You are on page 1of 9

Gas-freeing of crude oil tanks analysis using numerical simulation

Gas-freeing of crude oil tanks analysis using numerical simulation


K Chow, PhD, SIMarEST, AMRAeS, Narvik University College, Narvik, Norway

Marine gas-freeing is a seldom-examined operation that takes place on board oil tankers ahead of some cargo transitions and tank maintenance/inspection cycles. Since the introduction of inert gas systems (IGS), tanker explosions have been minimised, but the issue of toxic vapours inside the tank still exists. Another issue relating to gas-freeing concerns the actual time required to ensure gas-free conditions; there are no regulatory considerations for the minimum times for gas-freeing, which stands in contrast to regulations concerning other areas on board a vessel. Through the use of modern computational methods, gas-freeing processes have been modelled and simulated in order to determine the progression of gas-freeing with regards to examining the concentration and distribution of vapours inside the tank during the procedure, as well as studying the decrease in total average concentration with regards to forming a simple qualitative model to predict the time required to gas-free given certain initial conditions.

AUTHOR'S BIOGRAPHY Kevin Chow is a computational fluid dynamics researcher at Narvik University College, and has performed research into internal fluid flows during gas-freeing processes. He has recently concluded a doctoral study on this topic, having completed a Bachelors degree in Aerospace Engineering in 2004. INTRODUCTION

as-freeing operations have been performed for many years in order to sanitise a tank and make it possible to support life for different situations, such as inspection, maintenance, and to load cargoes. Due to the concentrations and types of gases involved in gas-freeing, such an operation is always associated with a high degree of danger. A number of tanker explosions in the past (Kong Haakon VII and Marpessa being the most infamous) underlined the difficulties involved in gas-freeing/tank cleaning prior to the introduction of IGS. Such systems are pivotal to crew and ship safety by removing the oxidising agent, rendering ignition impossible. However, even today there are still a number of tanker explosions during venting or purging due to improper usage or non-functioning inerting systems.1, 2 Although an explosion is the most prominent example of the inherent dangers during

tank cleaning, the fumes from the inert gas and the volatile organic compounds (VOCs) are also highly dangerous to any crew members exposed to them. In the very least, the inert gases will displace oxygen, quickly asphyxiating crew members unless they are moved to more oxygen-rich surroundings; the VOCs themselves will cause a variety of symptoms depending on their concentration, and can quickly cause life-threatening injuries. To date, there has been very little work concerning the gas-freeing process on board oil tankers. Savage3 prepared a useful overview of the gas-freeing process, including the importance of the marine chemist to the overall process. Logan & Drinkwater4 provided a detailed examination of the gas-freeing procedure, as well as exploring two methods of tank ventilation; mixing and displacement ventilation. Whilst displacement ventilation methods are more used for tank cleaning (ie, removal of liquid oily residue), mixing/dilution ventilation methods are employed for gas-freeing. A number of actual cases were provided, and the approximate gas concentrations (albeit of the evacuated air) recorded in order to present the progression of the tank venting process. As well as the information presented, it is interesting to note that although crude oil tank (COT) volumes examined in their work were small by today's standards (<10 000m3), the fans

No. A17 2010

Journal of Marine Engineering and Technology

Gas-freeing of crude oil tanks analysis using numerical simulation

used were of a high flow rate, so they were able to achieve much larger air change rates (ACR). More detailed information concerning one such process was examined by Oldervik et al, in support of their research into VOC emission control systems.5 They recorded gas concentrations of the evacuated air with respect to time, including the concentration of each gas species in time. Numerical methods have also been used in ventilation scenarios; for example, CFD has been used in conjunction with a fully 3-dimensional model to study the time required to inert and then re-oxygenate a typical ballast space on board a ULCC,6 suggesting that only two air changes are required for each cycle.

TYPICAL GAS-FREEING PROCEDURES


A typical gas-freeing process begins with an inerting cycle which removes the oxidant from the tank atmosphere. This reduces the oxygen content from 20% down to less than 8%. After the inerting cycle is complete, a number of operations may occur prior to gas-freeing, such as crude oil washing (COW), where oil is sprayed in a pre-defined pattern to remove emulsified oily residue from the superstructure. One side effect of COW is that the content of VOCs is increased due to the oil spray. Prior to the gas-freeing process starting proper, a purge cycle is carried out. Here, the inert gas fans are run, exchanging air inside the tank until VOC concentrations are below 2%. Gas-freeing starts after the purge cycle has completed. Typically, a fixed or portable gas-freeing fan (air or water driven) is mounted and secured on deck at one of the open hatches (typically 0.318m dia). The fan is used to supply air at high velocity into the tank, displacing both the inert gas and any remaining VOC vapours until a termination criteria is met, which is usually a function of the lower flammability limits (LFL); for example7: VOCs reduced to 40% of the LFL for the reception of cargo; VOCs reduced to 1% of the LFL for entry into the tank. The concentrations of various VOCs are determined experimentally by a marine chemist using a dipping probe, and the tank atmosphere is examined at a number of locations and heights. If the concentrations are found to be outside of the required criteria, gas-freeing is resumed until concentrations are inside allowable limits.

Prosser10 conducted experiments involving mixing tetra ethyl lead in varying tank sizes from 75m3 to 15 000m3. One aspect of their work showed that if the jet is of insufficient Reynolds number, it is unable to uniformly mix the entire contents of the tank, leading to stratification of the components. In more general ventilation scenarios, experimental analysis of residence time distributions of tracer gases used to visualise ventilation flows allows the formation of a generalised relative concentration expression. This can be used to empirically model the reduction of contaminants in a room. For example, Liao & Liang11 performed experiments measuring CO2 concentrations inside an environmental chamber, and then used a linear response method to model the tracer dynamics. This allowed the formation of mathematical expressions to model the relative change of concentration in time.

Scaling parameters
In order to define the flow field, and characteristics of the flow using non-dimensional quantities (eg, for scaling or comparison), a number of non-dimensional parameters exist which can be used here. The Reynolds number is a measure of the strength of the flow, in particular, the level of turbulence:

Re =

ul

(1)

Higher Reynolds numbers indicate higher levels of turbulence. If Re < 10 000, the flow transitions to laminar flow. In gas-freeing, the range of Reynolds numbers is expected to vary by several orders of magnitude. At the fan nozzle of the supply jet, Re ~= 1x106, whereas in areas deep in the tank that are obscured from the convective jet, the Reynolds number will be substantially lower, possibly below 50 000. Due to the presence of multiple species inside the tank, each with their own different densities, it is possible for the flow-field to stratify if not subject to external forces. Any stratification (or indeed the tendency to stratify) can be quantified using the Richardson number:

Ri =

g (T2 T1 ) g ( 2 1 ) g 'h or g ' = , where g ' = (2) 2 T u

Essentially, the Richardson number is a ratio of inertial forces to the gravitational potential energy of the fluids. Ri >1 indicate that the stratification is stable, whereas Ri <<1 results in unstable overturning motions until a stable configuration is found, or the contents mix. Analytical studies have suggested that stratified layers may be stable in Richardson numbers as low as 0.25.12

Industrial ventilation
Although there is a dearth of experimental information on the marine gas-freeing process itself, there is a much wider dissemination of industrial ventilation research. Both civil and chemical industries have studied mixing ventilation, using both experimental and computational methods. Chemical industries make extensive use of mixing processes, where the primary aim is to mix one or more species in as little time as possible. This is usually achieved by optimising the flow field to reduce zones of stagnant air,8 or by increasing the entrainment or shearing of the mixing constituents.9 In early work, Fossett &

NUMERICAL METHODOLOGY
Basis
There are a large number of difficulties surrounding the examination of gas-freeing. Part of the difficulty is in the large degree of uncertainty surrounding the initial state of the tank a priori; the concentration of gases inside the tank is dependent on the crude carried, how long it has been carried (ie, the time given to evaporate) and the tank pressures and temperatures during the voyage.

Journal of Marine Engineering and Technology

No. A17 2010

Gas-freeing of crude oil tanks analysis using numerical simulation

Equipping a cargo tank with instrumentation which does not affect the gas-freeing process, or cause a potentially dangerous situation is still hugely expensive in terms of charter, labour and equipment costs. If representative non-explosive gases are to be used instead, huge quantities of such gases would be required. One alternative method is to use a scaled model; however, depending on the scenario examined, reconciling all pertinent scaling parameters to ensure flow similarity becomes incredibly difficult to the point where such experiments can only model isolated phenomena in a piece-wise fashion, instead of in unison. Whilst not totally replacing experimental approaches, numerical methods such as computational fluid dynamics (CFD) allow the simulation of many different scenarios without the compromises required for scaled models or the expense in instrumentation and ship charter required for a full scale experiment. To ensure that the results produced by numerical predictions are usable, numerical simulations require close validation with other experiments to ensure that the predictions are accurate.

Computational overview
CFD involves taking a problem domain and breaking it down into many small cells, whereby the governing equations of fluid flow (the Navier-Stokes equations) are solved in each cell. By iterating the entire flow field a number of times, a solution can eventually be found. To capture the chaotic nature of high-Reynolds-number flows, turbulence models are employed to model the effects of turbulence on the flow without requiring the mesh to be overly fine. To account for separate species of gas, additional transport equations can be solved for each gas. The work presented in this paper was conducted using a commercial CFD code (Fluent 6.2.1613). Numerical discretisation used a third order scheme14 allied to a hexahedral mesh in order to ensure that numerical diffusion was minimised, so that the spreading and diffusion of different fluid species were not over-predicted.

Group reformulation of the original K-Epsilon model (referred to as the RNG model), K- , and Realizable K-Epsilon model. More complicated models exist, such as Reynolds stress models16 (RSM), which is another RANS-based turbulence model, but has seven additional equations; one for each of the Reynolds stress terms and the dissipation rate. The behaviour of the model and the number of extra equations increases the time taken to calculate a flow solution. A method called Large Eddy Simulation (LES) also exists, but requires an extremely fine mesh in order to directly simulate the effects not modelled in the turbulence model, making LES the most computationally-intensive turbulence modelling technique summarised here. Due to the large sizes of COTs, using a LES method would require an extremely dense mesh due to the high Reynolds numbers exhibited in parts of the flow field, possibly with cells as small as a centimetre in length. Whilst Reynolds-stress models do not require a fine mesh, the increased number of equations causes additional load, and the interaction of the RSM equations with the existing RANS equations makes the turbulence model susceptible to instability, especially at larger time-steps, reducing the speed of simulation progression. For example, a typical time-step for a K-Epsilon simulation could be as high as 0.2s to 0.5s, but with RSM, the largest time-step before the onset of instability may only be around 0.05s an order of magnitude lower. For these reasons, the Realizable K-Epsilon model was used for the majority of the simulations. A single simulation using the Reynolds stress model was carried out for calibration of the presented analytical model.

BOUNDARY CONDITIONS
Geometry and scaling
As the design of COTs vary from shipbuilder to shipbuilder, a generic tank design was adopted, which incorporates the majority of structural features present in modern doublehulled tanks. The geometry concerned a wing tank with a total volume of 23 500m3, with floor dimensions of 40x20m, a depth of 30m, and a 5m chamfer on the outboard side of the tank. A number of structural frames are also modelled to support the deck structure, but the inboard wall and floor area are only joined by half-width frames. Four vents are located at deck level. In order to create a computational model, the geometric model must be translated into a mesh. However, gas-freeing is a time-consuming process, and at the same time the sheer size of the tank is such that simulating a full three-dimensional model using a high fidelity turbulence model will make the problem intractable. To ensure that the computational model can model a gas-freeing scenario within a reasonable amount of time on modern computers, the domain was re-cast in a two-dimensional axis-symmetric manner; this was chosen over a simpler planar 2D representation due to the need to maintain similar volume and area ratios to the original 3D representation:

Turbulence modelling
Due to the complexity and pseudo-random nature of fluid turbulence, a large number of turbulence models are available, each particular model having its own advantages and disadvantages. One major category of turbulence models is the Reynolds-averaged type, which use the ReynoldsAveraged Navier-Stokes (RANS, ie, time-averaged) equations as a basis for the turbulence model. A result of Reynolds averaging is the appearance of the symmetric Reynolds stress tensor which contains six extra terms. In order to resolve these additional terms, the physical effects of turbulence are modelled as an increase in effective eddy viscosity. It is through this mechanism which the turbulence model acts. A large number of RANS turbulence models exist, but the most common are the two-equation type of models due to their low computational overhead and wide applicability. One of the first of these was the K-Epsilon model by Launder & Spalding,15 which solves two equations to calculate the turbulence kinetic energy k, and the rate of turbulence dissipation, . Other two-equation models exist, such as Re-Normalized

No. A17 2010

Journal of Marine Engineering and Technology

Gas-freeing of crude oil tanks analysis using numerical simulation

Fig 1: Geometry of 3D crude oil tank (left); Axisymmetric representation of the 2D model results in a cylindrical tank versus the original oblong shape of COTs (right)

Re2 D Re3 D ;

Ri2 D Ri3 D

(3)

Ajet Ajet Q jet Q jet (4) and VCOT 2 D VCOT 3 D Aplan 2 D Aplan 3 D
As a result of the 3D to 2D translation, the geometric equivalent of the computational model resembles a cylindrical vessel with the jet supply coming into the tank from the centre of the ceiling. The outlet is formed by a narrow ring with a 0.1m gap near the far edge of the tank ceiling between two upper frames (see Fig 1), which run concentrically. As the 2D model still lies on a rectangular plane, discretisation of the mesh can be created using only quadrilateral cells. To give a better prediction, cell density is increased in areas of higher convection (ie, around the supply jet and exhaust ports) and near-wall regions. The mesh is shown in Fig 2. The 2D mesh contains around 17 300 cells, and has overall dimensions of 15.8m wide and 30m tall; the first frame is at 1.75m from the line of symmetry, with the subsequent three frames pitched 3.5m apart.

The basis of the mesh used was derived by a plane cut through a 3D mesh (geometry shown in Fig 1) containing 3.24 million cells, which was determined through mesh sensitivity tests that compared results with mathematical models of jet flows. The protracted iteration time for the 3D model necessitated the derivation of the 2D model here.17

Initial conditions
Vapours evaporated from the remaining oil comprise a number of different hydrocarbon chains, from methane to pentane and hexane, as well as their isomers. Therefore, any particular crude can have a very large number of substrate species, and the amount of each gas evaporated depends on the oil blend. Light crudes will evaporate lighter gases (methane and ethane) whereas heavy crudes will evaporate heavier gases (pentane and hexane). The time, temperature and pressure of the tank during transport dictates the gas evolution from the oil; amounts of gas may also be liberated through pressure-relief valves during transport. Finally, a COW cycle will also liberate more gas. All of these variables make it virtually impossible to predict the tank contents a priori. Due to this wide variation in vapour content and concentration, the simulations were configured as multi-component models using a single gas (butane) as a representative VOC vapour, as its properties are roughly median with respect to the other different gases. Using a single VOC vapour also makes the definition of initial concentrations easier to define; instead of having to guess the concentration of multiple species, the amount of butane can be assumed to be a set value. If it is assumed that a correct purge was carried out, then the amount of VOC should be around 2%. If purging was not carried out, then the content can be much higher; prior to beginning their experiments, the total VOC emissions found by Oldervik5 were as high as 14%. In the simulations (listed in Table 1) this vapour content was assumed to be stratified at the base of the tank, displacing all of the air, forming a thick layer of gas. For the 14%/vol case, this amounted to a layer just over 4.5m thick deposited at the bottom of the tank. The remainder of the gas comprised of 2% oxygen, 15% carbon dioxide and the remainder nitrogen.

Fig 2:The equivalent computational mesh for the 2D axis-symmetric domain

Journal of Marine Engineering and Technology

No. A17 2010

Gas-freeing of crude oil tanks analysis using numerical simulation


Simulation 3B 3B (RSM) 3C 3D 3F 3H 3I 3J Flow Rate (m3/hr) 14 290 14 290 7 110 21 440 14 290 14 290 14 290 71 660 Mean Jet Vel (m/s) 50 50 32.5 75 50 50 50 25.3 Notes 14%/vol Butane 14%/vol Butane; Reynolds Stress Model 14%/vol Butane 14%/vol Butane 4%/vol Butane 2%/vol Butane; Initially fully mixed 4%/vol Butane; 2m Baffles 4%/vol Butane; 1m diameter jet entry

Table 1: List of simulations and initial boundary conditions Inlet velocity and turbulence profiles were based on the quantities downstream of a fan. As opposed to a pure jet, fan profiles have a higher velocity at the periphery due to the fan impeller having the highest tangential velocity at the tips, and the lowest velocities in the centre due to the motor hub. The basic inlet profiles are shown in Fig 3; these were normalised and scaled as required based on the overall volume flow rate.
Inlet Velocity and Turbulence Boundary Conditions
2.5
Axial Velocity (1st Axis)

6000

Vel ratio (u/U0) or TKE/10

Turbulent Dissipation (2nd Axis)

4000

1.5
3000

1
2000

Rate of Turbulent Dissipation

Turbulent Kinetic Energy / 10 (1st Axis)

5000

that no purging was performed and an amount of time has passed to allow full stratification of the gas). Each simulation used a different flow rate to examine the effect of using different fans upon the gas-freeing time. It can be seen that the flow rate directly affects the time required to reduce the concentration of butane inside the tank. With simulation 3C, it can be seen that the combination of lower flow rate and ducting has a substantial effect on the gas concentration graph (Fig 4), as well as creating a period of almost two hours at the beginning of the operation where no reduction occurs. With the other two simulations, 3B and 3D, it can be seen that the higher flow rate is quite well represented by the concentration curve, with concentrations decreasing around 1.5 times faster.
Comparison of Jet Velocity on Gas Removal
9000 3B - 50m/s Jet 8000 3C - 32.5m/s Jet, 15m Duct 3D - 75m/s Jet 7000 6000 3B - Layer Erosion (2nd Axis) 3D - Layer Erosion (2nd Axis) 4 3C - Layer Erosion (2nd Axis) 5 6

0.5

1000 0

0 0.00

0.20

0.40

0.60

0.80

1.00
Butane (kg)

Radial Distance (y/r)

5000 3 4000 3000 2000 2

Fig 3: Inlet boundary conditions for velocity, turbulence kinetic energy (m2/s2) and rate of turbulent dissipation (m2/s3) The walls of the domain are modelled as no-slip, isothermal boundaries with wall log laws prescribed to model the near-wall flow regions. The domain and the walls were considered isothermal, as previous work17 suggested that modern double-hull architecture, as well as the effects of thermal radiation, resulted in the tank exhibiting low variations in temperature range. This also allowed the usage of constant (as opposed to temperature-dependent) materials properties. Simulation 3Cs geometric model also included the usage of a 15m long duct which fed vertically down into the tank, to simulate the use of ducting. Although the inlet boundary condition was the same as the other simulations, the flow was channelled through a contraction to a 0.278m dia straight channel which was fed 15m into the tank. Due to the difficulty in kinematically describing the behaviour of a flexible duct whilst under pressure, it was modelled as a no-slip wall, allowing the flow to develop as it passed through the modelled duct.

1 1000 0 0.0 1.0 2.0 3.0 4.0 5.0 Time (hrs) 6.0 7.0 8.0 9.0 0 10.0

Fig 4: Plot of three total gas decay curves with respect to time at same initial conditions but with different fan supply rates; stratified layer decay plotted on secondary axis. Initial gas volume is 14% The second group of simulations (3F, 3H and 3J, shown in Fig 5) were run with a lower initial condition of 24% by volume with other variations. Simulations 3F and 3J started with initially stratified conditions (with 4%/vol of stratified vapour), whereas 3H begins with 2% of vapour uniformly distributed inside the tank to represent a fully mixed scenario. It can be seen in Fig 5 that although the initial gas amounts are reduced, the total gas graphs show similar characteristics to the previous simulations seen in Fig 4. Simulations 3F and 3J both have an initial period of no activity which was shown very clearly previously in simulation 3C. Here, the period of inactivity is much shorter, under 6 min for simulation 3J (Q = 71 660 m3/hr) and just over 12 min for 3F, which has the same flow-rate as simulation 3B. After this period of inactivity, the concentrations both begin to decrease.

RESULTS
An initial group of simulations, 3B, 3C and 3D were run assuming very high initial gas concentrations 14%/vol of butane totally stratified in the base of the tank (ie, assuming

No. A17 2010

Journal of Marine Engineering and Technology

Layer Thickness (m)

Gas-freeing of crude oil tanks analysis using numerical simulation


Total Gas Concentration - Gas Freeing Rate, Reduced Initial Concentration
1200 3F KE, 14,290 m3/h 1000
Mass Sum of Butane (kg)

Butane Mass in Floor Web Spaces


1000

Web Space 1 (Centre) Web Space 2

3H KE, 14,290 m3/h (Mixed) 3J KE, 71,660 m3/h, High Flow Rate

800

Web Space 3 Web Space 4

800
Mass (kg)

Web Space 5 (Outboard) 600 Total Gas In Tank Simulation 3F (No Obstructions) 400

600

400

200

200

0 0.00

0
0.50 1.00 1.50 2.00 2.50 Time (hrs) 3.00 3.50 4.00 4.50 5.00

0.5

1.5

2.5

3.5

4.5

Air Changes

Fig 5: Gas decay curves for 24% initial VOC concentration Simulation 3H, which began in a fully mixed state, does not have this initial period of inactivity; the concentration decreases immediately from the beginning of the operation.

Fig 6: Comparison of Simulation 3I (2m high baffles) to Simulation 3F; ventilation of individual web spaces is also shown
Maximum Concentration of CO2 during Gas-Freeing
0.2 0.18 0.16
Mass Fraction of CO2

0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 0.5 1 1.5 2 2.5 Air Changes 3 3.5 4 4.5 5 3F, 14,290 m3/h, 4% 3H, 14,300 m3/h, 2% Mixed 3I, 14,300 m3/h, 4% Baffled 3J, 71,660 m3/h, 4%

Effect of obstructions
Although modern double-hulled tankers are virtually devoid of tall structures that span the full width of the tank at the floor level, this structural arrangement can still arise with some FPSOs. With this in mind, simulation 3I has been run with the same initial conditions as 3F but with a series of 2m high rigid frames on the floor, and butane gas stratified between them. Due to the axisymmetric nature of the domain, the rigid frames manifest as a series of concentric rings (Fig 2) with the inboard spaces containing less volume than the outboard areas. Fig 6 shows several plots of the progression of gas-freeing. Compared to the control plot (simulation 3F), it can be seen that the addition of floor frames retards the rate of gas-freeing. Five additional plots are shown in Fig 6 which corresponds to the butane gas decay in the volume bounded by the obstructing frames. These show that web spaces 1 to 3 are ventilated rapidly (due to their small volumes), but spaces 4 and 5 take much longer, with space 5 still containing gas after 3.5 air changes. Interestingly, when web space 4 is almost ventilated, the decay of gas in space 5 is accelerated. One explanation for this is that the ventilation of spaces 4 and 5 are driven by the lateral spread of supply air after impingement. When space 4 is almost ventilated, the lateral spread of air will not contain as much gas. This increases the concentration gradient between the fresh air and VOC gas in space 5, allowing better entrainment.

Fig 7: Maximum CO2 concentration at any point in the tank during gas-freeing; high concentrations are persistent for the first 1.5 air changes
Minimum Oxygen Concentration during Ventilation
0.2 0.18 0.16
Oxygen Mass Fraction

0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 0.5 1 1.5 2 2.5 Air Changes 3 3.5 4 4.5 5

3F, 14,290 m3/h, 4% 3H, 14,300 m3/h, 2% Mixed 3I, 14,300 m3/h, 4% Baffled 3J, 71,660 m3/h, 4% 3H Average

Fig 8: Plot of absolute minimum O2 concentration in tank during gas-freeing; Simulation 3I takes longer to re-oxygenate due to slow ventilation of web-spaces by which time are below 1%. The plots of minimum oxygen concentration (Fig 8) are similar in that although air with 20% oxygen is supplied, the minimum concentrations in the tank are unaffected until after the first air-change. The addition of 2m tall baffles in the flow obviously disrupts the ability for the supply to re-oxygenate the atmosphere by obstructing the flow and reducing local air change rates. This can be seen by comparing the O2 plots of simulations 3I and 3F. Although 3F begins to re-oxygenise after two air-changes (corresponding to the time taken to entrain and remove the stratified layer), the addition of baffles at the base of the tank restricts the ability of fresh air to ventilate the area, resulting in a much longer time to re-oxygenate.

Variation of atmospheric gases


As well as evaluating the removal of the VOC vapours (in this case, butane), the oxygen and carbon dioxide concentration have also been examined in order to determine habitability of the atmosphere. As well as the average global concentration through time, absolute maximums of CO2 and absolute minimums of O2 at any point in the flow field domain are shown in Fig 7 and Fig 8 respectively. It can be seen that the maximum concentration of CO2 doesnt change immediately. Concentrations are still high during the first air change, but drop away towards two air changes. Past 4.5 air-changes, all of the plots show similar levels of CO2,

Journal of Marine Engineering and Technology

No. A17 2010

Gas-freeing of crude oil tanks analysis using numerical simulation

DERIVATION OF GAS-FREEING TIME MODEL


Examination of parameters
The similar behaviour shown in Fig 4 and Fig 5 indicates that the reduction in total VOC concentration of a given initial distribution occurs in a similar way. Most of the simulations appear to have two distinct regions; An initial period after the commencement of gas-freeing where the reduction of VOC is either low or zero; A decay region, after the initial region, where the VOCs are exhausted. The length of the initial region itself varies depending widely on the initial conditions. The time required to establish a steady flow-field is a possible contributory factor; however, in many cases, especially with large amounts of VOC vapour (Fig 4), it can be seen that this initial region can be as high as an hour or more, which suggests that other mechanisms contribute to this period. In Fig 5 it can be seen that when the flowfield is initially mixed, the initial region is virtually eliminated. This is in contrast to the other two plots which have initial regions of varying lengths between 5 and 12 mins. The major differences between these cases are both the flow rate and the initial VOC distribution; when the flow-field is initially mixed, there is a negligibly small initial region; when stratification exists, the initial region is dependent on the flow rate. Although it may seem intuitive to associate the total erosion of the stratified layer with the end of the initial region, Fig 4 shows that this is not necessarily the case, as the low flow-rate simulation 3C has a protracted initial region time surpassing the time required to erode the stratified layer. This reinforces the contribution of flow-rate to the initial region time. The length of the decay region is also variable, dependant on the initial amount of VOC and the flow rate. When this decay region is separated from the initial region and plotted on a logarithmic scale, the simulations all strongly correspond with an exponentially decaying function of time after this initial period.

uj =

B U ( x x0 ) / d

(8)

Model derivation decay region


As already shown, for a given initial condition, the decay region is strongly dependant on the flow rate. A logarithmic trend allows the relative concentration to be defined using an exponential expression: C (t ) = e t thus: Ln C (t ) (9)

= texp

(10)

is an efficiency factor which is used to denote the effectiveness of the ventilation process. For example, Simulation 3I has a much longer ventilation time due to the presence of the baffles, which can be reflected by modifying the efficiency factor. Additionally, this factor allows the model to be calibrated with a simulation of higher accuracy. Comparisons with the single Reynolds stress model simulation (Fig 9) shows that the behaviour of the K-Epsilon model causes an over-estimation of the diffusion and transport of the VOC gases out of the domain, resulting in a non-conservative prediction of the time to gas free. This anomaly of the K-Epsilon model is fairly well known;20 however, it is possible to calibrate the results using the Reynolds stress prediction. Using = 2/3, it can be seen that the analytical method is far closer to the Reynolds stress simulation without being un-conservative.
Comparison of Analytical and Numerical Gas-Freeing Concentration Curves
1 0.9 0.8
Normalised Concentration

3B (CFD, KE) 3B (CFD, RSM)

0.7 0.6 0.5 0.4 0.3 0.2

3B (Analytical)

Model derivation initial region


The examination of parameters showed that the initial region is dependant on both the stratification strength and air change rate, V/Q;

0.1 0 0 5000 10000 15000 Time (s) 20000 25000 30000

V tir = Rii ; Q
where Rii, is the interface Richardson number defined by: Rii = g ' / u 2 ; j g, the reduced gravity, can be found from the following: g'= g ( 2 1 )

Fig 9: Comparison of K-Epsilon models to RSM model, and calibration of analytical efficiency factor Using this analytical model (Equation 10) with an efficiency of = 2/3, it is possible to predict the time required to gas free a tank, given an initial concentration and the required final end-state. Table 2 presents a series of predicted gas-freeing times for a number of scenarios; the two Initially Stratified, m0 = 4%/vol cases correspond to the previously simulated 3F and 3J cases respectively. One consequence of the = 2/3 calibration is that it affects all of the simulated processes, including the return-to-habitability of the internal environment with regards to the O2 and CO2 content. This suggests that a more suitable number of air changes to ensure habitability is around 6.75, instead of 4.5.

(5)

(6)

(7)

Both g and can be determined experimentally using gas sensors coupled with a dipping probe, or theoretically, based on laboratory experiments on crude evaporation and settling. uj can be determined from experimental methods, manufacturer data sheets, or using an analytical jet model: 18, 19

No. A17 2010

Journal of Marine Engineering and Technology

Gas-freeing of crude oil tanks analysis using numerical simulation


State Fully Mixed Fully Mixed Initially Stratified Initially Stratified Fully Mixed mo 2%/vol 2%/vol 4%/vol 4%/vol 4.758% m(t) 0.72%/vol (40% 0.018%/vol (1% 0.72%/vol (40% 0.72%/vol (40% 0.2% C(t) 0.36 0.009 0.18 0.18 0.042 ACR (1/h) 0.608 0.608 0.608 3.04 4.2 tir (hr) 1.01 0.205 texp (hr) 2.52 11.61 4.23 0.846 1.13 = 2/3 t (hr) 2.52 11.61 5.24 1.05 1.13a

LFL) LFL) LFL) LFL)

Table 2: Example calculation of predicted total gas-freeing times for several initial conditions, assuming
a

Initial data based on data from Logan & Drinkwater (1961).

Range of applicability
Although sufficient work was carried out to allow the generation of a simple model to predict the gas-freeing time, the variation of tank design and gas-freeing procedure makes it difficult to create a more general model to encompass all situations. As shown in Fig 6, the introduction of obstructions causes a reduction in the change of rates of exhaustion of the VOCs. As the empirical model given here uses a power-law curve to describe the decay region, gas-freeing times in any tank with obstructions in stratified zones will be under-predicted. The model also assumes that a single gas-freeing method is used throughout the gas-freeing procedure; the usage of ducting, or alternating the position of the gas-freeing fan mid way through the process, are not accounted for by the model, and are out of the scope of investigation of the initial work here. The accuracy of the model is also difficult to quantify due to the limited points of comparison. The decay region prediction texp was adjusted to fit the more accurate Reynolds Stress Model results, but this is still bounded by the scope of tank design and procedure defined previously. Due to the form of Equation (10), the prediction only functions when the flow rate is sufficient to entrain the VOC gas; therefore, low ACH may result in the inability to entrain heavier gases, resulting in non-conservative predictions. Simulation 3C has the lowest ACH in the simulations examined here, so the prediction of texp is at least conservative to ACH = 0.303. The initial region prediction tir is more difficult to predict, as it is a factor of the initial stratification as well as the flow rate, and is quite sensitive to the entrainment ability of the supply air, becoming un-conservative at low ACH and overpredictive when stratified layers are thin (<1m). This complex relationship makes it difficult to create a modification equation, which would be empirical in nature. However, the results suggest that when a stratified layer exists and a tank contains more than 5% VOC, it is advisable to increase tir by a factor of 2 if ACH < 0.5.17

Returning the internal tank atmosphere to habitable conditions may take 67 air changes, if CO2 levels are required to be <1%. Additionally, the simulations provided enough data for a simple analytical model to be created, approximately predicting the length of the initial region and the gas concentration decay due to gas-freeing, based on a straightforward exponential decay function. Calibration of the model using a Reynolds-stress simulation showed that an efficiency factor of 2/3 should be used to scale both the K-Epsilon model results and the analytical model (ie, = 2/3) in order to prevent un-conservative predictions.

REFERENCES
1. US Coast Guard, 2004. Report on the explosion and sinking of the chemical tanker Bow Mariner in the Atlantic Ocean on 28 February 2004. Marine Safety Office, Hampton Roads, Virginia. 56pp. 2. International Maritime Organization, 2003. Casualty statistics and investigations, FSI 12/4. IMO, London. 62pp. 3. Savage KM. 1976. Marine gas hazards control: cleaning and gas-freeing shipboard tanks. Environmental Research 11(2): 213219. 4. Logan A and Drinkwater JW. 1961. Gas concentrations in the cargo tank of crude oil carriers. Annual Report and Transactions of the Royal Institute of Naval Architects. The Royal Institute of Naval Architects. London. 5. Oldervik O, Neeraas BO, Strom T, Martens OM and Meek-Hansen B. 2000. VOC emission control systems for shuttle tankers and floating storage systems (VOCON) - Final Report, MT23 F00137. Norwegian Marine Technology Institute. Trondheim. 24pp. 6. American Bureau of Shipping. 2004. Guide for inert gas system for ballast tanks. ABS, Houston. 42pp. 7. OCIMF ICS and IAPH. 1996. International safety guide for oil tankers and terminals 4th edition. Witherby & Co, London. 285pp. 8. Jayanti S. 2001. Hydrodynamics of jet mixing in vessels. Journal of Chemical Engineering Science 56:193210. 9. Zughbi HD and Rakib MA. 2004. Mixing in a fluid jet agitated tank: effects of jet angle and elevation and number of jets. Journal of Chemical Engineering Science 59(4): 829842. 10. Fossett H and Prosser LE. 1948. The application of free jets to the mixing of fluids in bulk. Journal of the Institute of Mechanical Engineers 160: 224232.

CONCLUSIONS
Simulations of gas-freeing processes have been numerically simulated using CFD, and a number of interesting aspects have been found: Gas-freeing is strongly dependant on the air changes per hour; Stratified vapours result in an initial delay from the start of gas-freeing to the removal of VOC gas;

10

Journal of Marine Engineering and Technology

No. A17 2010

Gas-freeing of crude oil tanks analysis using numerical simulation

11. Liao CM and Liang HM. 2003. A linear model of the effects of residence time distribution on mixing pattern in a ventilated airspace. Building and Environment 38(1): 1121. 12. Miles JW. 1961. On the stability of heterogeneous shear flows. Journal of Fluid Mechanics 10: 496508. 13. Fluent Inc. 2005. Fluent 6.2 Users Guide, Lebanon, New Hampshire. 2216pp. 14. van Leer B. 1978. Towards the ultimate conservative difference scheme. V. A second-order sequel to Godunov's Method. Journal of Computational Physics 32(1): 229248. 15. Launder BE and Spalding BD. 1974. The numerical computation of turbulent flows. Computer Methods in Applied Mechanics and Engineering 3(2): 269289. 16. Launder BE, Reece GJ and Rodi W. 1975. Progress in the development of a Reynolds-stress turbulence closure. Journal of Fluid Mechanics 68: 537566 17. Chow K. 2009. Simulation and analysis of gas freeing of oil tanks. Ph. D Thesis, University of Coventry. 235pp. 18. Shy SS. 1995. Mixing dynamics of jet interaction with a sharp density interface. Experimental Thermal and Fluid Science 10(3): 355369. 19. Pope SB. 2000. Turbulent flows. Cambridge University Press, Cambridge. 770pp. 20. Beeck JPAJ and Benocci C. 2002. Introduction to turbulence modelling. Von Karman Institute for Fluid Dynamics, Rhode St. Genese, Belgium. 550pp.

k l m0 m(t) t u x x0

Turbulence kinetic energy Length scale Initial concentration Concentration variation with time t Time Velocity Downstream distance from jet inlet Jet virtual origin offset; assumed equal to one Stratified layer thickness Dissipation rate of turbulence kinetic energy Ventilation efficiency Fluid dynamic viscosity Air change rate; V / Q Fluid density

Subscripts
2d 3d Jet Plan COT ir exp j 2-dimensional 3-dimensional Jet Plan-area Crude oil tank Conditions at infinity/Free stream Initial region Decay region Jet

NOMENCLATURE
Acronyms
ACR CFD COT COW FPSO IGS LES LFL RANS RNG RSM ULCC VOC Air Change Rate; also Air Changes per Hour Computational Fluid Dynamics Crude Oil Tank Crude Oil Wash Floating Production Storage and Offloading system Inert Gas System Large Eddy Simulation Lower Flammability Limits Reynolds-Averaged Navier-Stokes (Equations) Re-Normalised Group Reynolds Stress Model Ultra Large Crude Carrier Volatile Organic Compound

Appendix
In Table 2, the calculation of the total predicted gasfreeing time can be calculated using equations (5) through (10). The following example can be used as a guide: A particular tank is 20 000m3 and a fan capable of delivering 8000m3/h is being used. The cargo was a light crude and the lighter gas fractions were vented off during purging, leaving a mixture of 4% propane and inert gas (the remaining 96%). Between purging and the beginning of gas-freeing, the remaining gas has stratified into the bottom of the tank with a density difference of 0.8. The stratified layer thickness had been determined to be 0.5m high and from manufacturer diagrams, the jet speed at the layer height is 2.5m/s. Using equation (7), the Richardson number can be calculated by first calculating the relative gravity: g'= g ( 2 1 ) 9.81(1.8 1) = = 7.848 kg/m 3 1

Equation symbols
B C(t) Q T U V Re Ri d g g h Jet spreading constant, determined equal to 6 Concentration variation as a function of time Volumetric flow rate Temperature Jet mean inlet velocity Volume Reynolds number Richardson number Jet diameter Gravitational acceleration Relative gravity Characteristic height or length scale

Rii = g ' / u 2 = 7.848 0.5 / 2.52 = 0.628 ; j This yields an initial region time of: tir = 20, 000 V Rii = 0.628 = 1.57 h 8, 000 Q

The decay time to gas free to 0.72% can thus be calculated from equation (10): texp = Ln 0.72 / 4 (8, 000 / 20, 000) ( 2 / 3) = 6.43h

Therefore the total gas-freeing time is 8 hours.

No. A17 2010

Journal of Marine Engineering and Technology

11

You might also like