It is all-too-well known that:

1. combustible gases may leak into oil-platform modules, forming combustible mixtures with air;
2. these mixtures may be inadvertently ignited, leading to explosive flame propagation;
3. damage to and partial destruction of the platform may result in the projection of missiles; and
4. consequent penetration of vessels containing liquid fuel may lead to spillage and the spread of fire.

Yet they MUST be estimated, if the satisfactoriness of safety precautions is to be assessed, and approvals to build and operate are to be given.

Computer simulations provide a means of making such estimates, by way of predictions of:

1. the gas-dispersion and mixing process;

2. the ignition, flame-acceleration and explosion phenomena;

3. missile formation and motion resulting from structural damage;

4. the spread of fire resulting from subsequent oil spillage.

Moreover, being an attachment to the general-purpose fluid-flow and solid-stress-analysis code PHOENICS, it can call on all the further capabilities of that code which it requires.

An example: the escape of gas from a leaky flanged joint

The following pictures illustrate the use of PHOENICS, with its "fine-grid-embedding feature", to investigate one element of the Piper Alpha scenario considered in the Cullen report (Ref.1).

Fine grids, ie large numbers of computational cells, are often needed for such simulations, because small details, for example the size of the gap between two flanges, often have very significant effects.

Computer-based prediction methods are used for hazard assessment, and for estimating the utility of counter-measures, because experimental studies are far too expensive.

Experimental studies ARE carried out, on laboratory, intermediate or full scales; but their main use is for testing of the validity of the computer-simulation methods.

The recent full-scale-module experiments conducted by the Steel Construction Institute at Spadeadam exemplify the difficulty:

• the results have confirmed that the explosion process is affected by the number, shape and size of the objects which are present; an
• the expense has been such as to preclude investigation of more tha a few configurations.

The predictive techniques used for explosions have been categorised (Ref 2) as being based on:

1. "empirical models",
2. "physically-based" models, and
3. "numerical" models;

1. Empirical models are essentially correlation formulae, which seek to derive from such experimental data as are available the relations between:
   • a quantity of interest (eg the peak over-pressure) and
   • conditions of the experiment believed to be important (eg the number of obstacles and the ratio of their size to their spacing).

2. Physically-based models make more direct use of the laws of physics, to provide guidance as to the form of the correlations; but the geometrical postulates are usually very simple. For example, a few connected chambers may be considered, in each of which conditions are supposed uniform.

   The programs CLICHE, SCOPE and EXTRAN fall into this category.

   They can be regarded as rudimentary numerical models, falling far short of them in realism and detail, but possessing the merits of speed and cheapness.

3. Numerical models are also "physically-based", but they are designed to take ALL significant geometrical and material-property data into account, and so to provide superior realism, albeit at much greater expense of time and money.

   Computer codes in this category are FLACS, EXSIM, REAGAS, COBRA and EXPSIM, all of which have been specially designed for the simulation of explosions.

   EXPLOITS belongs to the "numerical-model" group; but it differs from the others in being a special-purpose embodiment of a general- purpose code, rather than a one-use-only computer creation.

   All such codes seek to obtain SOLUTIONS OF THE FUNDAMENTAL EQUATIONS of science applied to a sufficiently finely-divided set of imaginary "cells" so configured as to represent, geometrically and in material properties, the real-life object of study.

   This attempt can be called: THE CFD APPROACH; for it employs the techniques of Computational Fluid Dynamics.

   With the increasing power of computers, and a lengthening record of practical achievements, it is this approach which can be expected to dominate hazaed analysis in the future.

1.4 Why computer predictions cannot be fully trusted

• Predictions based on METHOD (1) could be be trusted rather completely if the correlated experimental data pertained to essentially the same conditions as those for which the predictions were required.

  However, this is so rarely the case that the method is used only by those who have nothing better available. It may however become useful again as the correlator of numerical predictions, when sufficient of these have been carried out.

• Predictions based on METHOD (2) are perhaps a little more reliable, because the laws of conservation of mass, momentum and energy, which they do respect, can certainly generate upper and lower limits to some of the quantities which are te be predicted.

  However the differences between a containable accident and an intolerable disaster can easily depend on precisely those details (for example the shapes of turbulence-promoting obstructions) for which these models have no space.

• Predictions based upon METHOD (3), ie the CFD approach, could be trusted if:-
  • the number of imaginary cells were sufficiently large;

  • the shapes and properties of the materials (solids and gases), and their starting and boundary conditions, were correctly introduced; and

  • the scientific basis of the equations solved were sufficiently well-established.

Therefore, despite their greater sophistication and expense, the "numerical models" can also not be trusted completely.

Nevertheless,

• partly because the gaps in scientific knowledge with regard to turbulence are being steadily closed,
• partly because the SCI experimental programme has shown that the higher-than-expected over-pressures predicted by numerical models can indeed occur, but also
• because the extremes of disbelief and over-credulity are nowadays less frequently encountered,

Provided that its predictions are treated as probabilities rather than certainties, CFD renains the BATWEC, ie

"The Best Available Technique Without Excessive Cost".

It is reasonable for potential users of computer codes to ask their vendors: Has your code been adequately validated?

And it is understandable that each vendor seeks to make his "Yes" more convincing than that of his competitors, preferring a resounding affirmation to the more cautious (and honest) "Up to a point..."

Most codes can claim NUMERICAL VALIDATION, in the sense that they produce error-free solutions to the equations which they solve.

Some can claim PHYSICAL-BENCHMARK VALIDATION, defined as showing that the code's built-in models of turbulence fit a wide range of well-researched experimental data.

Others make much of what might be called HINDSIGHT VALIDATION, having "tuned" some adjustable parameters to fit a few data.

EXPLOITS possesses the first two kinds of validation in high degree; for it is simply a special-purpose version of PHOENICS, which, if it is not the most thoroughly tested CFD code in existence, certainly should be; for it has the longest pedigree, and is the most widely used.

Hindsight validation it does NOT possess, the "tuning" of parameters to fit a few experimental data being regarded by its creators as not conducing to its credibility.

In its fields of application (gas-dispersion, explosions, missile projection, fire spread) it seems best to regard EXPLOITS as an exploratory instrument, which enables the effects of making upper- and lower-limit assumptions to be worked out quantitatively.

Especially in relation to turbulent flame acceleration, it is wise to adopt the NOKFOS principle that "Nobody Knows For Sure". The reason is that it is SAFEST to do so; and safety is what risk- assessment exercises are intended to secure.

(a) Hardware

EXPLOITS will run satisfactorily on any personal computer or work- station, its speed (of course) depending on the power of the machine.

However, because oil-platform geometries are often of considerable geometrical complexity, requiring many computational cells for their proper representation, and because many time steps have to be considered, super-computer power is needed.

Otherwise the computations take an intolerably long time.

The most cost-effective means of acquiring this power is to use a parallel computer, or a cluster of work-stations.

The use of "remote computing"

It is not to be expected that any organization will buy a parallel computer solely for hazard-analysis calculations.

Even a dedicated work-station is more than some organizations will judge it right to acquire.

Fortunately, telephone-line access to remote parallel (or other super-) computers is being made available, especially within the European Commission.

Users adopting this route require only a personal computer for setti up problems and reviewing results; their computations are performed remotely on a parallel machine.

More is written on this topic below, in section 8.3.

(b) Software

The software to be used should therefore be capable of making effective use of parallel-computer capabilities.

The other main requirements of the software are:-

• embodiment of the most realistic models of turbulent mixing and chemical reaction;

• the capability of representing adequately the presence and influence of solid obstacles immersed in the gas;

• possession of efficient numerical schemes, with error-estimating attachments;

• ease of use, both at the problem-set-up and results-inspection stages.

(c) Human

Provided that the software satisfies the ease-of-use requirement, it is not necessary for its user to have any computer-specific skills other than the ability to use a keyboard and a mouse.

It is therefore NOT necessary for organizations considering the use of CFD to suppose that they must hire, and maintain in employment hard-to-find CFD specialists.

EXPLOITS has been provided with an extremely user-friendly interface of the "virtual-reality" kind, which makes "child's play" (almost literally) of the setting up of simulations.

One of the special merits of this interface is that the results of the calculation, as they appear on the screen, are immediately intelligible to non-technical viewers.

wbs