Explosion Hazard Assessment
A wide variety of types of explosion may occur on offshore installations. These include unconfined explosions (overpressure generated by presence of obstacles), confined explosions (overpressure generated through a combination of confinement and obstacles), external explosions (a phenomenon associated with confined, vented explosions), internal explosions (e.g. within a flare stack), physical explosions (e.g. a failing pressure vessel), solid phase explosions (e.g. associated with use of well completion explosives), mist explosions and BLEVE's.
Explosions represent a significant component of the topsides fire and explosion risk on most installations. Over the 25 year period 1973-97 there were 10 significant (> 0.2 bar) explosions on offshore installations in the North Sea, 8 of which occurred in the UK sector, most during the 1980s (Vinnem, 1998). More recent data for the period 1992-99 shows that there have been 10 explosion incidents on UK installations, most internal explosions associated with gas turbines or flare systems.
Large Scale Experimental Data
There is now a considerable body of experimental data from large-scale tests that has greatly improved our understanding of how gas explosions behave in offshore installations. New phenomena have been observed which were not predictable by the models available at the time, for example the significant enhancement of the explosion pressure by the small-scale obstacles. In many cases, however, no in-depth analysis of the results has been undertaken and as a result the opportunity to gain even further insight into the explosion mechanism remains. Data from the large scale test is largely confined to measurements of overpressure and flame arrival time, however more detailed information (e.g. on flame speeds) is required for rigorous model evaluation and development.
Realistic Release Cases
The large-scale tests undertaken in Phase 2 and 3a of the JIP were concerned with quiescent, stoichiometric clouds filling the entire volume of the module. However there remains a question as to whether a realistic release, involving significant turbulence, could give rise to greater explosions overpressures than those observed in the tests. Phase 3b of the JIP addressed realistic release cases and the initial findings from this indicate that in some cases rapid filling of the module could occur with gas at near-stoichiometric concentrations. While lower overpressures generally occurred in the realistic release tests uncertainty still remains concerning whether this is always the case.
For equipment containing volatile liquids at elevated pressure, two-phase or 'mist' releases are possible. There has been limited experimental study of mist explosions and the mechanisms operating in mist explosions remain poorly understood. Because of the difficulty in characterising the initial conditions, their study is also experimentally challenging. In particular there is uncertainty in whether mists for certain drop sizes could generate explosions more violent than that of an equivalent vapour concentration.
The explosion models currently available may be categorised as follows:
- empirical models (e.g. TNO Multi-Energy model, Baker-Strehlow method, Congestion Assessment Method COMEX/NVBANG);
- phenomenological models (e.g. SCOPE and CLICHE); and
- CFD models (e.g. FLACS, EXSIM, AUTOREAGAS, CFX, COBRA and various research codes).
A critical review of modelling and suggestions for future areas of research and development up to 2000 is given in HSE (2004).
The empirical models have a limited range of applicability, cannot deal with complex geometries and have simplified the physics considerably. Nevertheless, these methods are useful for quick order-of-magnitude calculations and for screening of scenarios warranting further investigation with more sophisticated tools.
The phenomenological models are slightly more complex than the empirical models. They have a less limited range of applicability than empirical models, are essentially fits to experimental data and have a lower level of uncertainty than empirical models. They do not attempt to model the actual scenario geometry but instead represent it in a simplified manner, e.g. as boxes connected by corridors. The models are relatively easy to use, with modest computational requirements and therefore are suitable for use where large numbers of calculation runs must be made, as in exceedance curve generation.
CFD models can be divided into two groups; simple and advanced models. The distinction between the two groups, albeit somewhat arbitrary, is that the advanced models will attempt a more complete description of the physical and chemical processes involved, including better representation of the geometry and the accuracy of numerical schemes. To illustrate this difference, consider one of the key findings of Phase 2 of the JIP, i.e. the importance of taking into account small scale obstacles. The simple CFD models make use of porosity/distributed resistance (PDR) models in order to avoid having to resolve the smaller scales obstacles, while the advanced CFD models attempt the resolve the objects using adaptive mesh refinement. Compared to the phenomenological and empirical models, CFD offers the prospect of greater accuracy and flexibility, however computational run times are long and the scope for errors is greater. Particular areas of uncertainty are:
- Representation of geometry (as noted above).
- Modelling of combustion process. Commercial CFD codes use simple correlations derived from experiments without attempting to model the detailed combustion kinetics.
- Modelling of fluid flow. The turbulence model most often used in CFD explosion codes is not strictly applicable to high speed combusting flows.
- Application in overpressure exceedance calculations. This may involve:
- the use of the same CFD code for both dispersion and explosion calculations (each having different modelling requirements);
- Long run times prevent a sufficient number of separate CFD simulations (with associated concerns as to the quality of each simulation) to be carried out to make results statistically meaningful;
- The application of uncertain 'symmetry arguments' or 'physical reasoning' to generate the cloud shapes and overpressure data for all the various leak scenarios of interest.
- Uncertainties associated with the generation of 'equivalent stoichiometric clouds' within some CFD models can offset any gains in accuracy arising from sophisticated gas dispersion modelling and explosion overpressure calculation.
- Validation and verification. There has been a lack of disclosure of validation data for some codes. This issue of auditability is of particular concern where codes are being regularly updated and new versions issued on a regular basis.
Explosion hazard assessments undertaken for offshore installations may vary widely from simple assessments using empirical models to complex analyses using multiple CFD simulations. Key issues identified from a review of industry practice in this area are as follows:
- Explosion hazard assessments are sometimes undertaken in detail for some installations then by 'difference' for others. This raises questions as to whether the installations in question are indeed comparable from an explosion hazard perspective.
- A common assumption made is that if the explosion analysis is undertaken on the basis of a module filled entirely with gas at stoichiometric composition that this must represent the worst case. This neglects the important influence of:
- congestion in determining localised peak overpressures and that the turbulence associated with a realistic release case may give more severe overpressures (even if the gas does not completely fill the module);
- in calculating explosion risks in relation to escalation and TR impairment, calculation of low frequency worst case explosions will potentially ignore higher frequency, lower overpressure incidents that are capable of significantly contributing to the overall explosion risk.
- In assessing realistic release cases, dutyholders may assume that explosion overpressures scale in a simple manner with gas cloud volume, but this again neglects the important influence of congestion.
- The methodologies used for the development of exceedance across the industry curves are not consistent. The treatment of uncertainties is not clear and the wide range of methodologies employed, ranging from use of generic curves, combinations of phenomenological and CFD modelling and solely CFD modelling, prevents comparison between approaches used.
- The development of exceedance curve approaches for gas explosion modelling does not appear to be systematically documented or auditable.
- For explosion analyses undertaken during design, assumptions regarding the level of congestion in the module (in the absence of detailed design information) are critical. Explosion overpressures have, in the past, been significantly underestimated during the early stages of design.
- The interaction of explosion overpressures with structural response is not well understood.
Strategy development issues
Understanding of explosion phenomena
- Undertake a critical review of the large-scale experimental data from the JIP tests in order to maximise the information available from the tests.
- Improve understanding of the results of the JIP Phase 3B realistic release tests and interaction of gas releases with ventilation.
- Assess how representative the so-called realistic gas clouds used in explosion tests are of the actual conditions that could occur in a real incident.
- Encourage the development of diagnostic techniques for characterising transient mist clouds.
- Promote small-scale experimental work on mist explosions to gain a better understanding of the basic mechanisms and aid the assessment of the consequences of these explosions.
- In the longer term, consider the undertaking of larger scale experiments on mist explosions.
- Promote best practice in the application of CFD and other modelling techniques to gas explosion modelling.
- Encourage organisations to ensure that the person carrying out CFD calculations has a thorough understanding of fluid mechanics, combustion and CFD - possibly also putting a 'buddy' system into place, thereby ensuring that quality checks are carried out.
- Investigate possible differences in results between different versions of codes used in industry.
- The level of inconsistency between exceedance curve approaches is of concern and requires further discussion and appraisal.
- Encourage more openness from the code developers with regard to:
- the results of the validation exercises - perhaps with a document outlining the validation cases, etc; and
- the numerical and modelling techniques implemented in the CFD codes.
- Encourage CFD code developers to:
- incorporate better physical sub-models for ignition, laminar flame growth and turbulent combustion,
- incorporate better turbulence models, an accurate model for the transition from laminar to turbulent flow, and improved two-phase flow models
- incorporate more accurate differencing schemes and more efficient solvers, which are robust,
- introduce mesh refinement and de-refinement so that flame fronts and obstacles can be resolved properly.
- Encourage more validation, i.e. not calibration or 'tuning', of the CFD codes and their constituent sub-models.
- Improve understanding and modelling of the interaction of explosions with structural response.