Simulation of explosion experiments is needed to assure that the models in the explosion code are representative and correct implemented. They should also give input to extend submodels for areas where no or few data exists for them.
This chapter will summarize/evaluate the FLACS code’s ability to simulate a real explosion. A range of data from explosion experiments, as pressure-time data and time of flame arrival in a range of monitor points are used in the validation of explosion codes.
The pressure - time data from experiments and simulation could be compared directly, but these data are easier presented by three figures:
• peak pressure
• duration of pressure pulse (defined normally from pressures above 10% of peak pressure) • time from ignition to peak pressure
Except for explosions with two, or more peaks, these three figures give a good description of the time - pressure history in a point.
In several experiments the peak pressures exceed the maximum value the pressure transducers can measure. The maximum pressure is then not known. The pressures from simulations are static pressures. In experiments the total pressure is measured. Depending on the location of the pressure transducers used in the experiments, they may also include the dynamic part of the pressure.
The peak pressures from many explosion experiments have an uncertainty of at least 20%. This uncertainty has been observed when experiments has been repeated. Experiments in large scale offshore modules have normally not been repeated, but when they have, a large scatter in the results have been seen. The background for this difference is that it is not possible to have two of these large scale experiments 100% identical. The gas clouds can be inhomogenous or there can be scatter in the stoichiometric ratio, which may make the flame more instable. Wind may influence and give turbulence or the ignition source may give different initial results (if e.g. a chemical igniter is used instead of an electric spark).
The sub models included in the explosion code FLACS do not include models which describe shock/compression ignition, transition to detonation or flame acoustics interactions. The code’s ability to give a representative simulation of gas explosions where one or more of these phenomena are important is therefore limited. The flame acoustics interactions can be important for combustion in nearly empty boxes, like the empty SOLVEX experiments. Shock/ compression ignition can typically occur for pressures above three barg for methane-air mixtures and above one barg for mixtures containing other hydrocarbons. Explosion experiments have been performed in simple test geometries as well as in more realistic geometries. The most simple geometries are presented first, later the more realistic geometries.
8.1.1
Validation of submodels
The explosion code consist of a range of submodels as presented in the previous chapters, like the turbulence models (ongrid and subgrid), the flame model, the burning velocity models and the flame folding models. Most of these models include some uncertain factors which must be determined.
Turbulence and flame models should first be tested with some basic tests, as shown in Chapters 4, 5 and 6. In the evaluation of flame models, the flame’s ability to move according to the specified burning velocity in 1D, 2D and 3D was tested. The ability of the turbulence models to give representative turbulence fields was shown in the presentation of turbulence modelling. All validation and adjustment of explosion models and submodels should be done with variation of grid resolution to secure a code with little grid dependency. Different fuel mixtures should be tested to ensure that the code also can handle different fuels. The validation may include the explosion experiments presented in the next sections, and may typically be divided into six steps as shown below:
• The validation/adjustment of explosion models against explosion experiments should start with free flame propagation from a point. The flame location as function of time should be compared with experimental values, to secure that the ignition and quasi laminar burning velocity models are satisfactory.
• Next step is calculation of explosion experiments done in empty boxes (with one open wall and no obstacles inside) like the Sotra pipe, the empty SOLVEX box (two scales) and the B.G./Mobil box. These tests will verify e.g. that the flame has a representative burning towards wall.
• Simulation of experiments where all obstructions/geometry can be resolved ongrid, like the Sotra radial vessel, some of the MERGE geometries, the TNO experiments and the SOLVEX and B.G./Mobil boxes with obstructions inside. These tests will secure that the generated ongrid turbulence field and the turbulent burning velocity models are representative. Eventually a better extrapolation of the burning velocity model, with respect to higher pressures, turbulent intensity and length scales, can be made. Simulation with different grid resolutions will validate the ongrid flame folding model.
• Simulation of experiments were the obstructions are represented subgrid. The experiments to be simulated can be the same as in the previous step (but on a coarser grid) and the rest of the MERGE geometries. The constants in the subgrid turbulence generation and subgrid flamefolding models will be adjusted in these simulations.
• Simulation of experiments in scaled down versions of real offshore modules, like the M24 on Sotra. This validation secures that real obstructions/geometry are represented well. Simulation with ignition at the end of modules should secure that boundary conditions, as well as the flame is represented satisfactorily.
• The validation should finish with simulation of the experiments in full scale offshore modules, as the SCI module. This simulations will show that the sub models handle the effect of scale (like the extrapolated turbulent burning velocity model) and the inclusion of considerably smaller detailed geometry.
8.1.2
Effect of scale in the experiments
A range of experiments have been done in small scale and the experimental data are tried extrapolated up to full scale. Some parameters are similar, and some are different for different scales.
The turbulent fluctuating field is likely to depend only on the velocity field, but the turbulent length field is proportional with scale. Since the turbulent burning velocity increases with the turbulent length, due to less strain, it will also increase with scale. Based on (7.30), increasing the laminar burning velocity gives the same effect as increasing the scale. Several experimental programs have included this for scaling of experiments. Through oxygen enrichment, the product temperature and burning velocity were increased. A drawback with this method is that also the volume expansion increases. A better solution can be to replace the gas with another gas mixture with higher burning velocity but equivalent volume expansion.
The temperature drop in the burned gas due to heat loss from the gas to the surrounding geometry will decrease with scale, since the heat loss is proportional with the area and the energy content is proportional with the volume. The turbulence production behind round cylinders may however decrease with scale for Re>105, since the wake becomes thinner. The explosion codes were previously, before the SCI experiments, validated against experiments done at medium scale, as the M24 module, with grid size around 0.2 m in simulations. Real offshore modules has a scale four to five times larger. In simulations of them, should the grid size or the number of grid cell needed to represent the module be kept constant if the results of a code is grid dependent? This question shows that the code results must have minimal grid dependency to avoid the problem of scaling