i n f o a b s t r a c t
Article history: Liquefied Natural Gas transport and storage has become very important due to its ability to occupy 1/600th
Received 7 May 2010 of the volume that compressed natural gas would occupy at room temperature and atmospheric pressure.
Accepted 30 July 2010 In the present work, an LNG storage tank has been computationally simulated and probabilistically eval-
Available online 10 August 2010
uated in view of the several uncertainties in the fluid, structural, material and thermal variables that govern
the LNG storage tank. A finite element code ALGOR was used to couple the thermal profiles with structural
Keywords:
design. The stresses and their variations were evaluated at critical points on the storage tank. Cumulative
Probabilistic analysis
distribution functions and sensitivity factors were computed for stress responses due to fluid, mechanical
Liquefied natural gas
Storage tank
and thermal random variables. These results can be used to quickly identify the most critical design vari-
ables in order to optimize the design and make it cost effective. The total heat gained by the liquid part in the
tank has been evaluated and the amount of boil-off was calculated. Various methods have been proposed to
minimize thermal stresses.
Ó 2010 Elsevier Ltd. All rights reserved.
1359-4311/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2010.07.033
2764 R.S. Reddy Gorla / Applied Thermal Engineering 30 (2010) 2763e2769
tank, it would soak into the insulation. In the event of such an axisymmetric model of the tank in ALGOR, a FEA package which
accident, some insulation layers would not be able to function, and performs design analysis, simulation and the optimization. The
the outer concrete tank could be structurally compromised by maximum von Mises stresses, and cumulative distribution functions
a quick temperature drop [3]. In conclusion, the safety of the tank and sensitivity factors of stress were evaluated for the 54 random
can be ensured only through a thorough thermal analysis. variables. The maximum boil-off in terms of weight percent per day
Graczyk et al. [6] performed a probabilistic analysis of the was evaluated at the mean value of the random variables.
sloshing-excited tank pressures in LNG tankers. They found that ship
motion results in violent fluid motion in the tank which causes high 2. Analysis
tank pressure. The pressure was measured in a series of a model
tests, and the important issues of structural responses, such as the 2.1. Finite element solution for heat transfer
significance of spatial and temporal characteristics of sloshing loads
as well as the model scaling problem were addressed. Let us consider a two-dimensional partial differential equation
Design specification of the tank has to be studied thoroughly in of the form
order to do a simulation of the storage tank. The properties of all the
parts of a storage tank have been described by Jeon and Park [7].
1 v vT v vT
They discussed safe and economical construction for the above- rKrr þ Kzz þ PT þ Q ¼ 0 in A (1)
r vr vr vz vz
ground LNG tanks. As the capacity increases, special attention needs
to be given to the design code and an efficient procedure needs to be with the boundary conditions
established to design an LNG tank with structural and cost efficiency.
Various analyses have been carried out by KOGAS technology [8],
T ¼ T0 on L1 (2)
including static analysis, wind loading, modal and seismic analysis,
temperature modeling, leakage modeling, pre-stress/post tensioning, or
burn-out modeling, relief valve heat flux modeling and soilestructure
interaction. They used the LUSAS finite element modeling software to vT vT
design a 200,000 m3 above-ground tank. They considered an Krr nr þ Kqq nz þ aT þ b ¼ 0 on L2 (3)
vr vz
axisymmetric model for the static stress and thermal analysis, and 3D
The corresponding functional is
shell elements for modal analysis and seismic analysis. For burn-out
modeling, they performed a transient thermal analysis and predicted ZZ ( 2 2 )
1 vT 1 vT 1
the time required for the fuel to burn-out completely. I ¼ Krr þ Kzz PT 2 QT 2pr dA
2 vr 2 vz 2
Natural convection causes circulation of the LNG within the A
Z
storage tank which tends to maintain a uniform liquid composition. 1 2
The addition of new liquid can result in the formation of strata of þ aT þ bT 2pr dL2 : (4)
2
slightly different temperature and density within the LNG storage L2
tank. “Rollover” refers to the rapid release of LNG vapors from
Here, nr and nz are direction cosines of the outward normal to L2 .
a storage tank caused by stratification. The potential for rollover arises
Simplex elements for this problem are axisymmetric rings
when two separated layers of different densities (due to different LNG
whose properties are independent of the angleq.
compositions) exist in a storage tank. LNG rollover phenomena
Element interpolation functions are taken as linear, of the form
received considerable attention following a major unexpected vent-
ing incident at an LNG receiving terminal at La Spezia, Italy in 1971 [9].
T ðeÞ ¼ Ni Ti þ Nj Tj þ Nk Tk (5)
The main hazard arising out of a rollover accident is the rapid release
of large amounts of vapor leading to potential over-pressurization of where the pyramid functions are
the tank. It is also possible that the tank relief system may not be able 8
1 < ai ¼ Rj Zk Rk Zj
to handle the rapid boil-off rates, and as a result the storage tank will
Ni ¼ ðai þ bi r þ ci zÞ; b ¼ Z Z
fail leading to the rapid release of large amounts of LNG. 2A : c i ¼ R j Rk
i k j
Salem and Gorla [10] performed a probabilistic finite element
thermal analysis of a water tank to determine critical design
8
parameters and to perform design optimization. Thacker and 1 < aj ¼ Rk Zi Ri Zk
Nj ¼ a þ bj r þ cj z ; b ¼ Z Z
a team of researchers [11] at the Southwest Research Institute 2A j : c j ¼ Rk R i (6)
j i k
described the development of the NESSUS probabilistic engi- 8 ak ¼ Ri Zj Rj Zi
neering analysis software. Gorla et al. [12] performed the first <
1 bk ¼ Z i Z j
probabilistic study that interconnected computational fluid Nk ¼ ðak þ bk r þ ck zÞ;
2A : ck ¼ Rj Ri
dynamics and finite element structural analysis. In this study
a combustor liner was simulated by using the finite element
and 2A ¼ bi cj þ bj ci
method and evaluated probabilistically. The inlet and outlet
temperatures were found to greatly influence the hoop stress. Gorla Here, Ri and Zi denote the coordinates of the node i.
and Gorla [13] performed a probabilistic analysis of a non-gasketed The element minimization equations are
flange. Cumulative distribution functions and sensitivity factors
were computed for heat loss due to 11 random variables. Gorla and
Haddad [14] performed a finite element heat transfer and structural 8 9ðeÞ
> vI >
analysis of a cone-cylinder shell pressure vessel. They presented >
> >
>
>
> vT >
>
sensitivity factors for the stress versus the random variables. >
< vIi >
=
The objective of the present work is to design a robust LNG ¼ ½BðeÞ fTgðeÞ fCgðeÞ (7)
>
> vTj >
>
storage tank system which continues to function well when the >
> >
>
> vI >
>
>
operating conditions are not ideal. The location and magnitude of : ;
the maximum thermal stress was evaluated using a finite element vTk
R.S. Reddy Gorla / Applied Thermal Engineering 30 (2010) 2763e2769 2765
The effect of temperature can be accounted for by considering The vector e0 is the strain due to the average temperature change in
the element. q represents the element nodal load distributions that
e
the strain energy term.
Z must be added to the global force vector.
1 The stresses in an element are then obtained in the form
U ¼ ðe e0 ÞT Dðe e0 ÞtdA
2Z
1 (19)
¼ eT De 2eT De0 þ eT0 De0 tdA s ¼ DðBq e0 Þ (23)
2
The first term in the previous expansion gives the stiffness
matrix derived earlier. The last term is a constant, which has no
effect on the minimization process. The middle term, which yields 2.3. Linear steady-state heat transfer analysis
the temperature load, is now considered in detail. Using the strain-
displacement relationship e ¼ Bq, Linear steady-state heat transfer occurs when the material’s
conductivity is not dependent on temperature. But in our case, the
Z X properties of outer tank are dependent on temperature. Nonlinear
eT De0 tdA ¼ qT BT De0 te Ae (20)
heat transfer analysis is considered in this paper.
e
A Fourier’s law of heat conduction is given by Q ¼
This step is obtained using the Galerkin approach where eT will kAVT.where, Q ¼ heat flow; k ¼ thermal conductivity (a
be eT ðfÞ and qT will be JT . The symbol f defines the shape function constant) entered as a material property. Isotropic materials fall
and J defines the weight function. It is convenient to designate the under this category; A ¼ cross sectional area of an element face;
element temperature load as VT ¼ the temperature gradient in the direction normal to the
area, A
qe ¼ te Ae BT De0 (21) The convection heat transfer is given by Q ¼ hADT.
where, h ¼ convective heat-transfer coefficient (constant) entered by
Where, the user; A ¼ area of the element subject to convection; DT ¼ TsT;
Ts ¼ surface temperature of the element (calculated); T ¼ temperature
qe ¼ ½q1 ; q2 ; q3 ; q4 ; q5 ; q6 T (22) of the fluid (assumed to be constant) and entered by the user.
R.S. Reddy Gorla / Applied Thermal Engineering 30 (2010) 2763e2769 2767
The heat flux (heat transfer/time/area) experienced by a surface increasingly important in design and certification efforts. Traditional
subjected to thermal radiation is described by the following equations: design approaches typically use worst case assumptions and safety
q00 ¼ Frad s T4s T4rad factors to certify a design. This approach is overly conservative, does
not quantify the reliability; nor does it identify critical parameters or
where, q" ¼ thermal radiative heat flux; Frad ¼ radiation view failure modes affecting the system performance.
factor, which includes absorptivity, emissivity and view factor A probabilistic analysis approach characterizes input vari-
effects; s ¼ StefaneBoltzmann constant; Ts ¼ calculated surface ability using probability density functions and then propagates
temperature on an absolute scale; Trad ¼ ambient temperature. these density functions through the performance model to yield
Due to the higher outside ambient temperature relative to that of uncertain model outputs, which can be related to failure metrics
the cryogenic liquid, heat is always being transferred to the cryogenic such as fatigue life, rupture, or stress intensity. The approach
liquid. The bottom layer of the LNG tank is always at the higher quantifies the reliability, can reduce over-conservatism, and
temperature than the top layer of LNG. A buoyancy force is generated identifies critical parameters and failure modes driving the reli-
in the fluid when it is heated or cooled by a surface. Since hot LNG is ability of the system.
less dense than the cold LNG, natural convection occurs due to The programmers and researchers try to achieve the following
differences in density. This causes motion in the fluid as the warm in the development of the analysis algorithm.
fluid rises and the cool fluid is then moved to the surface where it will
be heated. Identifying sources of errors and uncertainties.
Developing probability distributions for input variables.
Determining spatial and temporal variations.
2.4. Probabilistic analysis Developing probabilistic load modeling.
Tailoring failure models for modeling uncertainty and obtain-
The ability to quantify the uncertainty of complex engineered ing appropriate system performance measure.
systems subject to inherent randomness in loading, environment, Creating system models (multiple failure mode and
material properties, and geometric parameters is becoming components).
2768 R.S. Reddy Gorla / Applied Thermal Engineering 30 (2010) 2763e2769
NESTEM enables designers to achieve reliable and optimum value of 0.99. Only twenty five of the 54 variables that influence
designs subjected to a life constraint with a probabilistic treatment stress the most are presented in this figure. For the variables
of key uncertainties. The NESTEM code has been under develop- number 7 to number 25 counting from left to right a special
ment at the NASA Glenn Research Center for over 15 years. NESTEM notation was used to specify if the random variable name refers to
uses deterministic analyses together with probabilistic methods to the bottom or the side of the tank. The random variable name with
quantify the probability of failure of structural components which a “1” after it (for example “thickness of insulation1”) means that it
are subjected to complex mechanical and thermal loading. The refers to the side of the tank whereas if there is no number after the
NESTEM code was developed to perform probabilistic analyses of random variable name it refers to bottom of the tank. No variables
structures subjected to either steady state or random thermal and pertaining to the top or roof are presented in the figures since they
mechanical loads. Probabilistic methods are becoming more and do not significantly contribute to the stress.
more useful due to the salient features of consistency, reliability From Fig. 3, we note that the coefficient of expansion of the inner
and economy. tank has the most influence on stress. The inner temperature,
modulus of elasticity of bottom and side of inner tank, coefficients of
3. Results and discussion expansion of the side and bottom of the inner tank, Poisson’s ratio of
the bottom of inner tank, thermal conductivity of insulation, and
The problem is solved iteratively by using a scattered set of values inside convection coefficient also influence the maximum stress. The
obtained by varying the mean variables by of the LNG storage tank somewhat arbitrary choice of the outside convection coefficient (10
10%. In the current work, a probabilistic analysis has been per- W/m2) can be justified in terms of the small influence it has on stress.
formed for the maximum von Mises stress. The outside ambient Also the outside temperature choice can be justified in that it has
temperature boundary condition is set to 35 C degrees and the a small bar in the sensitivity factor graphs.
inside LNG temperature is set to 162 C degrees. The inlet pressure The cumulative probabilities of stress in Fig. 4 show the range of
of the tank is set to 689.48 kPa. The tank is fixed at the bottom base. probability value from 0.001 to 0.999. The 50 percent probability
All the random variables are assumed to be independent and represents the stress produced for the case when all of 54 variables
a normal distribution is assumed for all random variables. are at the mean value which is the deterministic case. The stress at
The maximum stress location is determined in the deterministic the 0.001 probability level is 2.97 108 N/m2 and at the 0.999
calculation and this location is used to evaluate the cumulative probability level is 6.90 108 N/m2.
distribution functions and the stresses produced in LNG tank. A
3.2. Boil-off calculation
typical von Mises thermal stress distribution is shown in Fig. 1 and
Fig. 2 depicts the temperature profile in the tank for the mean
The method of calculating the maximum quantity of boil-off gas
random variables. The probabilistic stress analysis is performed at
generated is given. The total heat input to the LNG tank is the sum
the point of maximum stress which occurs at the bottom of the
of the heat input to the roof, sides and bottom and is given by:
tank near the outer edge.
Q T ¼ 249917 J=sec
3.1. FEA results where, QT: total rate of heat input (J/sec); The quantity of boil-off
gas is calculated by
Fig. 3 shows the sensitivity factors for stress for each random
variable obtained from NESTEM and are plotted for a probability q ¼ Q T =hfg
R.S. Reddy Gorla / Applied Thermal Engineering 30 (2010) 2763e2769 2769
temperatures and von Mises stresses for each run. The NESTEM
probabilistic engineering analysis software was then used to
simulate uncertainties in the random variables. Probabilistic
design is a way to formally quantify the effect of uncertainties.
Probabilistic design is necessary because the effect of the variables
on maximum temperature and stress have to be described accu-
rately. In sum, a design can be cost effectively accomplished if the
effects of uncertainties are known.
The rate of boil-off per day is 0.051% by weight found by
calculation in this analysis is within the range found in the litera-
ture for similar sized LNG tanks.
The sensitivity factors versus random variables for the prob-
abilities from 0.001 to 0.999 were found and the longer bars in
the plots indicated variables with a large impact on stress. All the
variables have at least some effect on the von Mises stress
whereas some variables have a high impact, which include the
inner temperature, the coefficients of expansion of the base and
side of the inner tank, inside convection coefficient, height of the
tank, and the moduli of elasticity of the side and the base of the
Fig. 4. Cumulative probability of stress.
inner tanks. Most of the other variables have a much smaller
contribution to the stress. Evaluating the sensitivity factors will
where, q: quantity of boil-off gas (kg/sec); QT: total heat input (J/ enable the identification of the most critical design variables in
day); hfg: latent heat of vaporization (J/kg); order to optimize the design and make it more cost effective. In
order to reduce the thermal stresses, the designer has the
q ¼ 249917=506169 ¼ 0:4937 kg=sec freedom to select materials for the inner tank that will have a low
value of thermal coefficient of expansion and modulus of
The maximum boil-off of gas in weight percent per day is
elasticity.
given by
R ¼ 24ðh=dayÞ3600ðsec=hÞ100 q=m
References
R ¼ 2436001000:4937 kg=sec=84589272 kg [1] H. Lun, F. Fillippone, D. Cobos Roger, M. Poser, Design and construction aspects
of post-tensioned LNG storage tank in Europe and Australasia, New Zealand
Concrete Industry Conference, September 2006.
R ¼ 0:051% wt=day
[2] Q.-S. Chen, J. Wegrezyn, V. Prasad, Analysis of temperature and pressure
changes in Liquefied natural gas (LNG) Cryogenic tanks, Cryogenics 44 (2004).
where, R: maximum boil-off gas rate in percent (wt/day); m: mass [3] S.-J. Jeon, B.-M. Jin, Y.-J. Kim, Consistent thermal analysis procedure of LNG
of LNG (kg); q: quantity of boil-off gas (kg/sec). storage Tank, Structural Engineering and Mechanics 25 (4) (2007).
Yang [8] examines a similar size tank to the one in this work [4] B.T. Oh, S.H. Hong, Y.M. Yang, I.S. Yoon, Y.K. Kim, The Development of KOGAS
Membrane for LNG Storage Tank, in: Proceedings of the Thirteenth Interna-
(200000 m3). He gives a design boil off per day of 0.05% wt/day
tional Offshore and Polar Engineering Conference, Honolulu, Hawaii May
which is close to the boil-off found in this study which is 0.051% 25e30 (2003).
wt/day. [5] S.J. Jeon, C.H. Chung, B.M. Jin, Y.J. Kim, Liquid Tightness Design of LNG Storage
Gorla [15] recently presented an analysis for determining the Tank Incorporating Cryogenic Temperature-Induced Stresses.
[6] M. Graczyk, T. Moan, O. Rognebakke, Probabilistic analysis of characteristic
boil-off calculation for LNG. pressure for LNG tank, Journal of Offshore Mechanics and Arctic Engineering
128 (2005) 128e133.
4. Concluding remarks [7] S.-J. Jeon, E.-s. Park, Toward a Design of Larger Above-ground LNG Tank, LNG
Journal, March/April.
[8] Y. Yang, Development of the world’s largest above ground full containment
A full containment LNG storage tankdwith typical dimension LNG storage tank, in: 23rd World Gas Conference, Amsterdam 2006.
abstracted from several research papersdwas modeled in ALGOR. [9] A. Bashiri, Modeling and Simulation of Rollover in LNG Storage Tanks. Inter-
national Gas Union, 2006.
The novelty of this work is the probabilistic evaluation of the finite [10] A. Salem, R.S.R. Gorla, Probabilistic finite element thermal analysis applied to
element solution for a thermally and mechanically loaded cryo- a water tank design, International Journal of Fluid Mechanics Research 31 (2)
genic fluid containing enclosure. Cumulative distribution functions (2004) 131e142.
[11] B.H. Thacker, D.S. Riha, S.H.K. Fitch, L.J. Huyse, J.B. Pleming, Probabilistic engi-
and sensitivity factors were computed for stresses generated due to
neering analysis using the NESSUS software, Structural Safety 28 (2006) 83e107.
54 random variables in the areas of thermal, material, and struc- [12] R.S.R. Gorla, S.S. Pai, J.J. Rusick, Probabilistic study of fluid structure interac-
tural variables that govern the LNG tank. Additionally, the boil off in tion, International Journal of Engineering Science 41 (2003) 271e282.
[13] R.S.R. Gorla, N.R. Gorla, Probabilistic finite element analysis in fluid mechanics,
weight percent of the tank per day was calculated at the mean of
International Journal of Numerical Methods for Heat & Fluid Flow 13 (7)
the random variables. (2003) 849e861.
One aim of this work was to predict the uncertainty in the [14] R.S.R. Gorla, O. Haddad, Finite element heat transfer and structural analysis of
stresses of the LNG tank under non-ideal conditions due to vari- a cone-cylinder pressure vessel, International Journal of Applied Mechanics
and Engineering 12 (2007) 951e963.
ation in the random variables. The first step was to perform a finite [15] R.S.R. Gorla, Rapid calculation procedure to determine the pressurizing period
element analysis using ALGOR to determine the maximum for stored cryogenic fluids, Applied Thermal Engineering 30 (2010) 1997e2002.