You are on page 1of 52

Risk ID & Assessment Training Course DNV Consulting

Module 22: Consequence Analysis (3) – Effect Modelling

MODULE 22

CONSEQUENCE ANALYSIS (3)


EFFECT MODELLING

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page i

CONTENTS

22. CONSEQUENCE ANALYSIS (3) – EFFECT MODELLING....................................1

22.1 TOXIC EFFECT MODELLING ............................................................................1


22.1.1 Toxic Effects of Clouds and Plumes ................................................................1
22.1.2 Modelling Toxic Effects in a QRA...................................................................2

22.2 FIRE MODELLING AND EFFECTS.....................................................................6


22.2.1 Fireballs/BLEVEs ...........................................................................................6
22.2.2 Jet Fires ........................................................................................................12
22.2.3 Pool fires.......................................................................................................21
22.2.4 Flash Fires.....................................................................................................28
22.2.5 Confined and Obstructed Fires ......................................................................29

22.3 EXPLOSIONS .....................................................................................................32


22.3.1 Introduction to Gas Explosions .....................................................................32
22.3.2 Confined Explosions .....................................................................................34
22.3.3 Semi-Confined Explosions.............................................................................35
22.3.4 Unconfined Explosions..................................................................................38
22.3.5 Detonations...................................................................................................40

22.4 CONFINED AND OBSTRUCTED FIRES...........................................................42


22.4.1 General Description.......................................................................................42
22.4.2 Modelling Approaches ..................................................................................42
22.4.3 Gas Pool Fires...............................................................................................42
22.4.4 Confined Fires...............................................................................................43
22.4.5 Obstructed Pool Fire .....................................................................................44

22.5 SMOKE GENERATION......................................................................................45


22.5.1 Application ...................................................................................................45
22.5.2 Accident Experience......................................................................................45
22.5.3 Smoke Generation.........................................................................................45
22.5.4 Smoke Composition......................................................................................46

22.6 REFERENCES.....................................................................................................49

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 1

22. CONSEQUENCE ANALYSIS (3) – EFFECT MODELLING

22.1 TOXIC EFFECT MODELLING

The toxic effect consequence toxic effect models take the output from the dispersion models
and use it to calculate the effects on people (or effect zone), given that the specified release
and weather conditions occur. Toxic consequences are expressed as one set of results which
give the variation in fatality risk with distance from the release point.

22.1.1 Toxic Effects of Clouds and Plumes


The effects of exposure to a toxic material can be divided into two categories according to the
duration and concentration of exposure.

• Acute Toxic Effects. These arise from short-term exposure at high concentrations.
Carbon monoxide poisoning is an example of an acute toxic effect.

• Chronic Toxic Effects. These arise from long-term exposure at low concentrations.
Asbestosis and lead poisoning are examples of chronic toxic effects.

Hazard assessments are usually concerned with releases of large quantities of material, which
disperse after an hour or so; therefore, hazard assessments consider acute effects only.

The effects of acute exposure include:

• Irritation. Irritation can be felt by the respiratory system, by the skin, and by the eyes.
With some materials the irritation occurs at low concentrations and can be a warning to
people to seek shelter, e.g. chlorine.

• Narcosis. Some materials affect people's responses, in such a way as to make them slow
to seek shelter or to warn others, e.g. hydrocarbon vapours.

• Asphyxiation. Most gases can cause asphyxiation by displacing oxygen from the
atmosphere; others, such as carbon monoxide, cause asphyxiation by displacing oxygen
from the blood and thus preventing oxygen from reaching the tissues.

• Systemic damage. Some materials cause damage to the organs of the body. This damage
may be temporary, or it may be permanent, e.g. hydrogen chloride.

The severity of the effects depends on the concentration and duration of exposure, and on the
toxic properties of the material. However, the toxic properties of many materials are not well-
established because the data about their effects of humans are very sparse. It is difficult to
derive properties from real incidents because the concentrations and durations in these
incidents are usually not known. Most toxicity data are based on experiments on animals, in
which concentrations and durations can be controlled. However, the applicability of the
results of such experiments to humans is questionable, given the differences in body weight
and physiology. Therefore the analyst should regard toxicity data with caution.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 2

Another problem in predicting toxic effects is the variation in vulnerability among the
population. Because of this variation, expressions for toxic effects give the proportion of the
population which would be expected to manifest the effect after a given exposure. The above
problems make it difficult to summarise the toxicity of materials in order to compare them,
since duration, concentration, and percentage-affected all have to be standardised.

One way of comparing the toxicity of materials is to use the LC50 for a stated duration of
exposure; this is the Lethal Concentration which would be expected to cause fatal effects to
50% of the exposed population over the exposure period. Another way of comparing or
expressing toxicity is to specify a harmless level of exposure for a given period as the limiting
exposure criterion; these levels are known as Threshold Limit Values (TLVs) and are specified
for periods such as a normal working life and short-term periods for emergency exposure.

Yet another criterion is the IDLH value for a 30 minute exposure, as described in NIOSH
(1994); this is the dose which is Immediately Dangerous to Life or Health at the exposure
duration. The IDLH value is usually a good choice for the minimum concentration of interest,
except in cases where the release may persist for a long time and evacuation is difficult. In
such cases high levels of fatality may be possible at concentrations below the IDLH value.

All of these criteria can be useful when choosing the lowest concentration of concern in
dispersion calculations. Some consequence models use LC50 in order to illustrate how toxic
effects can be analysed but he analyst might want to use a different concentration; the analyst's
chosen criterion can simply be substituted for the LC50. The LC50 value must be considered
too high for a limiting concentration, but may be a useful parameter in certain special
circumstances.

22.1.2 Modelling Toxic Effects in a QRA


The toxic effect consequence model takes the output from the dispersion models and use it to
calculate the effects on people (or effect zone), given that the specified release and weather
conditions occur. Toxic consequences are expressed as a set of results which give the
variation in fatality risk with distance from the release point.

In general, the impact of a toxic release decreases continuously with distance as illustrated in
Figure 22.1. Use of the LC50 is equivalent to replacing the continuous distribution by a “step
function” as illustrated in Figure 22.2. When estimating the number of fatalities that could
potentially result from a release, use of the LC50 will usually give a good estimate only when
the population distribution is uniform over the area covered by the cloud. In reality this is very
rarely the case.

In a QRA the toxic effects are normally calculated using the “probit” equation. This relates
the probability of fatality to the dose (concentration-time relationship) received. In this way
curves of probability of fatality versus downwind distance can be produced for releases of
toxic chemicals.

Eisenberg et al. (1975) published a Vulnerability Model for toxic agents, amongst other
hazards, based on a probability function or “probit” of the toxic load. This probit equation is
in the form of:

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 3

Pr = a + b ln (TL)

TL = Toxic Load = ∫ c(t)n dt

  Pr − 5 
Probability of Death = 1 1 = erf  
2  2 

Where: Pr = Probit function value


C = Toxic gas concentration (ppm)
t = Duration of exposure (minutes)
a, b and n = constants, depending on the material.

The probit function is a normally distributed function with a mean of 5.0 and a variance of 1.0.
A probit of 5.0 corresponds to 50% fatalities, 3.36 to 5% fatalities and 6.64 to 95% fatalities.
Implicit in the toxic load method is the concept of variable vulnerabilities to a given
concentration of toxic material for a specific duration. The very young and the elderly are
most vulnerable.

Table 22.1 shows the relationship between probability of fatality values and probit values.

Table 22.1 Fatality Probability versus Probit Values

Probability 0.01 0.02 0.05 0.10 0.20 0.50


Probit (Pr) 2.67 2.95 3.36 3.72 4.16 5.0

The three toxicity parameters a, b and n may be combined with tabulated probit values to
determine the concentration that can be expected to produce a stated fatality level for a stated
duration of exposure, for example 50% fatalities (i.e. the LC50).

Figure 22.1 Toxic Effect vs. Distance

1
Probability of death

0.5

0
Distance from release source

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 4

Figure 22.2 Step Function Equivalent

1 Step function used

Probability of death

0.5

0
Distance from release source

The concentration (ppm), in terms of the toxicity parameters, the duration (minutes) and the
probit value is:

1/ n
 e (Pr −a ) / b 
C = 
 t 

The concentrations in ppm corresponding to the probit values can be calculated by substituting
the constants for a, b and n and the exposure time (t) with the values for that particular
material.
The approach described above can be used for time varying exposures as might occur
following a large instantaneous failure. The toxic load can be accumulated for increments of
time as the cloud passes over.

Table 22.2 gives a selection of toxicity data for a range of toxic materials for which QRAs are
often undertaken. Because toxicity is a function of both concentration and exposure, toxicity
criteria give both a concentration and a time to give a certain level of risk. For the probits, the
units of concentration, C, are ppm, and the units of time, t, are minutes.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 5

Table 22.2 Single Value and Probabilistic Criteria for Common Substances

Single Value Criterion


Material Time LC50 Fatality IDLH HSE DTL Probit
(min) (ppm) Threshold (ppm) (ppm)
(ppm)
Ammonia 10 10800 6000 6000 Pr = -9.28 + 0.71 ln(C 2t)
30 6200 3600 500
Chlorine 10 940 100 100 Pr = -10.1 + 1.11 ln(C1.65t)
30 490 60 30
Hydrogen fluoride 10 5900 1200 1200 Pr = -48.33 + 4.853 ln(Ct)
30 2000 400 30
Hydrogen sulphide 10 950 670 670 Pr = -31.42 + 3.008 ln(C1.43t)
30 440 510 300
Hydrogen cyanide 10 165 40 Pr = -9.56 + ln(C 2.4t)
30 105 25 50
Methyl isocyanate 10 620 160 Pr = -5.642 + 1.637 ln(C 0.653t)
30 120 30 20
Sulphur dioxide 10 1900 470 Pr = -15.67 + 2.1 ln(Ct)
30 630 160 100

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 6

22.2 FIRE MODELLING AND EFFECTS


When a flammable material is released and ignited, there are several potential outcomes
depending on the time and location of ignition, the material phase, the type of release, and the
surroundings. The possible outcomes and corresponding hazardous effects are:

l Flammable outcomes: l Hazardous effects


v Fireballs/BLEVEs
v Jet fires v Radiation levels
v Pool fires
v Flash fires v Burning gas/liquid
v Explosions v Overpressure levels

The various fire outcomes are described in the remainder of this section. Explosions are
addressed in Section 22.3. Fires also generate smoke, which can have significant effects on
offshore installations in particular. This is addressed in Section 22.5.

22.2.1 Fireballs/BLEVEs
22.2.1.1 Description

A fireball is an intense spherical fire resulting from a sudden release of pressurised liquid or
gas which is immediately ignited, burning as it expands, forming a ball of fire, rising in the air.
It gives rise to high levels of thermal radiation at a considerable distance.

The best-known cause of a fireball is a BLEVE (boiling-liquid expanding-vapour explosion),


which is where a tank of liquefied flammable gas under pressure ruptures due to fire
impingement. This produces a large fireball, often projecting fragments of the tank for large
distances. BLEVEs are possible from many sources on onshore sites, but could only occur on
an offshore installation in the condensate system, because the other gas is not liquefied.

Nevertheless, the fireball model may also be appropriate to represent two other types of fire
which may occur offshore, and which have roughly spherical effects. These applications are:

• The failure of gas equipment under flame impingement. Alternatively these could be
modelled as immediately ignited flash fires, but the BLEVE model is believed to be more
realistic. Only the mass of the gas content in the process equipment is entered into the
fireball. A pool fire is used for the liquid content.

• The initial release in an ignited full-bore gas riser failure. Alternatively this could be
modelled as a jet fire, but a fireball is believed to be more realistic for the first stage of the
release when the flow rate is extremely high.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 7

22.2.1.2 Accident Experience

Sudden large fires are often described by witnesses as fireballs, but this does not necessarily
imply that a fireball model is the most appropriate theoretical representation of them. The key
characteristics of a fireball are that it is spherical (as distinct from a jet fire) and that it is
sufficiently intense to cause fatalities outside the fire envelope (as opposed to an immediately
ignited flash fire). Few accidents produce evidence of sufficient quality to determine this.

The Piper Alpha accident provides the best evidence for the occurrence of fireballs on
offshore installations. Two fireball types were observed:

• The hemispherical fire which was photographed emerging from B Module shortly after the
first explosion, was due to a full-bore failure of a condensate pipe under fire impingement.
However, it could reasonably have been represented by a confined fire model.

• The fires which were filmed emerging from the gas risers when they failed under fire
impingement were less directional than a jet fire model, but not as spherical as a pure
fireball model. Either model might be appropriate, but a fireball model better accounts for
the uncertainty in predicting the direction of this type of event.

22.2.1.3 Fireball Modelling

The fireball normally is assumed to be a sphere resting on the ground, see Figure 22.3,
although models are now available which account for the fireball rising.

Figure 22.3 Geometry of a Fireball/BLEVE

Fireball

R1
REF

Observe
r

R0

Fireball diameter is based on correlations with observed fireball size in experiments and
actual accidents. The most recent correlation (IChemE 1989) is:

D = 5.8 M0.333
where: D = fireball diameter (m)

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 8

M = mass released (kg)

Fireball duration is based on a mixture of small-scale experiments and dimensional analysis


(I.Chem.E 1989):

0.45 M 1c / 3 M c ≤ 37000 kg
t=
2.59 M 1c/ 6 M c > 37000 kg

where: t = fireball duration (s)

The fireball is assumed to be constant over this duration, although in reality some growth and
decay would occur.

Fireball rise above the ground is evident in filmed BLEVEs, but is not addressed in most
models, i.e. the fireball is assumed to be a sphere resting on the ground.

For offshore applications, the fireball may be restricted by the structure surrounding the
release point. For releases within modules, the fireball may be large enough to fill the module,
and any remaining volume may be assumed to form hemispherical fires outside the open sides.
For full-bore riser releases, the fireball may be assumed to be a sphere tangential to the point
of release and the sea surface.

Fireball Radiation

Surface Emissive Power

The surface emissive power of a fireball is usually assumed to be in the range 150-300 kW/m2.
The HSE use values for LPG of 270 kW/m2 for releases below 125 tonnes and 200 kW/m2 for
larger releases (Kinsman 1991). Experimental measurements of average surface emissive
power of butane fireballs have been reported as 300-350 kW/m2 (Shell Research 1991a).

Surface emissive power accounts for much of the uncertainty in fireball modelling. There are
two main approaches:

• A correlation giving the fraction of heat energy radiated based on initial vapour pressure of
the fluid at ambient temperature (Roberts 1982):

Fr = 0.27 P0.32

where: Fr = fraction of heat energy radiated, constrained to lie between 0.2 and 0.4
P = vapour pressure of fluid (MPa abs)

Roberts (1982) makes clear that P is intended to be the vapour pressure when the failure
occurs. In a BLEVE, this will be the relief valve setting Po, whereas in a fireball following
an impact failure it will be the vapour pressure at ambient temperature, as is used in
PHAST. For a fireball following a release of gas (as opposed to liquefied gas) P should be
the storage pressure.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 9

From geometric considerations the surface emissive power is:

M
E s = Fr 2Hc
πD t

where: Es = surface emissive power (kW/m2)


Hc = heat of combustion (J/kg)

This is used in PHAST.

• A correlation based directly on vapour pressure of the fluid (Moorhouse & Pritchard
1982):

Es = 235 P0.39

This is based on 6.2kg pentane BLEVE experiments. It can be generalized as:

Es = 5.2 x 10-6 Hc P0.39

According to Kinsman (1991), HSE use the relief valve setting pressure for P instead of
the initial storage pressure which the correlation was based on.

In either case, the surface emissive power should be constrained to between 150 and 300
kW/m2.

Radiation Calculation

As a BLEVE is spherically symmetric, for a given received radiation level FDOSE the observer
distance R0 is given by:

Es
R0 = Rmax −1
FDOSE

Figure 22.4 shows the hazard zone modelled for a fireball/BLEVE. The “radiation dose of
concern” as shown in this figure is discussed in Module 23.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 10

Figure 22.4 Fireball/BLEVE Output

Effect Radius
Radiation Dose
of Concern

Release
Source

Fireball of BLEVE

Transmissivity is the fraction of radiant heat absorbed by water vapour in the atmosphere
before reaching the target. PHAST uses an expression from Raj (1977):

τ = 1.389 - 0.135 log10 Pw R

where: τ = atmospheric transmissivity


Pw = partial pressure of water vapour in air (Pa)

22.2.1.4 Example Fireball Modelling Radiation Results

Fireball radiation fields have been estimated by a spreadsheet implementation of the PHAST
model described above using the following inputs:

• Material = Methane
• Heat of combustion = 50 MJ/kg
• Release pressure = 20 barg
• Water pressure = 1628 Pa
• Maximum surface emissive power = 300 kW/m2

The fireball radius and duration, and the distances to radiation levels of 37.5 kW/m2,
12.5 kW/m2 and thermal dose of 375 kJ/m2 are given in

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 11

Table 22.3. The radiation levels are calculated for a person on the ground, i.e. level with the
bottom of the fireball, and the distances are measured from the release point, i.e. the middle of
the bottom of the fireball.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 12

Table 22.3 Fireball Hazard Distances

Mass in Fireball Fireball Distance (m) to


2
Fireball (kg) Radius (m) Duration (s) 375 kJ/m 37.5 12.5 kW/m2
2
kW/m
100 13 2.1 8 20 54
300 19 3.0 18 43 75
1,000 29 4.5 37 62 113
3,000 42 6.5 68 88 160
10,000 62 9.7 127 130 235
30,000 90 13.9 220 183 330

22.2.2 Jet Fires


22.2.2.1 Description

Jet fires are burning jets of gas or atomised liquid whose shape is dominated by the momentum
of the release. On an offshore installation, jet fires typically result from gas or condensate
releases from high pressure equipment or a blowout. A jet fire is an intense fire that gives rise
to high levels of thermal radiation at a considerable distance. If directed at an angle to the
wind, the flame will be bent by the wind as illustrated for a vertical release in Figure 22.5.

Figure 22.5 Shape of Jet Flame

Wind Direction

For process streams containing a mixture of oil and gas, a combined jet and pool fire may
occur.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 13

22.2.2.2 Jet Flame Modelling

Jet Flame Shape and Size

The shape shown in Figure 22.5 is the shape on which the API model (API 1982) is based,
from which thermal radiation can be calculated. The API model has the following features:

• Assumptions: • Affected by: • Flame properties:


• Jet extended to LFL • Humidity • C5 and less - luminous
• 10 point sources • Windspeed • C5 and greater – smoky
• Orientation

If a jet flame length is all that is required (for example, to see if a jet flame could reach a
critical piece of equipment and cause escalation, the so-called ‘Wertenbach’ model provides a
simple correlation between jet flame length L and mass release rate q, independent of the
material:

L = 15 Q 0.41

The Shell (Chamberlain, l987) model is a more recent and better model for thermal radiation
from jet flames. The model describes the shape of a jet flame as a frustrum of a cone as shown
in Figure 22.6. The parameters describing the frustum, shown in the figure, have been derived
from comparisons with experimental data from laboratory and field tests. The flame shape
does not correspond to the visible flame but is suitable as input for modelling the hazard effect
zone.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 14

Figure 22.6 Shell Jet Flame Model

Z Wind Direction

W2

α RL
W1
L

b
θ X
The base of the jet flame is usually not attached to the release point, due to the high velocity
and richness of the fuel near the source. This lift-off distance has been measured on flares to
be 20% of jet length (Chamberlain 1987), but a value of 10% is typically used in QRA to
account for the extra turbulence around the edges of real pipe failures compared to a flare.
This effect is important in reducing the predicted radiation level on the leak source, which
might otherwise cause a small leak to escalate to a full-bore failure.

For gas releases through small holes, some discharge conditions may prevent a stable flame
from forming (Shell Research 1991a). If ignited, the flame blows off downstream and is
extinguished. However, nearby surfaces may stabilise the flame, and normally in a QRA any
size of hole is assumed to be capable of forming a stable jet flame.

Jet Flame Radiation

Surface Emissive Power

Heating outside the jet flame is dominated by radiation arising from hot soot particles and
gases within the flame. There are several well-validated semi-empirical prediction techniques
for the radiation field. They are reviewed and compared with experimental data by Shell
Research (1991a). This concludes that multiple point source and surface-emitter models (such
as PHAST) give reasonable predictions to within about a flame length of the fire. The
predictions from these models are mainly within a factor of about 1.5 of the experimental
radiation measurements. They are not valid for impingement conditions. More sophisticated
models are being developed, but are not yet validated for routine predictions.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 15

The radiation consequence zones are calculated by assuming the flame to be a surface emitter
with a surface emissive power given by:

FS m∆H C x10 −3
ES =
S

where: Fs = Fraction of heat emitted as radiation


m = mass flow rate (kg/s)
∆HC = heat of combustion (J/kg)
S = surface area of flame (m2)

Jet Flame Radiation

Having calculated the shape of the flame and its emissive power, the radiation incident at a
point can be calculated by integrating numerically over the surface of the flame the thermal
radiation from small flame surface elements. The integration takes account of the distance
between each element and the point as well as the projection of the element as seen from the
point. The atmospheric transmissivity, i.e. the absorption of radiation between the source and
observer, is taken from Raj (l977).

The hazard effect zone of the flame is taken as the ground-level contour of a specified
radiation intensity level. For modelling purposes the effect zone modelled for a jet flame is
idealised as an ellipse, as shown in Figure 22.7 (NB this shows the true flame shape rather than
the cone frustrum of the Shell model). “a” and “b” are the semi-major and semi-minor axes of
the ellipse respectively. The centre of the hazard effect zone ellipse is some distance “x”
downwind of the release point. The “radiation level of concern” as shown in this figure is
discussed in Module 23.

Figure 22.7 Jet Flame Output

Radiation Level
of Concern

x a
Release Source

Jet Flame

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 16

22.2.2.3 Example Jet Fire Modelling Results

Jet fire sizes are given in Table 22.4 and Figure 22.8 for a high-pressure methane release. The
results were obtained using the standalone jet fire model in PHAST (implementing the
Chamberlain model) with the following inputs:

• Material = methane
• Source pressure = 70 × 105 N/m2
• Source temperature = 20oC
• Wind speed = 5 m/s
• Jet orientation = horizontally downwind
• Hole diameters = from Table 22.4

The jet radius, temperature and density after expansion to ambient pressure are calculated
from the hole diameter and the pressure and temperature before the release, using the
discharge modelling in PHAST.

Table 22.4 Gas Jet Fire Characteristics

Release Rate Hole Diameter Flame Length Surface Emissive


Power
(kg/s) (m) (m)
(kW/m2)

1 0.011 11.4 117

10 0.033 31.4 152

100 0.105 87.6 195

1000 0.330 240.2 259

As in the Wertenbach correlation, the Chamberlain model is mainly influenced by the mass
release rate. Although parameters such as fuel type, wind speed, hole size and upstream
pressure do influence the results, their importance is minor compared to other uncertainties in
the overall analysis. The resulting flame lengths can be represented by the following
correlation:

Lf = 11.14 Q0.447

This may be used as an approximation to the flame length in the Chamberlain model for typical
offshore conditions.

The internal angle of the flame cone in each case was 16.4o.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 17

Figure 22.8 Jet Fire Hazard Ranges

1000
Maximum Distance (m)

100

10
1 10 100 1000
Release Rate (kg/s)

Flame Length Dist. to q=5 Dist. to q=12.5 Dist. to q=37.5

Contours of heat flux density are shown for a typical fire in Figure 22.9, calculated by
PHAST. The standalone fire models included in the PHAST software package estimate a
surface emissive power from the fuel flow rate and velocity. This is assumed to be uniform
over the flame surface. Typical values are given in Table 22.4. It then determines the view
factor and transmissivity between the flame surface and the observer location.

As a result of the conical flame shape, the contours cluster close to the flame near the point of
release. As a result of the assumed flame lift-off, the heat flux at the point of release itself is
minimal. This conforms with practical fire-fighting experience that it is possible to approach
close to a flame from behind the point of release.

The PHAST model strictly only determines the thermal radiation field around the flame. It
might be appropriate to adjust this to account for convective heating. This effect would be
greatest at the flame tip. Previous studies have modelled it by a 50% increase in flame length.
However, this is considered to be an unduly conservative approach.

The maximum distances (along the axis of the frustum) to radiation levels of 37.5, 12.5 and 5
kW/m2 and the flame length are plotted as a function of release rate in Figure 22.8. The
distances between the flame surface and the radiation contours are given in Table 22.5 as
fractions of the flame lengths, since these fractions are virtually independent of release rate.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 18

Figure 22.9 Jet Fire Radiation Zones

5kW/m2
12.5kW/m2

37kW/m2

Table 22.5 Gas Jet Fire Hazard Ranges

Radiation Level Distance to Radiation Level


(kW/m2) Flame Length

37.5 1.2
12.5 1.45
5.0 1.75

The jet flame hazard zone can be represented more simply by an ellipse. The geometry is
shown in Figure 22.10.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 19

Figure 22.10 Jet Flame Output

Radiation Dose Level


Note: d =x/a
of Concern

x a
Release Source

Jet Flame

22.2.2.4 Heating of Impinged Objects

Extensive experiments on jet flame impingement are described by Shell Research (1991a). No
predictive techniques are available for this at present. The surface emissive power estimated in
jet fire models (see above) is sometimes used but is not strictly applicable.

In natural gas flames, total heat fluxes have been measured in the range 50-300 kW/m2 for
flow rates up to 10 kg/s. The areas of maximum flux on the target surface are relatively small
and only occur a certain locations within the flame. Maximum average heat fluxes over the
total flame impinged areas were approximately 200 kW/m2 (Shell Research 1991a).

In two-phase propane flames, total heat fluxes have been measured in the range 50-250 kW/m2
for flow rates up to 20 kg/s, with maximum average heat fluxes over the total flame impinged
areas of approximately 150 kW/m2 (Shell Research 1991a). These are comparable to those in
pool fires.

These values are significantly lower than assumed in previous QRAs.

22.2.2.5 Effect of Obstacles

Available semi-empirical models such as PHAST cannot model the effect of obstacles
distorting the flame and the radiation contours. This is particularly important for jet fires
inside modules or pointing upwards from underneath a platform.

A simple judgmental representation of the effects of obstacles within the radiation zones is as
follows (Figure 22.11):

• Solid obstacles such as walls on enclosed spaces are considered impervious to thermal
radiation, and are assumed not to affect the radiation zones outside them.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 20

• Partial obstacles such as process equipment or decks on open-sided modules are


approximated by reducing the radiation zones as follows:

37.5 kW/m2 becomes 12.5 kW/m2


12.5 kW/m2 becomes 5 kW/m2
5 kW/m2 becomes 2 kW/m2

These reductions are only applied more than 10m from the edge of the obstructed zone.

Figure 22.11 Effect Of Obstacles On Radiation Zone

Solid 5kW/m2
12.5kW/m2
Barrier

37kW/m2

Partial
Barrier

10m

Obstacles within the flame zone tend to increase the area impinged by the flame compared to
idealised predictions. The effect can be investigated by CFD modelling, but this is not yet
properly validated.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 21

22.2.3 Pool fires


22.2.3.1 Description

Pool fires are burning pools of liquid which has collected on a horizontal surface. On an
offshore installation, this may be a deck of the platform or the sea surface. For an onshore
installation this may on the ground or in a bund. The liquid may be crude oil, condensate or
refined products such as gasoline, kerosene or diesel oil. The pool fire models are applicable
to fuel-controlled (i.e. unconfined) fires, which may occur in the open, on the ground, external
decks or the sea surface.

Pool fires may result from oil blowouts, riser or pipeline leaks, process leaks, collisions
causing oil spills, helicopter crashes, or storage fires.

For fires in two-phase fluid, containing a mixture of liquid and gas, several approaches are
possible:

• Select a model (e.g. pool fire or jet fire) for the component which has the largest mass, and
enter the full released mass in the calculations. For mainly gas releases, this assumes the
liquid is entrained in the gas. For mainly liquid releases, it assumes the gas adds to the
burning of the liquid.

• Model each component (gas and liquid) separately, and combine the results in the impact
modelling. This tends to produce superimposed pool fires and jet fires. It is more time-
consuming, but gives a better representation of the confused nature of real fires.

22.2.3.2 Pool Fire Ignition

Basic Principles

The evaporation of light hydrocarbons from a pool forms a cloud of vapour above the pool
surface. In a release of two phase or volatile material, the gas which flashes from the liquid
before the pool is formed may create a larger flammable cloud than the vapour evaporated
from the pool surface.

The flash point of the liquid indicates the ease with which it is ignited. The flash point is the
lowest temperature at which the liquid will give off vapour sufficient to form a flammable
mixture with air. If ignited, a flame will flash through the vapour to the liquid surface. If the
liquid is close to its flash point, the flame will then go out.

If the liquid temperature is higher, above its fire point, the burning vapour will generate heat
which evolves more vapour and achieves stable combustion.

The flash points of various liquid hydrocarbons are given in Table 22.6 (SINTEF 1992). The
fire points are in general 20-40ºC higher. Condensate, naphtha and gasoline have flash points
below ambient temperature and are readily ignited. Fuel oil, kerosene and stabilised crude oil
have flash points above ambient temperature, and can only be ignited if heated. Unstabilised
crude oil has a flash point close to ambient temperature, and ignition in the absence of heating
is uncertain.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 22

Table 22.6 Hydrocarbon Flash Points

Material Flash Point

(oc)

Stabilised crude oil 125

Unstabilised crude oil 1 - 10

Fuel oil 133

Kerosene 49

Gasoline -45

n-Nonane 31

n-Octane 13

n-Heptane -4

n-Hexane -23

n-Pentane -117

Ignition by Heating

Oil pools on the sea surface may be heated to their flash point by the flare (as in the Ekofisk
Central accident) or by other nearby fires (where the oil spill is due to escalation). For oil at
ambient temperature, heating at 10 kW/m2 enables ignition within about 1 minute. Ignition
times in other heat fluxes are inversely proportional to (heat flux)2 (Scandpower 1992).

Pool Fire Burning Rates

When first ignited, the fire spreads rapidly across the full extent of the hydrocarbon pool and
proceeds to consume the liquid at a characteristic burning rate. Burning rates for hydrocarbon
fuels on land are given in Table 22.7 (Mudan & Croce 1988). More extensive data is given by
Shell Research (1991a).

Table 22.7 Hydrocarbon Burning Rates On Land

Material Mass Burning Rate Burning Velocity


(kg/m2s)
(mm/s)

Gasoline 0.05 0.07

Kerosene 0.06 0.07

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 23

Hexane 0.08 0.11

Butane 0.08 0.13

LNG 0.09 0.20

LPG 0.11 0.21

On water, liquefied gases burn more quickly than on land due to the heating from the water.
The burning rate for LPG on water is twice that on land (Mudan & Croce 1988).

For liquids, the initial burning rate on water is the same as on land. In the later stages of the
fire, the burning rate is reduced due to cooling by the water.

Burning rates for crude oils on water are given by SINTEF (1992). They are in the range 0.02
to 0.07 kg/m2s (0.02-0.08 mm/s). These are mainly based on tests on pools of 2m diameter or
less. Data on the effect of pool size indicates that the burning rate reaches a maximum at 10m
diameter, where it is twice the burning rate at 2 m diameter (SINTEF 1992).

The burning rate for the 10m diameter fire was approximately 0.05 kg/m2s, and this is
considered appropriate for offshore crude oil fires. A typical burning rate for condensate is
taken as 0.08 kg/m2s (based on hexane in Table 22.7).

Since the pool fire size is inversely related to the burning rate, a value at the lower end of the
range should be selected in order to be conservative.

22.2.3.3 Pool Fire Modelling

The flame for a pool fire can be modelled as a sheared cylinder, having a circular cross-section
in a plane parallel to the ground, as illustrated in Figure 22.12.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 24

Figure 22.12 Pool Fire Flame Shape

Wind Direction

H
θ

D
The mass burning rate can be taken to be independent of the diameter of the pool, consistent
with experimental data for liquid hydrocarbon pool fires with diameters in excess of l m. The
correlation of Burgess and Hertzberg (l974), based on experimental burning rate
measurements, can be used for the mass burning rate m& for fires on land. This is typically of
the order of 0.05 kg/m2s.
For materials with a boiling point below the ambient temperature, their heat of vaporisation is
just equal to the normal heat of vaporisation. A correction has been made for materials with a
boiling point above the ambient temperature to account for the heat required to raise the
temperature of the liquid to its boiling point.

The correlation with the modified heat of vaporisation has been compared with experimental
data (Mudan and Croce, l988). The burning rate calculated above is increased by a factor of
2.5 for fires of cryogenic liquids on water in accordance with test results discussed in this
reference.

The flame height is obtained from the correlation of Thomas (1965) for the mean visible height
of turbulent diffusion flames from laboratory-scale wooden crib fires:

0.61
H  m & 
= 42 
D  ρ gD 
 a 

where: H = flame height (m).


D = pool diameter (m).
m& = mass burning rate (kg/m2s).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 25

ρa = density of air = 1.2 kg/m3.


g = acceleration due to gravity = 9.8 m/s2.

The flame is assumed to be tilted from the vertical by an angle φ which is calculated (American
Gas Association, 1974) as:

0 u* ≤ 1
φ =  −1 *−1/ 2
cos (u ) u* > 1

where u* is the non-dimensional wind velocity given by:

−1 / 3
&D
 gm
u = U w 
*

 ρv 

and: Uw is the wind speed (m/s).


ρv is the vapour density (kg/m3).

Radiation Calculations

Surface Emissive Power

Fuels for liquid pool fires are assumed to burn with either a luminous or a smoky flame. IN
general, hydrocarbons lighter than pentane burn with a luminous flame, those heavier burn
with a smoky flame. Based on experimental data (Mudan & Croce, 1988) the surface emissive
power Es for a luminous flame is calculated from:

E s = E 0 [1 − exp( − D / L s )]

where: E0 and Ls are respectively the maximum emissive power and a characteristic length
describing how rapidly the emissive power increases with pool diameter. Typical
values of these for different hydrocarbons are:

Material E0 (kW/m2) Ls (m)


LPG 220 2.5
Other with luminous flame 170 2.75
Material with smoky flame 140 8.33

For fuels that burn with a smoky flame, the smoke obscures the flame and absorbs a significant
part of the radiation. However, there may be gaps in the smoke coverage exposing luminous
spots on the flame surface. As the diameter of the pool increases the absorption of the
radiation by the smoke and soot increases, until for very large diameter fires the emitted
radiation is due only to that radiated by the black soot.

The behaviour of surface emissive power with pool diameter for smoky flames is given
(Mudan and Croce, 1988) by:

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 26

E s =140 exp( −0.12D) + 20[1 − exp( −0.12D)]

Radiation Calculations

Having calculated the shape of the flame and its emissive power, the radiation incident at a
point can be calculated in the same manner as for jet flames (Section 22.2.2.2).

The hazard effect zone of the flame is taken as ehe ground-level contour of a specified
radiation intensity level. For modelling purposes the effect zone modelled for a jet flame is
idealised as an ellipse, as shown in Figure 22.13. “a” and “b” are the semi-major and semi-
minor axes of the ellipse respectively. The centre of the hazard effect zone ellipse is some
distance “x” downwind of the release point. The “radiation level of concern” as shown in this
figure is discussed in Module 23.

Figure 22.13 Pool Fire Output

Radiation Level
of Concern

x a
Release Source

Pool Flame

Wind Direction

Side View of Pool Fire

22.2.3.4 Example Pool Fire Modelling Results

The flame sizes for typical fires of a C7 material predicted by the standalone fire modelling in
PHAST are given in Table 22.8. The results shown were obtained using the input values listed
below:

• Ambient temperature = 293K


• Ambient relative humidity = 70%
• Wind speed = 3 m/s

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 27

Table 22.8 Oil Pool Fire Characteristics

Release Rate Pool Diameter Flame Length Tilt Angle

(kg/s) (m) (m) (•)

0.4 2.5 8 58

1.7 5 13 53

7 10 21 48

42 25 39 39

60 30 44 37

167 50 63 29

427 80 88 19

699 100 102 11

The radiation fields have been estimated by PHAST for pool fires of a C7 material using the
following inputs:
• Ambient temperature = 293K
• Ambient relative humidity = 70%
• Wind speed = 3 m/s
• Accuracy parameter = 2%
• Vertical (upright) observer

The distances from the centre of the pool to radiation levels of 12.5 and 5.0 kW/m2 are given
in Table 22.9. The 37.5 kW/m2 contour is within the pool or very close to its edge. These
hazard distances are plotted in Figure 22.14.

Table 22.9 Oil Pool Fire Hazard Ranges

Release Pool Surface Distance to Distance to


Rate Diameter Emissive 12.5 kW/m2 5.0 kW/m2
(kg/s) (m) Power (m) (m)
(kW/m2)
0.4 2.5 109 6.3 12
1.7 5 86 9.4 19
7 10 56 11.0 28
42 25 26 13.2 30
60 30 23 15.6 32
167 50 20 25.2 43
427 80 20 40.2 64
699 100 20 50.0 77

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 28

Figure 22.14 Oil Pool Fire Hazard Ranges

100
Distance (m)

Pool Radius
10 12.5 kW/m2
5.0 kW/m2

1
0.1 1 10 100 1000
Release Rate (kg/s)

22.2.3.5 Heating of Engulfed Objects

Total heat fluxes within heavy hydrocarbon fires have been measured in the range
100-160 kW/m2 (Shell Research 1991a). Heat transfer is mainly by radiation and partly by
convection, since the flow velocities in such fires are relatively low.

22.2.4 Flash Fires


22.2.4.1 Description

A flash fire occurs when a cloud of gas burns without generating any significant overpressure.
The cloud is typically ignited on its edge, remote from the leak source. The combustion zone
moves through the cloud away from the ignition point. The duration of the flash fire is
relatively short, but it may stabilise as a continuing jet fire from the leak source.

22.2.4.2 Flash Fire Modelling

A flash fire can only be ignited within the flammable cloud (i.e. the zone between the lower
and upper flammable limits, LFL and UFL). The flame then propagates through the cloud
wherever the concentration is above approximately 0.9 LFL (DNV ARF T15). Since the
cloud may not be uniform around the edge, pockets of flames may occur where the average
concentration is only 0.6 LFL.

As it burns, the gas expands to 8 times its original volume. This expansion is in the least
constrained direction, which is usually upwards, although there may be some expansion
horizontally. Overall, the volume affected may be 8 times the volume of the unignited cloud
within the LFL.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 29

The effect area of a flash fire can be taken to be the area covered by the cloud above the LFL
(lower flammable limit) concentration.

22.2.5 Confined and Obstructed Fires


22.2.5.1 General Description

The models of jet fires, pool fires, flash fires and fireballs outlined above are all based on
experiments on gas and oil releases in open environments. There is much less knowledge
about fires whose development is limited by the surrounding structure. Research on fires
inside rooms has mainly addressed cellulosic fires, and is not readily applicable to hydrocarbon
fires. Research on gas explosions inside offshore modules (see below) has concentrated on
predicting the overpressure which results, and gives no information on the size of the burning
gas cloud. Available data on confined and obstructed fires is presented by Shell Research
(1991b), but is difficult to generalise.

Obstructions such as process equipment, decks, firewalls conductors etc. may distort gas and
oil fires from the size and shapes in the unobstructed case. Where the obstructions are located
in the radiation field but not the flame, they may be modelled by simple allowance for shielding
(e.g. Section 25.5.4). Where the obstructions are within the flame itself, their effect may be
substantial, and they may make the fire's effects worse, e.g. by diverting it towards an escape
route. Therefore, some form of modelling is desirable.

22.2.5.2 Modelling Approaches

The ability to predict the effects of confinement and obstructions on large fires in buildings and
offshore installations is limited at present. Possible modelling approaches are:

• Physical models. These give the most realistic representation of the effects of
obstructions. Several tests have been carried out on equipment in impinging jet fires,
but no analysis is available of the effects of the equipment on the jet shape.

• Numerical models. CFD models have been used to represent impinging fires. They
are able to model the flow of hot combustion gases, and give a good representation of
the shape of the fire, but do not at present predict the radiation field around it.

• Phenomenological Models. Theoretical models of the combustion within confined


spaces have been developed, but are not yet validated or available commercially.

In the absence of an exhaustive treatment of the subject, several individual fire scenarios are
described below for which an empirical approach is available.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 30

22.2.5.3 Gas Pool Fires

Application

A fire in a low momentum gas leak will be controlled more by the buoyancy of the combustion
gases than by the momentum of the fuel gas release. It may be represented by the pool fire
model (Section 22.2.3.3) although in reality there will be no pool.

There are two main applications:

• Fires from subsea gas releases


• Fires in gas jets obstructed by dense equipment in a building or module

Fires from Subsea Gas Releases

Subsea gas releases are in effect gas jets obstructed by the water. The water reduces the
momentum of the gas release, and in principle it may behave like a passive release when it
reaches the surface.

When the gas plume reaches the surface, its diameter is estimated to be 22% of the depth of
the release below the water surface (Section 25.3.3). Thus for low release depths or high
release rates, very high fuel flow rates per unit area can be achieved. If this is less than 0.106
kg/m2s, a pool fire model is appropriate (Technica 1990), using a constant surface emissive
power of 153 kW/m2. Otherwise, the gas jet fire model is used (Section 25.5). For a typical
release depth of 50m, this change-over occurs at a release rate of 10 kg/s. Since most release
rates are higher than this, and since the difference between the gas pool and gas jet radiation
fields is small for lower release rates, the gas jet model is appropriate throughout.

Fires in Obstructed Gas Jets

Obstructions to a gas jet reduce its momentum and allow the buoyancy from the hot
combustion gases to bend the jet upwards. A pool fire model represents an extreme version of
this process, which is probably only appropriate if the jet is impinging on a solid wall close to
the point of release. Otherwise it may be appropriate to use the pool or jet model, depending
on which is the most conservative.

22.2.5.4 Confined Fires

A fire inside a module which has a plentiful air supply behaves like an open (fuel-controlled or
well-ventilated) fire.

A fire inside a module which cannot draw in sufficient air to burn all the fuel is known as a
ventilation-controlled (or under-ventilated) fire. If this occurs, unburned fuel will leave the
module with the combustion gases, and burn outside. The size of the external fire is difficult
to predict and only CFD or empirical models are available (Shell Research 1991b, DNVR
1994).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 31

Empirical definitions of the ventilation regime are (DNV Technica 1992):

• For pool fires:


Fuel controlled - air flow rate (kg/s) ≥ 44 x pool dia (m)
Ventilation controlled - air flow ≤ 7 x pool dia (m)

• For jet fires:


Fuel controlled - air flow rate (kg/s) ≥ 160 x fuel flow rate (kg/s)
Ventilation controlled - air flow ≤ 16 x fuel flow

These definitions leave a large range where there is intermediate behaviour.

Confined fires may change behaviour abruptly if the air supply increases, but no models are
available of this effect.

22.2.5.5 Obstructed Pool Fire

A pool fire which is unable to develop its full height because of a solid boundary above it (e.g.
the platform for a pool fire on the sea, or the module roof for a fire inside a module) will
spread out sideways along the obstruction.

The radius of the expanded area where the fire impinges is indicated by the empirical formula
(You & Faeth 1979):

0.96
 H - Hm 
R = 0.5 D  
 D 

where: R = Radius of impinged area on obstacle (m)


D = Diameter of pool (m)
H = Height of flame if unobstructed (m)
Hm = Height of module (m)

The increase in pool fire burning rate due to radiation reflected from the obstruction is
believed to be small (Shell Research 1991b). Due to lack of data, this is normally neglected.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 32

22.3 EXPLOSIONS

22.3.1 Introduction to Gas Explosions


22.3.1.1 Definitions

An explosion is defined as a rapid release of energy which causes a significant pressure pulse
capable of causing damage (I Chem E 1992). A vapour cloud explosion (VCE) (also known
as a gas explosion) is an explosion resulting from the combustion of a cloud of gas. A VCE is
distinguished from a flash fire by the fact that the combustion of the gas gives rise to damaging
levels of overpressure.

22.3.1.2 Explosion Mechanisms

If a stationary flammable gas / air mixture is ignited, a smooth spherical flame is formed. The
flame consumes the unburned gas ahead of it, converting the chemical energy into heat,
forming an expanding sphere of combustion gases. The hot combustion gases occupy a
volume approximately 8 times that of the unburned gas, and their expansion generates
pressure which causes the unburned gas to flow ahead of the flame. At stoichiometric
conditions, when the proportions in the gas / air mixture are those required for complete
combustion, the expansion effect is greatest.

If the flame is laminar (i.e. smooth), the burning velocity (the speed of the flame relative to
the gas) is relatively low. For example, a stoichiometric mixture of methane and air has a
laminar burning velocity of 0.45 m/s, and an expansion ratio of approximately 8, which gives a
flame speed of 3.6 m/s and maximum flow velocities of 3.2 m/s. Such flames are unlikely to
cause any damage.

If the expansion process is confined due to being inside an enclosure, the pressure inside the
enclosure will rise. The theoretical maximum pressure based on the expansion ratio is
approximately 8 bar. In practice, most structures would lose integrity before this value was
reached. Usually, offshore modules have vents which allow the unburned gas to escape and
hence limit the pressure inside.

If the flame becomes turbulent (i.e. varying in speed and direction), the flame front becomes
wrinkled, increasing in area and hence in burning velocity. Turbulence may be caused by the
flow of the gas from the initial release, or by confinement or obstacles in the flame path. If
there is sufficient congestion, this may cause a feedback mechanism as follows:

• Combustion causes pressure


• Pressure causes expansion flow in unburned gases
• Flow around obstacles causes turbulence
• Turbulence increases flame area and speed
• Increased flame area increases combustion

This causes rapid flame acceleration, leading to high overpressures. This may occur whether
or not there is any confinement. The overpressures are very sensitive to the size of the cloud,
the arrangement of obstacles, and the location of the ignition point relative to the obstacles
and the vent.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 33

If unburned gas is vented from the module, this may cause highly turbulent flow which, if
ignited, may generate an external explosion with high overpressure.

The above mechanisms are all types of deflagration. The most damaging explosion
mechanism is detonation, where the combustion wave is supersonic and the shock wave and
combustion wave are coupled. In a fuel / air cloud the detonation wave will propagate at a
velocity of 1500 to 2000 m/s and the peak overpressures is typically 15 to 20 bar. The
likelihood of a deflagration propagating to a detonation depends on the gas, but is fairly low
for methane.

The pressure generated by an explosion is normally referred to as overpressure (i.e. pressure


in excess of ambient). The pressure pulse (or blast wave) generated by an unconfined
explosion has a leading shock front followed by an experimental decay to zero, and then a
longer duration but smaller magnitude negative overpressure phase. In general, the peak
overpressure is the most important parameter. The overpressure pulse due to a semi-confined
explosion is more complex, with several possible peaks associated with venting through the
relief areas.

22.3.1.3 Types of Gas Explosions

The consequences of a gas explosion will depend on the environment in which the gas cloud is
contained. Therefore gas explosions are classified as follows:

• Confined explosions (or confined-vented explosions). These are where the burning gas is
largely confined, typically inside a largely empty enclosed module, an oil tank or a leg of a
concrete platform. The main mechanism of pressure build-up is the expansion of the gas as
it burns, exceeding the vent capacity of the space.

• Semi-confined explosions. These are where the gas is partly confined in a typical offshore
module, but the main mechanism of pressure build-up is turbulence generated by obstacles
such as process equipment in the path of the expanding gas. The Piper Alpha accident is
an example of this type of event. The term semi-confined is used because only partial
confinement is necessary for the explosion to occur.

• Unconfined explosions. These are where the gas cloud is largely unconfined, but there are
sufficient obstacles to generate turbulence and start the build-up of pressure. The term may
therefore be a bit misleading since some degree of confinement or obstruction is necessary
to cause the explosion. This type of explosion has mainly occurred onshore, typically in
chemical plants, but it might occur offshore above the upper deck or the sea surface. These
explosions are often referred to as unconfined vapour cloud explosion (UVCE).

Strictly, all these are types of vapour cloud explosion, but in practice the term VCE is often
reserved for explosions other than confined ones, and sometimes for unconfined explosions
alone.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 34

22.3.2 Confined Explosions


22.3.2.1 Application

Confined explosion models outlined below are typically applied to explosions inside enclosed
areas, in fuel storage tanks, process equipment or in legs of concrete platforms. In principle
they can be applied to explosions within other compartments with solid walls, but experience
in this is limited.

22.3.2.2 Probabilities of Confined Explosions

A gas / air mixture within its flammable range inside a confined space will explode if an
ignition source of sufficient strength is applied. The frequency of such an explosion can
therefore be found from:

• The frequency of a flammable mixture forming;


• The probability of a suitable ignition source being present.

These can be calculated using fault tree analysis. For example, this might address the
frequency of a fuel storage tank’s inert gas system failing, and hot work taking place without
adequate gas detection. The probability of a given gas leak forming a flammable atmosphere
in a confined space with a known ventilation rate can be determined from gas concentration
models.

Alternatively, the frequency of explosions can be obtained directly from historical data such as
on oil tanker ships.

22.3.2.3 Overpressures in Confined Explosions

For most hydrocarbon fuels the theoretical maximum overpressure, for a stoichiometric fuel /
ar cloud at initial conditions of 25°C of 1 atm, in a confined explosion is 8 to 10 bar (CMR
1993). In practice, most structures would lose integrity before this value was reached. The
actual overpressure at the source can then only be determined from an analysis of the strength
of the structure against blast loading.

Much of the energy in a confined explosion is absorbed in deforming the structure, and some
is converted to kinetic energy of projectiles, so the remaining energy in the blast wave is often
rather small. Damage caused by the external blast wave is therefore commonly neglected.

22.3.2.4 Projectiles from Confined Explosions

Projectiles (or missiles) are the main external hazard from confined explosions. Many different
projectile effects may occur. Simple design guidelines could be used to predict the hazard
zones and risk levels (SCI 1991).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 35

22.3.3 Semi-Confined Explosions


22.3.3.1 Introduction

Semi-confined explosions are explosions where the gas cloud is partly confined. The main
mechanism of pressure build-up is turbulence generated by obstacles in the path of the
expanding gas. This type of explosion represents the main explosion risk on a typical offshore
installation. The recent Joint Industry Project - Blast and Fire Engineering for Topside
Structures has greatly improved the industry’s understanding of semi-confined explosions.

22.3.3.2 Joint Industry Project - Blast and Fire Engineering for Topside Structures

The Blast and Fire Engineering Project for Topside Structures is one of the largest research
projects undertaken following the Piper Alpha disaster. The ultimate objective of this joint
industry project was to improve the industry’s confidence in its safety systems with respect to
hydrocarbon explosions and fires.

As of January 1998, two phases of the project have been completed:

• Phase 1: Literature review to gather up-to-date knowledge on the causes and effects of
fires and explosions offshore.

• Phase 2: Full-scale fire and explosion experiments were carried out using a rig designed to
represent the construction and equipment layout of a typical module on an offshore
installation. This data was used to evaluate the accuracy of fire, explosion and structural
response models and to provide a basis for design guidance.

Phase 2 of the project served to determine the ability of current explosion models to
characterise explosion scenarios. Twelve models participated in the exercise.

The initial explosion testing revealed that the majority of models were underpredicting the
overpressures, sometimes by an order of magnitude, with a few overpredicting by a factor of
over 2. It seemed that different models differed widely in their predictions of the same
explosion scenario. In addition for some models there wasn’t any consistency, with the
predictions grossly underpredicting for some tests and overpredicting for other tests. This
initial batch of testing resulted in several of the models being modified taking account of the
information produced from the testing.

Further testing showed greater agreement between the predicted and experimental results.
The difference between predicted and experimental results was generally below a factor of
two. Predictions of overpressure rise time and duration showed a similar agreement between
predicted and experimental results.

The models considered to have most accurately predicted the experimental results are the CFD
models FLACS, EXSIM and AutoReaGas. DNV’s empirical model COMEX+NVBANG also
performed well, although it was not able to produce the same amount of output results as the
CFD models.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 36

22.3.3.3 Semi-Confined Explosion Modelling

COMEX and NVBANG

COMEX is an empirical model used for the calculation of maximum explosion overpressure
pressure for partially confined modules containing flammable gas. The model was developed
by DNV based on a series of scaled experiments carried out between 1978 and 1983 and has
been updated after the Joint Industry Project - Blast and Fire Engineering for Topside
Structures.

NVBANG is the DNV model that is normally used in conjunction with COMEX for the
calculation of the explosion overpressure for vented deflagrations as a function of time.
Combustion (deflagration) is considered to be a quasi-static process with pressure equalisation
within the confinement at any time and is modelled in a series of small time steps. A number
of thermodynamic processes are performed within each time step.

FLACS

FLACS is a complete dedicated three-dimensional computational fluid dynamics (CFD)


ventilation, gas dispersion and explosion simulation tool. The simulator is produced,
distributed, supported and maintained by Christian Michelsen Research (CMR) and
development is currently sponsored by bp, TotalFinaElf, Exxon, Gaz de France, Mobil, Norsk
Hydro, Phillips, Statoil, Norwegian authorities (NPD) and British authorities (HSE).

The model covers such phenomena as gas leakage and dispersion, explosion (deflagration),
blast waves, HVAC and natural ventilation, waterspray and the effect of inert gas. Inputs to
the FLACS are geometrical information, scenario information (i.e. gas cloud position, size and
composition, ignition location or, if the dispersion facility is used information about leak size,
direction and position, and wind characteristics) and size of the simulation volume and grid
size. FLACS solves the full Navier - Stokes equations in the explosion region. It contains a
combustion model that consists of two parts, a burning velocity model and a flame model. A
k-ε is turbulence model is used to calculate the turbulence field. The outputs generated are
scalar-time curves of a large number of parameters such as local pressure, pressure impulse,
drag force, average wall pressure and approximately forty five others. The simulator also
produces two- and three-dimensional plots of approximately fifty variables such as combustion
products, gas cloud, pressure, velocities and turbulence parameters.

22.3.3.4 Semi-Confined Explosions in Quantitative Risk Analysis

An explosion within a certain module can result in a considerable range of overpressures,


depending on the size of the gas could, the stoichiometry, the degree of mixing, the location of
the ignition source and the strength of the ignition source. The size of the gas cloud itself,
also depends on a large number of independent variables, such as leak size, leak location, leak
direction, wind speed and wind direction.

In the context of a QRA for an offshore installation it is impossible to perform detailed


dispersion calculations and detailed explosion overpressure calculations for all the possible
values of these variables. Two methods to deal with this problem are discussed below.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 37

QRA Using Representative Scenarios

For this method a limited number of representative explosion scenarios need to be identified.
These scenarios should represent the complete range of explosion characteristics resulting
from variations in the above mentioned variables.

The range of explosion scenarios, resulting from varying stoichiometries, degrees of mixing
and strengths of the ignition sources, are typically represented by the worst case scenario, i.e.
uniformly mixed, stoichiometric clouds and high strength ignition sources. The range of cloud
sizes would be represented be a limited number of cloud sizes, typically 100%, 50% and 25%
of the maximum possible could size. For location of ignition sources the module volume is
divided in a few areas with similar explosion characteristics. The worst ignition location
within each of these areas are chosen to be representative.

For this limited number of representative scenarios a detailed explosion consequence analysis
is carried out, using FLACS or a similar software package.

An impact analysis needs to be carried out. The impact analysis will analyse the ability of
various critical structures (such as blastwalls) to withstand explosions.

Finally the impact analysis and the explosion analysis need to be combined, the probability of
each of the potential outcomes needs to be determined and the results need to be included in
the QRA’s event trees.

QRA Using a Probabilistic Explosion Overpressure Model

Simple Approach

For this method the maximum explosion overpressure is determined from a empirical
explosion model, using COMEX + NVBANG or a similar software package.

The probability of exceeding any of the potential explosion overpressures is estimated using a
probabilistic explosion overpressure model, such as DNV’s PROEXP model, which is
included in Neptune Offshore. This model takes the following variables into account when
estimating the explosion overpressure exceedance probability distribution: leak rate,
ventilation rate, module volume, ignition time and location of the ignition source, ESD and
blowdown delay time, relative explosion pressures for semi-filled modules.

An impact analysis also needs to be carried out.

Finally the impact analysis and the probabilistic explosion analysis need to be combined and
included in the QRA’s event trees.

Detailed Approach

For this method a limited number of carefully selected scenarios are modelled with CFD
codes. The results are used to determine response surfaces which describe gas cloud
dispersion and explosion overpressures as a function of a large number of independent
variables. These response surfaces are used in a probabilistic analysis to determine the
explosion overpressure exceedance probability.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 38

An impact analysis also needs to be carried out.

Finally the impact analysis and the probabilistic explosion analysis need to be combined and
included in the QRA’s event trees.

22.3.4 Unconfined Explosions


22.3.4.1 Application

Unconfined explosion models outlined below are applied to largely unconfined clouds, which
might occur outside a platform.

22.3.4.2 Probabilities of Unconfined Explosions

A hydrocarbon gas / air mixture is only likely to explode with significant overpressure if all the
following conditions are satisfied:

• The hydrocarbon gas / air mixture is within its flammable limits and an ignition source is
applied.

• A high degree of congestion from turbulence-creating obstacles.

• A large area, allowing the flame front to accelerate to high velocities.

• A large gas cloud, also needed for flame front acceleration.

These conditions are less likely to occur to gas clouds in open areas, unless these are ignited
after venting from internal explosions.

Probability data on explosions offshore could be applied to unconfined explosions as well as to


semi-confined explosions. The data on blowouts is probably the most appropriate, as many of
these would have been unconfined gas clouds. Explosion probabilities are considered in
Module 29, Section 29.7.

Experiments on methane / air mixtures have indicated great difficulty in producing unconfined
explosions (except by venting from internal explosions), and on this basis some studies of
natural gas have neglected the unconfined explosion probability.

22.3.4.3 Overpressures in Unconfined Explosions

An explosion model calculates the distances to various orverpressure or damage levels


resulting from the explosion of a vapour cloud. These distances are the radii of the circles, see
Figure 22.15, within which the damage resulting is defined by the analyst as part of the
parameter input.

TNO Vapour Cloud Explosion Model

In the TNO correlation model the total energy content of the cloud is calculated based on the
output from the dispersion model and data in the material data base. The working range of the
model is (between 100 kg and 100 tonnes of hydrocarbons). If the cloud contains less than

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 39

100 kg the overpressure level is expected to be too small to cause an serious damage. Very
little is known about chlorides above 100 tonnes so this is taken as the upper limit.

A yield factor is applied in the calculation (as the TNT equivalent model explained below)
since only a small part of the total combustion energy in an explosive can be available for
shock wave propagation. From this the radius to various damage levels can be calculated; no
explicit predictions of overpressures are made although sometimes a rough equivalence
between damage levels (e.g. ‘heavy building damage, R1’ and overpressure (of the order of
0.35 bar) is made.

Figure 22.15 Example of Explosion Effect Area

Flammable Cloud
Centre of Cloud and
Explosion

Release
Source

Explosion R1 Effect Area


Explosion R2 Effect Area

TNT Equivalence Model

Another (well known but now rather dated) approach to calculating overpressures from an
explosion is to convert the mass of vapour in the cloud to an equivalent TNT mass. Extensive
research has been undertaken on TNT explosion where overpressure levels have been
increased with various quantities of TNT. Hence by being able to convert the explosive
potential of a vapour cloud to a known quantity consequence distances can be easily
calculated. The TNT method gives reasonable predictions of overpressures outside the cloud
for relatively reactive materials, but it is sensitive to the choice of parameters. Within the gas
cloud, the TNT method overestimates the overpressure, and constant values are normally
assumed instead. For propane, a value of 1 bar within the cloud might be appropriate.
Experiments on methane explosions in unconfined areas have indicated a maximum
overpressure of about 0.2 bar (van Wingerden et al. 1989). This can be used as a constant
peak VCE overpressure.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 40

Multi-Energy Explosion Model

The Multi-Energy Explosion method (van Wingerden et al. 1989) treats a vapour cloud
explosion as a number of sub-explosions. These sub-explosions are vented on parts of the
cloud that are either in intensely turbulent motion or are partially confined or obstructed.
Sources of sub-explosions are identified and assigned a blast strength between 1 (lowest) and
10 (highest). Each sub-explosion is modelled separately and the results of each blast are
analysed separately (they cannot be added together). Unconditioned parts of the cloud do not
detonate; the sub-explosions may be detonations or deflagrations but are most likely to be the
latter as there is usually insufficient confinement on a process plant for detonation to occur.

Examples of potential centres of strong sub-explosions are:

• High velocity jets releasing fuel at high pressure as a result of a tube or vessel leak.

• Densely configured objects such as:


− Process equipment
− Pipe racks
− Piles of crates or drums

• Spaces between long parallel planes such as:


− Concrete platforms carrying process equipment
− Beneath cluster of parked cars
− Open multi-story buildings like parking garages

• Within tube-like structures such as:


− Tunnels
− Bridges
− Corridors
− Sewage systems
− Culverts

Output from the Multi-Energy Explosion Model consists of the potential explosive power of
the vapour cloud, represented by the combustion energy of each charge, and the over-pressure
and pulse duration at user-selected distances from each blast source.

22.3.5 Detonations
22.3.5.1 Application

In principle, any type of VCE can undergo a transformation from deflagration to detonation,
with a corresponding large increase in overpressure. Detonations can occur in confined
volumes such as pipes, or relatively open situations such as pipe racks, and also when jet
flames are shot out from a confined volume to an unconfined cloud.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 41

22.3.5.2 Probabilities of Detonations

The probability of a VCE developing into a detonation is particularly difficult to quantify, as


no theory exists to predict the conditions when such a transition might occur. There is no
experience of such events offshore, except in confinement such as pipes.

The probability is greatest for reactive fuels, such as hydrogen, acetylene or ethylene, and in
oxygen-enriched atmospheres. For natural gas consisting mainly of methane, detonation is
very unlikely, and the scenario is often discarded in offshore QRA. For propane, detonations
have been observed, and hence detonations of gas from condensate might be possible.

The probability is greatest for large clouds with a high degree of confinement or high density
of obstructions. Offshore modules are normally designed to minimise explosion probabilities,
and such design practices may be considered to produce negligible probabilities of detonation.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 42

22.4 CONFINED AND OBSTRUCTED FIRES

22.4.1 General Description


The models of jet fires, pool fires, flash fires and fireballs outlined above are all based on
experiments on gas and oil releases in open environments. There is much less knowledge about
fires whose development is limited by the surrounding structure. Research on fires inside
rooms has mainly addressed cellulosic fires, and is not readily applicable to hydrocarbon fires.
Research on gas explosions inside offshore modules (see below) has concentrated on
predicting the overpressure which results, and gives no information on the size of the burning
gas cloud. Available data on confined and obstructed fires is presented by Shell Research
(1991b), but is difficult to generalise.

Obstructions such as process equipment, decks, firewalls conductors etc. may distort gas and
oil fires from the size and shapes in the unobstructed case. Where the obstructions are located
in the radiation field but not the flame, they may be modelled by simple allowance for shielding
(e.g. Section 25.5.4). Where the obstructions are within the flame itself, their effect may be
substantial, and they may make the fire's effects worse, e.g. by diverting it towards an escape
route. Therefore, some form of modelling is desirable.

22.4.2 Modelling Approaches


The ability to predict the effects of confinement and obstructions on large fires in buildings and
offshore installations is limited at present. Possible modelling approaches are:

• Physical models. These give the most realistic representation of the effects of obstructions.
Several tests have been carried out on equipment in impinging jet fires, but no analysis is
available of the effects of the equipment on the jet shape.

• Numerical models. CFD models have been used to represent impinging fires. They are
able to model the flow of hot combustion gases, and give a good representation of the
shape of the fire, but do not at present predict the radiation field around it.

• Phenomenological Models. Theoretical models of the combustion within confined spaces


have been developed, but are not yet validated or available commercially.

In the absence of an exhaustive treatment of the subject, several individual fire scenarios are
described below for which an empirical approach is available.

22.4.3 Gas Pool Fires


22.4.3.1 Application

A fire in a low momentum gas leak will be controlled more by the buoyancy of the combustion
gases than by the momentum of the fuel gas release. It may be represented by the pool fire
model (Section 25.6), although in reality there will be no pool.

There are two main applications:

• Fires from subsea gas releases

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 43

• Fires in gas jets obstructed by dense equipment in a building or module

22.4.3.2 Fires from Subsea Gas Releases

Subsea gas releases are in effect gas jets obstructed by the water. The water reduces the
momentum of the gas release, and in principle it may behave like a passive release when it
reaches the surface.

When the gas plume reaches the surface, its diameter is estimated to be 20% of the depth of
the release below the water surface (Section 25.3.3). Thus for low release depths or high
release rates, very high fuel flow rates per unit area can be achieved. If this is less than 0.106
kg/m2s, a pool fire model is appropriate (Technica 1990), using a constant surface emissive
power of 153 kW/m2. Otherwise, the gas jet fire model is used (Section 25.5). For a typical
release depth of 50m, this change-over occurs at a release rate of 10 kg/s. Since most release
rates are higher than this, and since the difference between the gas pool and gas jet radiation
fields is small for lower release rates, the gas jet model is appropriate throughout.

22.4.3.3 Fires in Obstructed Gas Jets

Obstructions to a gas jet reduce its momentum and allow the buoyancy from the hot
combustion gases to bend the jet upwards. A pool fire model represents an extreme version of
this process, which is probably only appropriate if the jet is impinging on a solid wall close to
the point of release. Otherwise it may be appropriate to use the pool or jet model, depending
on which is the most conservative.

22.4.4 Confined Fires


A fire inside a module which has a plentiful air supply behaves like an open (fuel-controlled or
well-ventilated) fire.

A fire inside a module which cannot draw in sufficient air to burn all the fuel is known as a
ventilation-controlled (or under-ventilated) fire. If this occurs, unburned fuel will leave the
module with the combustion gases, and burn outside. The size of the external fire is difficult to
predict and only CFD or empirical models are available (Shell Research 1991b, DNVR 1994).

Empirical definitions of the ventilation regime are (DNV Technica 1992):

• For pool fires:

Fuel controlled - air flow rate (kg/s) • 44 x pool dia (m)

Ventilation controlled - air flow • 7 x pool dia (m)

• For jet fires:

Fuel controlled - air flow rate (kg/s) • 160 x fuel flow rate (kg/s)

Ventilation controlled - air flow • 16 x fuel flow

These definitions leave a large range where there is intermediate behaviour.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 44

Confined fires may change behaviour abruptly if the air supply increases, but no models are
available of this effect.

22.4.5 Obstructed Pool Fire


A pool fire which is unable to develop its full height because of a solid boundary above it (e.g.
the platform for a pool fire on the sea, or the module roof for a fire inside a module) will
spread out sideways along the obstruction.

The radius of the expanded area where the fire impinges is indicated by the empirical formula
(You & Faeth 1979):
0.96
R = 0.5 D 
H - Hm 

 D 

where: R = Radius of impinged area on obstacle (m)


D = Diameter of pool (m)
H = Height of flame if unobstructed (m)
Hm = Height of module (m)

The increase in pool fire burning rate due to radiation reflected from the obstruction is
believed to be small (Shell Research 1991b). Due to lack of data, this is normally neglected.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 45

22.5 SMOKE GENERATION

22.5.1 Application
Smoke from burning hydrocarbons consists of a mixture of gaseous combustion products and
unburned carbon particles. The term smoke is used here to describe the mixture, and soot to
describe the solid particles. Note that some sources use the term smoke to describe the solid
particles alone.

Smoke is generated by any burning hydrocarbon, but in an offshore QRA, where smoke
hazards are most relevant, it is most significant when produced by burning liquids such as
crude oil.

The nature of the smoke depends strongly on the supply of oxygen to the fire. Two basic types
are distinguished:

• Well-ventilated (or fuel-controlled) fires, which occur in the open, such as on the ground,
on an open deck or on the sea surface. These fires have sufficient air to achieve complete
combustion.

• Under-ventilated (or ventilation-controlled) fires, which occur inside buildings or modules.


These fires are unable to draw sufficient air to achieve ideal combustion.

22.5.2 Accident Experience


In most onshore and offshore fires, the main effects have been due to heat, but in a few events
the effect of smoke has been significant, and these few events have resulted in many fatalities.
The most important was Piper Alpha. Following an explosion, an oil pool fire generated a
large plume of smoke which engulfed the accommodation. Smoke inhalation caused at least
109 of the 167 fatalities.

Prior to Piper Alpha, there had been very little study of smoke in hydrocarbon fires. Since
then several experimental studies of smoke generation have been carried out, most with oil
pools less than 2 m diameter.

22.5.3 Smoke Generation


For most hydrocarbon fuels, the stoichiometric ratio is 15 kg air per kg of fuel. Thus, in well-
ventilated fires with sufficient air for complete combustion, the smoke production rate is 16
times the fuel burning rate.

The temperature of the generated smoke may be up to 1200oC, although it may also be much
less. Worst-case conditions may in fact be when the temperature is lower, causing the plume
to rise less. Theoretical combustion calculations for fires inside modules have indicated smoke
outlet temperatures of 400-700oC (SINTEF 1991). Other data suggests temperatures of
1000oC for smoke from well-ventilated fires and 600oC for under-ventilated fires (DNVR
1994).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 46

22.5.4 Smoke Composition


22.5.4.1 Combustion Gases

Combustion gases from burning hydrocarbons consist mainly of a mixture of nitrogen, carbon
dioxide (CO2), carbon monoxide (CO) and water vapour. Because of the high proportion of
nitrogen in the air, hot nitrogen tends to dominate in the combustion products.

The concentration of toxic gases in the smoke is important because of their potential impact
on personnel. For modelling purposes, it is necessary to know the initial concentration close to
the fire. The dilution in the smoke plume can then be predicted theoretically, to determine the
concentration at any point downwind.

The proportion of toxic gases in the smoke depends on the chemical structure of the burning
materials and the degree of ventilation to the fire. The differences between different
hydrocarbons are quite small, and ventilation has the main effect. In general, reduced
ventilation (e.g. for a fire inside a compartment) greatly increases the ratio of CO to CO2 in the
combustion gases. Typical gas concentrations close to the fire are given in Table 22.10 (Bonn
1993). In each case the oxygen concentration is zero.

Table 22.10 Initial Gas Concentrations In Smoke

Gas Concentrations In Smoke (Vol%)


Well Ventilated Fire Under Ventilated Fire
Gas Fire Liquid Fire Gas Fire Liquid Fire
CO 0.04 0.08 3.0 3.1
CO2 10.9 11.8 8.2 9.2

The two ventilation regimes are defined empirically as defined in Section 22.4.4, with linear
interpolation in the intermediate region.

In principle, the gas composition can be calculated from theoretical combustion equations for
hydrocarbons. In practice, there are too many different products to solve the equations
without input from experimental measurements. This approach appears to be less accurate
than using experimental measurements directly (DNVR 1994).

Empirical correlations giving the concentration of CO, CO2 and O2 as functions of the air/fuel
ratio are given by SINTEF (1991). However, in most QRA studies, the air flow to the fire is
unknown.

22.5.4.2 Soot

The solid carbon particles resulting from a hydrocarbon fire, which are entrained in the
combustion gases, are described here as soot. Soot is important because of its impact on
visibility.

The proportion of soot in the smoke depends on the fuel type and the degree of ventilation to
the fire. Soot yields are measured in terms of kg soot produced per kg fuel burned. Traditional
building materials and furniture have soot yields in the range 0.01 - 0.02. Hydrocarbon fuels
have much higher soot yields.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 47

Soot yields from hydrocarbon fuels have been reported as follows (SINTEF 1991):

• In under-ventilated fires, soot yield 0.07 - 0.1 kg/kg fuel for all liquid hydrocarbon fuels.
The value of 0.1 is appropriate for offshore QRA.

• In well-ventilated fires, soot yield was still 0.07 - 0.1 kg/kg fuel for heavy fuels (boiling
point 140 - 165oC), but only 0.02 for lighter fuels (boiling point 62 - 82oC).

Mass fractions of soot in the smoke may be estimated from these:

fm = Y/16

where: fm = mass fraction of soot in smoke (kg/kg)


Y = yield of soot (kg/kg fuel)

For smoke yields in the range 0.02 to 0.1, this gives mass fractions in the range 1.3 x 10-3 to
6.3 x 10-3 kg soot per kg smoke.

22.5.4.3 Overall Smoke Composition

The smoke consists of an aerosol of solid soot particles entrained in the hot combustion gases.
The relative proportions in smoke from 1kg of heavy oil are illustrated in Figure 22.16.

The density of the smoke is approximately that of the combustion gases. Assuming they are an
ideal gas with the same density as air, the density of the hot smoke is:

Ta
ρ smoke = ρ a
Tsmoke

where: ρsmoke = smoke density (kg/m3) at actual temperature


ρa = air density at ambient temperature, 1.19 kg/m3
Tsmoke = smoke temperature (K)
Ta = ambient temperature, 293 K

Assuming a smoke temperature of 600oC (873K), the smoke density is 0.4 kg/m3.

The density of soot is assumed to be 2250 kg/m3 as for solid carbon. The mass density of soot
in the smoke is then:

m = fm ρsmoke

or:

m = ρsmoke . Y/16

For smoke yields in the range 0.02 to 0.1, this gives densities of soot in the smoke in the range
5 × 10-4 to 2.5 × 10-3 kg/m3. An alternative source (Wighus et al. 1992) gives an average value
of 1.1 × 10-3 kg/m3.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 48

Figure 22.16 Smoke Production

AIR FUEL

M =15kg M =1kg
T =293K T =293K
ρ =1.19kg/m3 ρ =800kg/m3
V =12.6m3 V =0.001m3

SMOKE

M =16kg
T =873K
ρ =0.4kg/m3
V =40m3

COMBUSTION GAS SOOT

M =15.9kg M =0.1kg
T =873K T =973K
ρ =0.4kg/m3 ρ =2250kg/m3
V =40m3 V =4x10-5m3

The volume concentration of soot in the smoke is:

ρ smoke
fv = fm
ρ soot

For smoke yields in the range 0.02 to 0.1, this gives volume concentrations of soot in the
smoke in the range 0.2 to 1.1 ppm.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 49

22.6 REFERENCES
American Gas Association (174), LNG Safety Research Programme, Report IS-3-1.

API (1982), Guide for Pressure Relieving and Depressuring Systems, API RP521, 2nd
Edition.

Bonn (1993), Smoke and Gas Detectors for Temporary Refuges, 4th International Conference
on Offshore Loss Prevention, Aberdeen.

Burgess, D.S. and Hertzberg, M. (1974), Advances in Thermal Engineering, Ch. 27, p413,
John Wiley and Sons.

Chamberlain, G.A. (1987), Developments in Design Methods for Predicting Thermal


Radiation from Flares, Chem. Eng. Res. Des, 65, pp.299-309.

CMR, 1993. Gas Explosion Handbook, Christian Michelsen Research, Bergen, Norway.

Crossthwaite, P.J., Fitzpatrick, F.D., and Hurst, N.W. (1988) Risk Assessment for the Siting
of Developments near Liquefied Petroleum Gas Installations, IChemE Symp, 110,
pp.373-400.

DNVR (1994), Review of DNV Models for Fire and Smoke Risk Assessment, DNV Research
93-2068.

DNV Technica (1992), QRA Studies for Magnus, Clyde and Thistle, C3377A, August 1992 -
February 1993

IChemE (1989), Calculation of the Intensity of Thermal Radiation from Large Fires, IChemE
monograph, also published in Loss Prevention Bulletin 082, August 1988.

IChemE (1992), Major Hazards Onshore and Offshore, Proceedings of a Symposium, 20-22
Oct 1992, Manchester, IChemE Symposium Series no. 130, Institution of Chemical Engineers
(IChemE), Rugby.

Kinsman, P. (1991), Major Hazard Assessment: A Survey of Current Methodology and


Information Sources, Specialist Inspector Report No. 29, Health and Safety Executive (HSE).

Moorhouse, J and Pritchard, M J (1982), Thermal Radiation Hazards from Large Pool Fires and
Fireballs - A Literature Review, Symposium on the Assessment of Major Hazards, I.Chem.E,
Symp. Series 71.

Mudan, K.S. and Croce, P.A. (1988), Fire Hazard Calculations for Large Open Hydrocarbon
Fires, Ch. 2-4, Handbook of Fire Protection Engineering, Society of Fire Protection
Engineers.

NIOSH (1994), NIOSH Pocket Guide to Chemical Hazards, US Dept. of Health and Human
Services, National Institute for Occupational Safety and Health.

Quintiere, J.G. (1991), Pool Fires - A Review, Appendix B of Shell Research (1991b).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 22: Consequence Analysis (3) – Effect Modelling Page 50

Raj, P.P.K. (1977), Calculation of Thermal Radiation Hazards from LNG Fires. A Review of
the State-of-the-Art, AGA Transmission Conference.

Roberts, A. F. (1982), Thermal Radiation Hazards from Releases of LPG from Pressurised
Storage, Journal of Fire Safety Studies 4, pp.197-212.

Scandpower (1992), Handbook for Fire Calculations and Fire Risk Assessment in the Process
Industry, SINTEF-NBL, Kjeller.

SCI (1991), Blast and Fire Engineering Project for Topside Structures, Steel Construction
Institute, Ascot, UK.

Shell Research (1991a), Oil and Gas Fires - Characteristics and Impact, Work Package FL1 of
SCI.

Shell Research (1991b), Behaviour of Oil and Gas Fires in the Presence of Confinement and
Obstacles, Work Package FL2 of SCI.

SINTEF (1991), Modelling of Hydrocarbon Fires Offshore, Final Report, SINTEF Report
STF25 A91029, Trondheim, Norway.

SINTEF (1992), Fire on the Sea Surface - State of the Art and the Need for Future Research,
SINTEF Report STF25 A92035, Trondheim, Norway.

Thomas, P.H. (1963) The Size of Flames from Natural Fires, 9th Intl. Combustion
Symposium, Combs Inst. Pittsburgh, PA, pp.844-859.

van Wingerden, C.J.M., van den Berg, A.C., Opschoor, G. (1989), Vapor Cloud Explosion
Blast Prediction, Plant/Operations Progress, 8, 4, pp.234-238, October.

Wertenbach, H.G. (1971), Spread of Flames on Cylindrical Tanks for Hydrocarbon Fluids,
Gas and Erdgas 112 (8).

Wighus, R., Meland, Ø. and Vembe, B. (1992), Smoke Hazard in Offshore Platform Fires,
SINTEF Report STF25 A91007.

You, H.Z. and Faeth, G.M. (1979), Ceiling Heat Transfer During Fire Plume and Fire
Impingement, Fire and Materials, Vol 3, pp 140-147.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module22.doc ©2004 DNV, All Rights Reserved

You might also like