To predict the consequences of a gas explosion in an industrial environment is not a simple task. Nomograms or simple scaling laws can be useful for interpolation and scaling of experimental data. However, they may give misleading results if they are used outside their range of validity. (British Gas for the Department of Energy, 1990.) Parameters such as geometry (i.e. confinement, size, type of obstacles, geometrical layout), gas type and concentration affect the rate of flame propagation and thereby the explosion pressure. Moen et al. (1981) have shown that simple vent area recommendations may be totally inadequate for enclosures containing obstacles. More advanced tools than nomograms and simple scaling laws have to be applied for simulating gas explosions in industrial environment.
This was also the main conclusion of the report British Gas wrote on behalf of the UK Department of Energy (1990). Explosion venting guidelines, simplified theoretical models and complex numerical codes were reviewed.
Explosion venting guidelines were seen as largely inapplicable because of the small scale of the experimental data on which they are based and because of their inadequate treatment of turbulence generated by leaks and obstacles.
Simplified theoretical models (empirical and approximate theoretical models) could with further development become adequate techniques perhaps.
Numerical models can provide a framework for developing a more general offshore explosion model.
For the last ten years CMR has had a large activity on gas explosion research. Important knowledge has been generated and formalised through the development of numerical tools like FLACS (FLame ACceleration Simulator).
||Experimental results from CMR Raufoss experiments compared with scaling law. |
The FLACS code is a three-dimensional gas explosion and gas dispersion simulation tool. The model takes account of the interaction between the gas flow and complex geometries such as structures, equipment and pipework. The FLACS code produces quantitative information, e.g. in the form of pressure-time curves. By performing sensitivity studies alternative scenarios and layouts can be tested and their explosion hazard potential can be identified. FLACS has been applied in the design of more than 30 offshore platforms and for accident analyses after the West Vanguard and the Piper Alpha accidents. It is being increasingly used also for onshore applications.
According to the British Gas report (1990) the FLACS code "stands alone in being the most developed and validated on general offshore explosion modelling".
The objective of this section is to describe
This section is not intended as a user manual for FLACS and CASD. References to the FLACS and CASD user manuals are Storvik et al. (1990) and Langeland et al. (1988a, b, c).
12.1 Route through a FLACS Simulation
The first version of FLACS, used in the beginning of the 1980's, was a research tool with simple input and output facilities. However, the geometries that were studied became gradually more complex. It was realised that communication between the user and FLACS had to be improved. An advanced user interface to FLACS was developed (based on CAD and computer graphics technology) and given the name CASD (Computer-Aided Scenario Design).
Figure 12.2 shows the connections between CASD and FLACS. CASD generates the scenario definition to FLACS and presents the results from the FLACS simulation.
||The route through a FLACS simulation.|
The first step of a FLACS simulation is to generate the geometry (i.e. geometrical layout of the plant, compartment or offshore module) that is to be investigated. Gas cloud composition, size and location, location of ignition point, and specific output parameters have to be determined before the simulation of the gas explosion can start. The running of FLACS is an extensive numerical task which requires a fairly large computer. In FLACS simulations the three-dimensional Navier-Stokes equations are solved. The FLACS output is presented by the CASD program. Typical output can be time-series plots like pressure-, impulse-, and drag-time plots as well as coloured shaded-image contour representations of velocities, flame location, pressure etc. Three-dimensional animations of the explosion development can also be generated.
12.2 Geometrical Layout
A realistic representation of the layout of an industrial facility for a FLACS simulation, requires a fairly high degree of detail. In offshore modules objects with a size from 0.3 m and upwards will typically be included. In areas with high density of smaller objects, these smaller objects may also have to be taken into consideration.
Geometrical layout such as equipment, piping, walls etc. in the simulated geometries are represented as cylinders and boxes which are aligned with the main axes of the module. Pipes are represented as long cylinders. Beams which are not vertical or horizontal are represented by vertical or horizontal beams with blockage similar to the original beams. Figure 12.3 shows a line drawing of an input geometry for a FLACS simulation.
||Line drawing of an input geometry for FLACS simulation.|
Walls are represented by boxes with zero width in one direction. Porosity for walls and decks is a value between 0.0 and 1.0, defining the fraction of the area available for flow. A solid wall has a porosity of 0.0. Louvered walls have a porosity equal to the fraction of the area available for flow. The walls and decks can be modelled in four different ways:
This is an unyielding wall which is fully closed.
This is used for louvered walls and grated decks which are unyielding but partly open to flow.
3. Blow out panels / Explosion relief panels:
This panel is initially represented by a closed wall which opens up when the simulated explosion pressure reaches a specified value. The opening time of the panels is estimated from the pressure load. The opening pressure, maximum travel distance, weight of the panel and final porosity (after opening) can be specified by the user.
These are open areas which do not offer any resistance to either flow or pressure, except for the modelled beams and main structure.
Due to limitations in processing speed and memory capacity of today's computers, the control volume in FLACS-simulations is in full scale one cubic meter. In many industrial geometries, flame acceleration may be generated in areas where the geometrical details are too small to be resolved on the numerical grid. The geometrical details in these areas are represented by porosities and empirical formulas, depending upon obstacle type and shape which describe momentum loss and turbulence generation.
||The FLACS code uses porosity, resistance and turbulence generation parameters on a numerical grid to approximate the geometry. The numerical grid divides the simulation volume into control volumes. The calculations of the explosion parameters are done for each of these control volumes.|
As shown in Figure 12.4, one large obstacle may cover a number of control volumes in the calculation domain. FLACS can also calculate the contribution of a number of small obstacles or parts of an object to the porosity parameters for a single control volume. Verification tests for FLACS with control volume size corresponding to one cubic metre in full scale show good agreement with experiments in scaled-down (1:5 and 1:33) typical offshore modules.
12.3 Explosion Scenario
In explosion simulations using FLACS the following explosion scenario parameters may be investigated:
- Size and fuel concentration of the combustible cloud
- Type of fuel
- Location of ignition point
One can assume that a homogeneous stoichiometric cloud covering the whole volume is a worst case situation. The probability for this situation to occur must, however, be considered for each given scenario. Some data exist which may be helpful:
- Experimental results from gas explosions with homogeneous and nonhomogeneous clouds show comparable overpressures.
- Gas dispersion experiments and simulations indicate that large, high-momentum leaks in semiconfined areas will, shortly after the initiation of the leak, result in effectively uniform, flammable concentration within most of the interior volume.
- The ignition probability is largest close to stoichiometry, since the minimum ignition energy is at its lowest for this concentration.
- Nonhomogeneous gas clouds with concentrations in the flammable range may have lower ignition probability than a similar homogeneous stoichiometric cloud. However, the effects of a gas explosion might be equally severe for the two cases.
The FLACS89 code has the capability of simulating gas explosions with methane, propane, ethane, propylene, ethylene and hydrogen in air. The capabilities of FLACS to handle methane and propane have been extensively verified. As far as the other fuels are concerned, limited verification has been carried out and results from simulations with these fuels should therefore be used with care. Natural gas is treated as a mixture of methane and propane, where the ethane content of the natural gas is treated as methane and the higher hydrocarbons as propane. Assuming that CO2 behaves as an inert gas, the effect of CO2 will be marginal. Fuels with large quantities of inert gas, such as N2 and CO2 can be handled approximately as lean mixtures.
In most cases the ignition point location is uncertain. It may also be difficult to judge where the worst case location is, some knowledge of gas explosions is usually required. The typical scenario uses the expected worst case location or ignition in the centre of the area. It is also common to test out the sensitivity of moving the ignition.
Previously explosion scenarios have mostly been selected on the basis of worst case scenarios for ignition location and gas cloud. However, we foresee that in the future explosion parameters will be more related to risk analysis, where the postulated accident scenario is evaluated based on frequency of the event and where the simulation of gas explosions accounts for release rates, gas dispersion and most probable ignition location.
12.4 The FLACS Code
FLACS is a fluid dynamic code that calculates explosion pressure and other flow parameters as a function of time and space for different geometries and explosion scenarios. It takes account of the interaction between flame, vent areas and obstacles such as equipment and pipe work. Recent development of FLACS includes the ability to simulate dispersion in complex geometries, both with diffuse and high-momentum leaks, with or without wind.
||Partial differential equations solved in FLACS.|
The FLACS code solves the full gas dynamic partial differential equations for a set of control volumes, as shown in Figure 12.5. The effects of turbulence and chemical reactions are included in the differential equations. The equations are discretised using a finite-volume technique and a weighted upwind/central differencing scheme for the convection terms. Velocities are calculated on staggered grids. The effect of turbulence is included through the eddy-viscosity concept by solving equations for turbulent kinetic energy (k) and its rate of decay (e). Combustion is modelled a flamelet model which consists of a sub model for burning velocity as function of gas mixture, temperature, pressure and turbulence in the reactant. Ignition is modelled by assuming that 50% of the fuel in the control volume in which ignition occurs, is consumed. Thus the temperature is raised and the explosion starts.
12.5 Output from FLACS
A tremendous amount of data is produced when FLACS is solving the pressure, velocity, temperature, density, turbulent parameters and combustion rate in each control volume in time steps of typically 10 msec. All these data cannot be stored during the simulation. Some of the output parameters have therefore to be defined before the FLACS simulation is carried out. These output parameters are typically:
- Number and location of monitor points for pressure, impulse, drag and other parameters
- Location and size of areas for average wall pressure monitoring
- Specification of variables to be presented as field plots (cross-sectional plots)
The pressure-time curves are presented either as local pressure time curves or as average wall pressure curves. Short pressure spikes that may be observed on local pressure time curves will in the average pressure time curves be smoothed out. The average pressure-time may therefore be more relevant for assessment of the average load acting on walls and decks.
Local pressure-time curves
For a number of predetermined locations the local explosion pressure is monitored and presented as individual pressure-time curves. After the main positive pressure pulse, the simulations will then predict a small negative pressure pulse. The magnitude of this negative pulse will depend on the vent arrangement and the geometry.
These plots will then give the maximum explosion pressure in barg at this location along with the duration of the pressure pulse. This is vital information if a dynamic response analysis of the structure is to be performed later on. The curves are well suited for comparing the results of different sensitivity simulations in order to choose the best layout of the area.
||Typical pressure-time curves from two FLACS simulations of gas explosions in a compartment. In case 2 a wall was opened for explosion venting and the pressure was hence reduced. |
Area-averaged wall pressure curves
Area-averaged wall pressure curves can be generated at portions (panels) of the outer walls. For a porous wall or partly open wall, pressure in the open parts will not contribute to average pressure loading. The average pressure for a panel is calculated as the net force (F) acting on the panel divided by the net area (A) of the panel.
The appropriate portion of a wall for which the average pressure should be estimated will depend on the wall structure. Note that no time-averaging is performed, the pressure is still given as a pressure-time curve!
Drag (i.e. dynamic pressure)
Smaller objects such as pipework, cables etc. will mainly be subjected to drag forces due to the explosion wind. The net drag force on an object can be estimated by multiplying the drag (i.e. dynamic pressure) by the front area and the drag coefficient, CD, for the object. The local drag or dynamic pressure is presented in Pascal (Pa) (1 bar is 105 Pascal) and calculated for a number of predetermined locations by use of the following relation:
For these calculations the local density, r, and velocity, u, are used.
In a FLACS simulation the flow velocity vectors in x, y and z direction, i.e. u, v and w are predicted.
Maximum positive pressure impulse
The pressure impulse is the time integral of the local pressure-time curves. The pressure impulse is given in PaS (Pascal-seconds). The maximum positive pressure impulse is at the time when the pressure is ending the first positive pressure phase. The maximum positive pressure impulse is one way of characterising the pressure time curve, which takes the pressure and the duration of the pressure pulse into account. Maximum positive impulse and maximum pressure is often used to estimate structural response.
To visualise the development of the explosion, contour plots are presented. These contour plots can show variables as pressure, combustion products, fuel concentration and velocity vectors in various cross-sections of the module at specific time steps during the simulation.
The contour plots typically consist of five plots and a text header showing the time after ignition. The first plot shows the geometry in the specified cross section. The second and third plots show the fuel and combustion product concentrations. The overpressure is given in the fourth, and velocity curves are given in the fifth plot.
The information in contour plots is mainly used for visualisation of the flow phenomena and local pressure build up during the explosion. They are very useful as verification of scenario parameters, such as cloud size and location of blow-out panels. The contour plot can also be used for production of video animation of the results.
12.6 Benefits from FLACS Simulations
The FLACS code has been evaluated as the most validated code for prediction of gas explosions in offshore modules (British Gas for the Department of Energy, 1990). The FLACS code is based on the latest knowledge within gas explosion research and is most likely providing the highest quality of results currently available.
FLACS provides quantitative information, such as pressure-time curves for a given explosion scenario. The FLACS results can be applied for evaluation of structural response as part of a risk analysis. A joint project CMI/DNV has evaluated the possibilities of transferring FLACS data to the DNV Sesam code for structural response predictions.
By performing sensitivity studies with FLACS, different layouts such as explosion vent arrangements or location and orientation of equipment can be tested out. The best or the most acceptable solution can be established. In this way the FLACS code can be a very practical and useful design tool. For simple geometrical layouts, however, µFlacs is a more efficient tool for sensitivity studies than is FLACS.
The quantitative results in form of pressure-time plots, contour plots and video animation of the results, makes the results easily accessible. It is easy to understand the main results from a FLACS simulation. The code is therefore an effective tool for transferring knowledge about gas explosions to decision makers.
The benefits from FLACS simulations can be summarised as follows:
Safer design and operation through transfer of knowledge and practical results can be achieved.
12.7 Accuracy and Validity
The solution technique described above is generally first-order accurate in time and space. The relevance of using the order concept in determining the accuracy of simulations of flows in very complex geometries which are not completely resolved by the numerics is, however, somewhat questionable. It is hence very important to verify the performance of FLACS both against simple non-reacting flows which are well documented by others and against more complex flows involving flames propagating in obstructed environments.
There are mainly three factors influencing the quality of FLACS simulations:
- The quality and appropriateness of physical and chemical models used
- The accuracy and stability of the numerical schemes used
- The representativity of the geometry and scenario implemented by the user
These factors can be, and have been, addressed in FLACS validation studies. A comparison with shock-tube studies showed that FLACS predicted both shock strength and position well (Bakke, 1986; Bakke and Hjertager, 1986). Furthermore CMR has a large set of data from a wide range of gas explosion experiments studying the effects on flame speed and overpressure of parameters relevant to industrial plants offshore as well as onshore. FLACS has been extensively validated against this data set. More than 2000 experiments have been performed in the following geometries :
- 10 m tube
- 1 and 10 m wedge-shaped vessels
- 3 m cuboid vessel
- Scale 1:33 and 1:5 offshore modules
In addition to these FLACS was validated against experimental data generated at other institutions (British Gas (Catlin et al., 1993), TNO-PML (van Wingerden, 1989) and Shell Research (SOLVEX)).
The effects of varying the following parameters have been studied:
- Scale (1 m and 10 m long explosion vessels)
- Fuel gas type (hydrogen, methane, ethane, ethylene, acetylene, propane, propylene, butane, cyclohexane)
- Fuel gas concentration (between LEL and UEL)
- Fuel gas homogeneity (homogeneous clouds as well as "real" clouds)
- Fuel gas mixtures (realistic process streams)
- Ignition source strength (sparks, planar ignition sources and flame jets)
- Ignition source location
- Explosion vent size
- Explosion vent position
- Obstacle density (# of obstacles)
- Obstacle size
- Obstacle shape (rounded, sharp-edged, grids)
FLACS has been used to simulate a large number of the experiments listed above. The trends observed when varying the parameters were consistently predicted by FLACS. In most cases a certain amount of over- or underprediction can occur. However, the results are generally within 30 to 40 % of the experimental data. In some instances many repetitions of one experimental scenario have been performed, and particularly in the large-scale experiments the scatter is comparable to the figures quoted above (30-40%) for the FLACS simulations.
||Comparison of FLACS93 results and experimental results for a 1:5 compressor module. |
Run 1: central ignition lower deck (methane)
Run 2: central ignition upper deck (methane)
Run 3: central ignition upper deck (propane)
Run 4: central ignition lower deck (propane)
(Storvik og Hansen, 1993)
The FLACS code combustion model is based on a quasi-laminar formulation and a turbulent combustion concept. The FLACS combustion model does not account for Taylor-type instabilities, nor does it predict transition to detonation and propagation of detonation waves. Even though the FLACS code cannot predict transition to detonation, the result from FLACS can give an indication of whether a transition to detonation is likely to occur or not (high flame velocities and pressures; see Chapter 6).
The major uncertainty for the use of simulation results lies in the representativity of the parameter ranges used for verification, e.g. is the range of scales studied representative of industrial plants? No full-scale experimental data exist, hence scaling is a matter of some concern when FLACS results are used. However, scaling from 1 m long to 10 m long explosion vessels is handled well and it is reasonable to assume this behaviour to be valid for larger scales. Experiments which will be carried out at full-scale in the near future will show whether this assumption is valid or not (Steel Construction Institute, 1992).
||Scaling with FLACS.|
In spite of the uncertainties involved, a recent review of predictive methods for gas explosions concluded that at present FLACS is the best available tool for pressure prediction (British Gas for the Department of Energy, 1990).
12.8 FLACS Projects
This section illustrates the contents and timing of a typical project using FLACS.
CMR's consultancy service on gas explosions, GexCon, has done a large number of projects using FLACS. The following table shows a schedule for a typical GexCon FLACS project. The project consists of two simulations, one base case simulation and one sensitivity simulation.
The first task in a project is to get drawings and other input data from the client. The following drawings and input data are required, if possible:
- Plan view
- Primary steel work
- Secondary steel work
- Cable trays
- Equipment list
- Venting conditions for the explosion (i.e. cladding design, weight, opening pressure etc.)
A kick-off meeting is arranged, where explosion scenario, layout and required output data are discussed. Implementing the layout geometry for a typical offshore module takes up to one week for an experienced FLACS user. When the geometry has been implemented, a print-out is sent to the client for verification. After verification and possible changes are made, the FLACS simulation is carried out. The results are then approved and sent to the client. Running a sensitivity study, i.e. making minor modifications in geometrical layout, change the gas cloud or ignition location, takes typically from one to three days.
A report will include reference to data used and specify relevant assumptions. Detailed results will appear in an appendix in the form of:
- FLACS pressure-time curves at specified locations.
- FLACS drag (i.e. dynamic pressure / explosion wind) -time curves at specified locations.
- FLACS contour plots showing pressure, combustion products, fuel concentration and velocity vectors in various cross-sections of the module at particular time steps during the simulation.
A video showing the development of the explosion is optional.
Guidelines for FLACS projects:
- Contact your in-house FLACS user or GexCon as early as possible and discuss when and where running a FLACS simulation is advisable.
- Regular meetings are preferable. One kick-off meeting is an absolute minimum.
- Change one parameter at a time when sensitivity studies are performed.
The FLACS code is available through GexCon, CMR's gas explosion consultancy, and is being used in-house by BP, Elf, Esso (Exxon), Mobil, Norsk Hydro and Statoil.
12.9 Running FLACS on the Computer
In this section a sequence of tasks, from preparing input data via simulation to presenting results, is outlined. Following this sequence reduces the possibility for inconsistencies in the input data and partitions the work into manageable tasks. It is important that written or plotted documentation is produced following each task and before starting the next one. The sequence could be:
- State your problem
- Define possible parameter variations
- Define and verify the geometry
- Define and verify the grid
- Define and verify the scenario
- Check for inconsistencies in tasks 2, 3 and 4
- Calculate and verify the porosities
- Run the simulation
- Check the simulation log file for errors
- Present the results
- Store all data on tape for later use
If FLACS produces unexpected results it may be that some of the input data are incorrect or inconsistent. Below a check-list for pitfalls is presented:
Avoid large Courant numbers
Locate ignition in an unblocked volume
Locate monitors in unblocked volumes
Define realistic leakage parameters
Make sure vent areas are correct
Make sure gas composition is correct
Avoid strong transient wind build-up
Check disk space and access rights