ITSA

The QRAQ Project

Volume 7,
Probability of Ignition

Version 1 Issue 1
February 2012
J.R.Taylor
 QRAQ 7 Ignition Probability

© J.R.Taylor 2012
2010 i
i
 QRAQ 7 Ignition Probability

The QRAQ project

Quality of Risk Assessment for Process Plant
Taylor Safety Engineering,
Prunusvej 39,
3450 Allerød,
Denmark
Issue Date Author Approval Release
V1I1 Oct 04 JRT
V2I1 Aug 12 JRT

© J.R.Taylor 2012
2010 ii
ii
 QRAQ 7 Ignition Probability

© J.R.Taylor 2012
2010 iii
iii
 QRAQ 7 Ignition Probability

Preface
This report is a systematic examination of the models available for calculation of ignition
probability in the case of release of flammable materials within a process plant, and the
impact of this on risk assessment.

J.R.Taylor
Allerød 2004

© J.R.Taylor 2012
2010 iv
iv
 QRAQ 7 Ignition Probability

© J.R.Taylor 2012
2010 v
v
 QRAQ 7 Ignition Probability

QRAQ publications
1. The QRAQ Project – Introduction
2. Quality and completeness of hazard identification
3. Consequence calculation models
4. Risk assessment frequency data
5. Risk analysis methodologies
6. Risk acceptance criteria
7. Ignition frequency
8. Jet fire models
9. Fire water monitors as a risk reduction measure
10. Boilover and fire induced tank explosion
11. Self evacuation as a risk reduction measure
12. Major hazards scenarios - Model validation against actual accidents
13. Pool fire models
14. H2S classified areas red zones
15. Domino effects
16. Gas impoundment
17. Gas dispersion mapping for gas detector placement
18. In preparation
19. In preparation
20. Human error in process plant operations and maintenance
21. SIL assessment using LOPA
22. Assessment of simultaneous operations

© J.R.Taylor 2012
2010 vi
vi
 QRAQ 7 Ignition Probability

Updating history

Issue Date Affected Change

Initial version Oct 2004 Initial release
Update Aug 2012 Evaluation of Joint Industry
Project model added
Evaluation of model accuracy
and benefit from more advanced
models

© J.R.Taylor 2012
2010 vii
vii
 QRAQ 7 Ignition Probability

© J.R.Taylor 2012
2010 viii
viii
 QRAQ 7 Ignition Probability

Contents

1. Introduction ........................................................................................................................1
2. Methodology..................................................................... Error! Bookmark not defined.
2.1 Scoping analyses ....................................................... Error! Bookmark not defined.
2.2 Location of releases ................................................... Error! Bookmark not defined.
2.3 Exposure times .......................................................... Error! Bookmark not defined.
2.4 Calculation methodology .......................................... Error! Bookmark not defined.
3. Release frequencies .......................................................... Error! Bookmark not defined.
4. Modelling.......................................................................... Error! Bookmark not defined.
5. Risk to Persons ................................................................. Error! Bookmark not defined.
6. Other Considerations ........................................................ Error! Bookmark not defined.

© J.R.Taylor 2012
2010 ix
ix
 QRAQ 7 Ignition Probability

1. Introduction
For flammable releases, the overall frequency of an accident is determined by the frequency
of the release, multiplied by the probability of ignition. The determination of ignition
probability is therefore just as important as the determination of release frequency.

Three approaches have been used in risk analyses for determination of ignition probability.

1. Use of average ignition probability values for a whole plant (ref. 1)

2. Methods, which count ignition sources inside the release area. (ref. 2)
3. Parametric models, relating release size (release rate of flow) to ignition probability
(ref. 3)

Some of the published methods are described in the following sections. Then the methods are
compared with actual release data.

Some 150 causes of ignition have been identified for flammable releases (ref. 4 ). Ideally in
keeping with the principles of this study, the risk analysis methods should reflect all of these
sources, in order to provide a motivation for risk reduction. This is a difficult objective for
any method. The algorithm in section x of this report aims at taking a fairly wide range of
factors into account.

1.1 Average ignition probability value methods.

The simplest of the ignition probability analysis method is to allocate a single value for all
releases. Xx suggested a value of 0.1, which experience has shown, is quite typical for
medium size releases in chemical process plants.

An approach which is sensitive to the size of releases is given in the Dutch Purple Book, with
values for ignition frequency dependent on release size (table 1.1)

Source Substance
Continuous Instantaneous K1 liquid Gas, low Gas average to
reactivity high reactivity
<10 kg/s < 1000 kg 0.065 0.02 0.02
10 – 100 kg/s 1000 – 10000 kg 0.065 0.04 0.5
> 100 kg/s > 10000 kg 0.065 0.09 0.7

Table 1.1 Process plant ignition sources

Source Probability
Road tanker continuous release 0.1
Road tanker instantaneous release 0.4
Tank wagon continuous 0.1
Tank wagon instantaneous 0.8
Table 1.2 Transport ignition sources

© J.R.Taylor 2012
2010 1
1
 QRAQ 7 Ignition Probability

E&P Forum reviewed ignition data for oil production platforms, with values for ignition
probability varying from 0.07 to 0.3 (for blowouts). (ref. 5)

The E&P Forum review also includes platform data with ignition probabilities depending on
release size.

1.2 Flow rate based ignition models

One of the best documented models for ignition probability calculation is that due to Cox,
Lees and Ang. in classification of Hazardous Locations (IChem E 1990). The model relates
ignition probability to flow rates in the release. The model can be described by the formula

P = A + FB

Where P is the probability, F is the flow rate and A and B are constants. Values for the
constants are:

A B
Gas leaks, observed values 0.017 0.74
Ignition due to failure to control ignition sources 0.006 0.77
Ignition probability with no control of ignition sources 0.074 0.28
Ignition probability due to self ignition 0.29 0.003

The Joint Industry Project (ref. DNV, Scandpower, AEA, Cowi and CMR) developed a
model for ignition on offshore installations which is based in part on counting of ignition
sources, (see next section) and in part on the release size and duration.

1.3 Discrete ignition source models

The approach in ignition source counting is to actually map the ignition sources in the area
where a release may take place and then to actually identify these, multiply by a probability
of ignition for each, and combine all of the values. A simple approach for combining these is
simply to add all the contributions. A more accurate approach is to track the spread of a gas
or vapour plume or cloud with time, and to add contributions for each source i as:

Pan = Πn-1i=1 (1-Pi) Pn

Where Pin is the actual probability of ignition for source n and Pn is the inherent probability
of ignition by source n in the absence of other ignition sources.

This approach has the advantage that it allows a time development of the release to be
calculated, and a differentiation of fire and explosion sizes depending on whether ignition is
early or delayed.

The approach also avoids the problem of the ignition frequency exceeding 1.0 which can
happen when ignition source contributions are simply added.

© J.R.Taylor 2012
2010 2
2
 QRAQ 7 Ignition Probability

The first such approach to this was developed for the IFAL calculation method (ref. 2 ) in the
late 1970´s. The values used were based on actual fire statistics from 90 refineries, and on
experience from detailed studies at two refineries concerning the sources of ignition. Since
then, the present author has refined the data, tracking the fire history of some 30 plants over
20 years. Table 1.4 gives the basis values updated from the last publication in 1994 (ref. 6 )

Motors, switch gear, pumps (not flame-proof) 10*10-3

Switch gear, unprotected which is switched off during exposure 0.5
Electrical sub-station 100*10-3
Analysers (not overpressured from a safe location) 10*10-3
Control rooms, offices laboratories (not overpressured) 50*10-3
Fans 50*10-3
Operators 1*10-3
Lights 0.3*10-3
Instruments 0.3*10-3
Operational maintenance , hot working during maintenance 5*10-3
Each person during maintenance 2*10-3
Background items, static. 1*10-3
High static operations, drying etc. 1*10-3
Transport, visitors etc.(assuming normal process plant operation, very low 0.3*10-3
transport activity)

Table 1.3 Gas Ignition Probabilities in typical process plant situations (ref. xx,xx)

1.4 Ignition source density models

Spencer and Rew (ref. 1997) developed a model based on a Poisson model for non ignition.
The probability of no ignition at time t for an ignition of a particular type.In their publication
from 2004, Daycock and Rew extended the model giving the probability of non ignition for a
release as:
Ln  = ∑
 ∑ 
  1 −     λ   − 1
Where

is the area of the i’th region of the whole site
 is the density j’th ignition source within the area

is the average period between activations of the source
 is the ignition potential (probability of ignition) for the j’th source in area i
! is the average period of activation of the source
"
is the duration of time that the j’th ignition sourcer has been in contact the
gas cloud
λ is the frequency of activation of the j’th ignition source
a is the probability of the ignition source being initially active
a = ta / (ta + ti)
λ = 1 / (ta + ti)

© J.R.Taylor 2012
2010 3
3
 QRAQ 7 Ignition Probability

a = 1 and λ = ∞ if ti = 0, i.e the ignition source is continuous

This model is a mathematically precise reflection of the probability of ignition, provided that
the underlying assumptions are satisfied. These are in particular that:

• the ignition sources are uniformly distributed within each area

• the probability of presence of an ignition source is independent of the probability of
gas release.

Spencer, Daycock and Rew tabulate values of the parameters for different industries, and
whether the release is day or night. These parameters were derived by a very detailed study
of the presence and distribution of ignition source, with the probability of ignition for each
source being derived from literature, or by judgment, with well documented assumptions.
The entries for process access roads were derived for this project by direct observation in a
refinery and in a gas treatment plant

Land use Ignition sources Typical ignition source parameters

p ta ti a λ µ Loc
-1
min min min ha-1
Car park Rush hour vehicles 0.2 6 474 0.0125 0.0021 160 O
Other vehicles 0.2 6 54 0.1 0.0167 3 O
Smoking 1 10 470 0.021 0.0021 8 O
Road area Rush hour vehicles 0.1 6 474 0.0125 0.0021 160 O
Other vehicles 0.1 6 54 0.1 0.0167 3 O
Delivery vehicles 0.1 6 24 0.2 0.0333 20 O
Traffic control 1 0 15 0 0.0667 20 O
Controlled road Delivery vehicles 0.2 6 24 0.2 0.0333 20 O
Waste ground None 0 - - 0 0 0 O
Boiler house Boiler 1 120 360 0.25 0.0021 200 O
Flames (flares, Continuous (indoors) 1 - 0 1 0 200 I
furnaces, Continuous (outdoors) 1 - 0 1 0 200 O
Infrequent (indoors) 1 60 420 0.125 0.0021 200 I
Infrequent (outdoors) 1 60 420 0.125 0.0021 200 O
Intermittent (indoors) 1 5 55 0.0833 0.0167 200 I
Intermittent (outdoors) 1 5 55 0.0833 0.0167 200 O
Kitchen Smoking 1 5 115 0.042 0.0083 200 I
Cooking equipment 0.25 5 25 0.167 0.0333 100 I
Process areas “Heavy” eqp. levels 0.5 - - 1 0.028 50 I
“Medium” eqp. levels 0.25 - - 1 .035 50 I
“Light” eqp. levels 0.1 - - 1 0.056 50 I
Classified None 0 - - 0 0 0 I
Classified (Ex.) Materials handling 0.05 5 25 0.167 0.0333 10 O
Storage (Ex.) Materials handling 0.1 10 20 0.333 0.0333 10 O
Office “Light” eqp. levels 0.05 - - 1 0.056 20 I
Table 1.4 Ignition source parameters for the base case or typical quality of ignition control

© J.R.Taylor 2012
2010 4
4
 QRAQ 7 Ignition Probability

Table 1.4 is one of three tables prepared by Daycock and Rew, covering typical, poor and
good ignition control.

In use, a grid is developed and the ignition contribution is calculated as a function of time for
each cell on the grid. For a given release, the probability of ignition for each cell crossed by
the release can then be calculated.

The PHAST program, developed by DNV, addresses ignition either by using fixed ignition
probabilities (default values or individually specified for each scenario), or by assessing
ignition probabilities on a grid basis. Each ignition source may be specified as a point source,
or an area ignition density. The ignition probability is calculated (ref. ) as:

#$, &,  = '(,) 1 −  *+,,,-. 

Where

#$, &,  is the probability of ignition at localtion (cell) x,y up to time t

'(,) is the presence factor for the cell (the proportion of time that the source is present
and active
/(,),. is the ignition effectiveness for the cell
t is the time interval in question

Default values can be used as parameters in the model, or the user can specify the values.

As can be seen, the DNV model is similar in concept to the Simmons, Rew and Daycock
model (or vice versa), and has the advantage of being built in to one of the most widely used
ignition control packages.

The Simmons, Daycock and Rew (SDR)models focus on offsite ignitions (despite the title of
ref. xx) because the model development was intended as an improvement of UK HSE’s
RISKAT model for land use planning. Note for example that process areas are referred to as
indoors, which is appropriate for an industrial area, but not for a refinery or petrochemical
plant. The model does not provide data suitable for in plant ignition calculations, lacking
distinctions for example between gas turbine driven pumps and steam turbine driven pumps.

An adaptation of the Simmons, Daycock and Rew model is made here, in order to be able to
calculate ignition probabilities in petroleum and heavy chemicals plant. One of the first
adaptations is that many of the ignition sources, such as pump motors, electrical heaters, and
compressors, can be identified individually. Since one of the purposes of risk analysis may be
to optimize the location of these, it is worthwhile to regard these as point ignition sources.

There are other ignition sources for which detailed identification is not generally worthwhile,
such as operators, maintenance work, hot work and electrical cable ignitions. For these, the
approach of Simmons, Daycock and Rew, of determining an ignition source density, is more
appropriate.

The method of calculating ignition probabilities on a grid cell basis is also inappropriate for
calculating in-plant ignition probabilities. Significant plumes may be less than 1 m. wide at
their origin, and calculating on a 1 m. grid is computationally demanding.

© J.R.Taylor 2012
2010 5
5
 QRAQ 7 Ignition Probability

The approach adopted here is then to track each gas plume through the forest of individual
point ignition sources and area ignition sources, and the calculate ignition probability at each
time step. The equation for this is a simplification of the SDR model:

.

ln  = 1  1 −     λ 2 . − 1



with notation as above. J(t) is the number of ignition sources encompassed by the plume at
time t.

In more detail, the algorithm is to calculate for each release, each release direction and each
time step:

• The point ignition sources encompassed by the plume.

• The areas with potential for ignition crossed by the plume, and the corresponding
plume area overlapping the ignition area.
• Whether the concentration of gas at the point or area ignition source is between the
UEL and LEL
• The probability of ignition at time t, this being determined according to the equation
given above, for all the J(t) ignition sources at locations between the UEL and LEL
(or rather, with some vortices of gas between the UEL and LEL).

There are a number of discretisations which need to be made when applying this model.
Firstly, the number of directions to be taken into account needs to be chosen. For momentum
jet releases which do not develop into wind dispersed plumes, the jet angle is typically about
17˚, giving a number of directions of just over 20. The number is larger if upward directed
jets are taken into account, and lower if impinging jets are taken into account (because
impinging jets are generally wider. For releases such as those for LPG, which can disperse at
a wide angle, fewer directions may be needed, but it is important that the last part of the
plume is properly represented in the calculation, especially if there are high probability
ignition sources such as boilers at this distance.

The second discretization needed is the time step. In actuality, distance steps are needed for
reasonable accuracy, because the velocity of the gas can vary from supersonic close to the
release point, down to less than 1 m/s at the end of the plume if wind speed is low.
Calculation at 1 m. intervals should be more than adequate, given typical spacings of ignition
sources, or each distance intervals can be the ignition source spacing.

In order to derive time steps, the gas velocity along the plume needs to be known. This
involve either calculating the ignition probabilities in parallel with the gas dispersion
calculation, or storing the velocities as a function of time or distance. In the implementation
used in Ch. 4 below, the first of these alternatives is used.

The third discretization which needs to be chosen is that of the representative ignition times
and locations. For a very precise calculation, any of the ignition sources which is within the
plume could be the actual ignition point, and this could occur at any time step. Since the fire
size depends on the time of ignition, each possibility leads to a different consequence. This is

© J.R.Taylor 2012
2010 6
6
 QRAQ 7 Ignition Probability

even more important if there are congested areas along the plume path, since later ignitions
are then more likely to cause vapour cloud explosions.

With a single potential ignition point, the number of possible scenarios is equal to the number
of time steps after reaching the ignition point, with two it is the product of the first and
second number of active time steps. The number of scenarios quickly becomes
computationally excessive. Even if time step differences are ignored, there are J!
possibilities. This compares with a usual number of two possibilities with other methods,
early and late ignition. In the implementation described in Ch 4, the number of scenarios
chosen is one early ignition, in the interval in which the momentum jet dominates, one at the
time when the mid plume distance is reached, one at the time of emergency shut off, and one
at the time when the largest congested area is “filled” to the maximum overlap between
congested area and plume. This is a rule of thumb approximation and is investigated by
sensitivity study in Ch 4.

Typical values for p, a, λ and µ were derived from table 1.3 above and from UK HSE release
data in ref. xx

© J.R.Taylor 2012
2010 7
7
 QRAQ 7 Ignition Probability
Table 1.5 Ignition parameters for outdoor process plant areas with good ignition source control
Land use Ignition sources Typical ignition source parameters
p ta ti a λ µ Person Comment
-1 -2
hr hr yr m present
-4
Process Background 0.1 - - 1 ∞ 3*10 N Depends on typical weather, deserts
area, Ex, (1) have higher probability
-2
well Lighting 0.003 16 8 0.33 ∞ 1*10 N Lights up to 10 m. height considered
managed -2
Cabling, earth 0.001 - - 1 ∞ 1*10 N Cable races at 5 to 8 m. (2)
-4
Operator walk round 0.1 0.5 4 0.125 2190 3*10 Y Twice per shift, ½ hour (3)
-4
Operator sampling 0.1 0.25 8 0.031 1095 3*10 Y Once per shift, one sample point (4)
Pump start 0.001 - - - 52 - Y Pump start switch at typically 5 m.
Pump, motor, switch .001 - - 1 ∞ - N Continuously operating pump
Motorised valve .0001 - - 1 ∞ - N
Electrical heater .0001 - - 1 ∞ - N Heater may also start the release (5)
Lagging .0003 - - 1 ∞ - Y Lagging likely to start the fire (6)
Electrical heat trace .001 - - 1 ∞ - N Tracing may start the fire (7)
Hot piping or flanges 0.3 - - 1 ∞ 0.07 N Only on hot equipment (8)
Steam Turbine .004 - - 1 ∞ - N Based on 7 actual cases
Gas turbine (enclosed) .02 - - 1 ∞ - N Large gas clouds only
Gas turbine release .5 - - 1 ∞ - N Releases from gas turbine itself
Centrifugal compressor .002 - - 1 ∞ - N
Reciprocating compressor .002 - - 1 ∞ - N
Fin fan cooler .05 - - 1 ∞ - N
Instruments, transmitters .00003 - - 1 ∞ - N
Loading rack, per station .001 2 28 .08 365 - Y

! is the period of activation of the source


is the interval between activations of the source

© J.R.Taylor 2012
2010 1
1
 QRAQ 7 Ignition Probability
Table 1.6 Ignition parameters for outdoor process plant areas with poor ignition source control
Land use Ignition sources Typical ignition source parameters
p ta ti a λ µ Person Comment
-1 -2
hr hr yr m present
-4
Process Background 0.1 - - 1 ∞ 3*10 N Depends on typical weather, deserts
area, Ex, (1) have higher probability
-2
well Lighting 0.03 16 8 0.33 ∞ 1*10 N Lights up to 10 m. height considered
managed -2
Cabling, earth 0.01 - - 1 ∞ 1*10 N Cable races at 5 to 8 m. (2)
-4
Operator walk round 0.3 0.5 4 0.125 2190 3*10 Y Twice per shift, ½ hour (3)
-4
Operator sampling 0.3 0.25 8 0.031 1095 3*10 Y Once per shift, one sample point (4)
Pump start 0.003 - - - 52 - Y Pump start switch at typically 5 m.
Pump, motor, switch .003 - - 1 ∞ - N Continuously operating pump
Motorised valve .0001 - - 1 ∞ - N
Electrical heater .001 - - 1 ∞ - N Heater may also start the release (5)
Lagging .003 - - 1 ∞ - Y Lagging likely to start the fire (6)
Electrical heat trace .01 - - 1 ∞ - N Tracing may start the fire (7)
Hot piping or flanges 0.3 - - 1 ∞ 0.07 N Only on hot equipment (8)
Steam Turbine .004 - - 1 ∞ - N Based on 7 actual cases
Gas turbine (enclosed) .02 - - 1 ∞ - N Large gas clouds only
Gas turbine release .5 - - 1 ∞ - N Releases from gas turbine itself
Centrifugal compressor .002 - - 1 ∞ - N
Reciprocating compressor .002 - - 1 ∞ - N
Fin fan cooler .1 - - 1 ∞ - N
Instruments, transmitters .00003 N

Poor ignition source control is defined by examples in the photographs at the end of this chapter

© J.R.Taylor 2012
2010 2
2
 QRAQ 7 Ignition Probability
Table 1.7 Ignition parameters for outdoor utility area
Land use Ignition sources Typical ignition source parameters
p ta ti a λ µ Person Comment
-1 -2
hr hr yr m present
-4
Process Background 0.1 - - 1 ∞ 3*10 N Depends on typical weather, deserts
area, Ex, (1) have higher probability
-2
well Lighting 0.03 16 8 0.33 ∞ 1*10 N Lights up to 10 m. height considered
managed -2
Cabling, earth 0.01 - - 1 ∞ 1*10 N Cable races at 5 to 8 m. (2)
-4
Operator walk round 0.3 0.5 4 0.125 2190 3*10 Y Twice per shift, ½ hour (3)
-4
Operator sampling 0.3 0.25 8 0.031 1095 3*10 Y Once per shift, one sample point (4)
Pump start 0.1 - - - 52 - Y Pump start switch at typically 5 m.
Pump, motor, switch .01 - - 1 ∞ - N Continuously operating pump
Motorised valve .0001 - - 1 ∞ - N
Electrical heater .001 - - 1 ∞ - N Heater may also start the release (5)
Lagging .003 - - 1 ∞ - Y Lagging likely to start the fire (6)
Electrical heat trace .01 - - 1 ∞ - N Tracing may start the fire (7)
Hot piping or flanges 0.3 - - 1 ∞ 0.07 N Only on hot equipment (8)
Steam Turbine .04 - - 1 ∞ - N Based on 7 actual cases
Gas turbine (enclosed) .02 - - 1 ∞ - N Large gas clouds only
Gas turbine release .5 - - 1 ∞ - N Releases from gas turbine itself
Centrifugal compressor .02 - - 1 ∞ - N
Reciprocating compressor .02 - - 1 ∞ - N
Fin fan cooler .1 - - 1 ∞ - N
Instruments, transmitters .00003 N
Boiler 1 - - 1 ∞ - N

© J.R.Taylor 2012
2010 3
3
 QRAQ 7 Ignition Probability
Table 1.8 Ignition parameters for specific high probability and mechanical ignition sources

Land use Ignition sources Typical ignition source parameters

p ta ti a λ µ Person Comment
-1 -2
hr hr yr m present
High Fired heater 1 - - 1 ∞ - N The heater can be the source of fire
probab- Boiler 1 - - 1 ∞ - N
ility Melter 1 - - 1 ∞ - N
ignition Drier 0.1 - - 1 ∞ - N Non fired drier
sources Fired glycol regenerator 1 - - 1 - N The regenerator can be the source of
∞
fire
Ground flare 1 - - 1 ∞ - N
Flare stack 1 - - 1 ∞ - N Only applicable for elevated plumes
Mech- Transport belt 0.03 - - 1 ∞ - N
anical Screw conveyor 0.1 - - 1 ∞ - N
handling Crusher 0.1 - - 1 ∞ - N
Front loader .1 8 16 .33 ∞ - Y
Reclaimer .1 8 16 .33 ∞ - N
Fork lift truck .01 12 12 .5 ∞ - Y Fork lift approved for the area

Table 1.9 Ignition parameters for in plant transportation

Land use Ignition sources Typical ignition source parameters
p ta ti a λ µ Person Comment
-1 -2
hr hr yr m present
In plant Unit access road 1 - - 1 ∞ - N The heater can be the source of fire
transport Main plant access road 1 - - 1 ∞ - N
Perimeter road 1 - - 1 ∞ - N
Loading station access 0.1 - - 1 ∞ - N Non fired drier

© J.R.Taylor 2012
2010 4
4
 QRAQ 7 Ignition Probability
Table 1.10 Ignition parameters for maintenance
Land use Ignition sources Typical ignition source parameters
p ta ti a duration λ µ Person Comment
-1 -2
hr hr days yr m present
Mainten- Individual maintenance 0.002 6 18 .25 2 4 .01 Y Maintenance frequency and
ance worker (during plant duration to be adjusted
operation) according to plant
Hot work (during plant 0.1 .25 .25 .5 2 4 .01 Y Use of welding tent and fire
operation) watch assumed
Hot work (during plant 1 .25 .25 .5 2 4 .01 Y Use of fire watch assumed,
operation) no welding tents
Electrical hot work (during 0.1 1 7 .125 2 4 .01 Y Use of fire watch assumed.
plant operation) Only short periods with live
circuits
Filter change (per filter) .001 - - - .04 4 - Y
Individual maintenance 0.1 .25 .25 .5 2 0.33 .01 Y Plant unit under maintenance
worker (during major turn is assumed to be shut down.
round) No night shift assumed
Hot work (during major 1 12 12 .5 10 0.33 .01 Y Plant unit under maintenance
turn round) is assumed to be shut down.
No night shift assumed
Electrical hot work (during 0.1 1 7 .125 2 0.33 .01 Y Plant unit under maintenance
major turn round) is assumed to be shut down.
No night shift assumed
Only short periods with live
circuits

© J.R.Taylor 2012
2010 5
5
 QRAQ 7 Ignition Probability

1.5 Examples of poor ignition control

Figure 1,1 Poor cable support

Figure 1.2 Non Ex approved power cable used in classified area

© J.R.Taylor 2012
2010 1
1
 QRAQ 7 Ignition Probability

Figure 1.3 Welding cables trailed through plant

Figure 1.4 Hot oil dripping on lagging

© J.R.Taylor 2012
2010 2
2
 QRAQ 7 Ignition Probability

Figure 1.5 Non Ex fan used for ventilation of a kerosene tank

Figure 1.6 Broken junction box

© J.R.Taylor 2012
2010 3
3
 QRAQ 7 Ignition Probability

Figure 1.7 Ex junction box with no closure bolts. Closed with a twist of wire.

Figure 1.8 Electrical cables after washing down with a hose

© J.R.Taylor 2012
2010 4
4
 QRAQ 7 Ignition Probability

Figure 1.9 Broken junction box

Figure 1.10 Broken light

© J.R.Taylor 2012
2010 5
5
 QRAQ 7 Ignition Probability

Figure 1.11 Light partially filled with solvent

Figure 1.12 Transmitter cap left open in error on powder hopper

© J.R.Taylor 2012
2010 6
6
 QRAQ 7 Ignition Probability

© J.R.Taylor 2012
2010 7
7
 QRAQ 7 Ignition Probability

2. Data from industrial release ignitions

2.1 Offshore releases
UK HSE has published release data for off shore installations. The data also give ignition
incidence and ignition causes. Table 2.1 gives ignition percentages based on the total release
data, broken down according to equipment. Table 2.2 gives the data from HSE’s own
summary assessment.

Equipment Flammable Releases Ignitions %

substance
Compressor 2717 16 0.6
Lubrication oil 12
Gas 2
Dehydration unit 768 8 1
Glycol 768 6
Condensate 2
Diesel generator 3063 22 0.7
Diesel 20
Lube oil 2
Utilities Gas 1023 7 0.7
Process Gas 724 39 5.4
Separation 1686 3 0.18
Condensate 1
Gas 2
Vent Gas 3345 9 .27
Gas turbine 4172 22 0.5
Lube oil 6
Gas 9
Diesel 7
Table 2.1 Ignition probabilities for offshore installations from ref. 5

LIQUIDS GAS 2-PHASE

Major Sig Minor Major Sig Minor Major Sig Minor Total
Ignitions 0 24 67 0 16 24 0 0 0 131
Releases 14 301 422 99 739 312 30 133 21 2071
% 0 8 15.9 0 2.2 7.7 0 0 0 6.3

Table 2.2 Summary data for ignitions in offshore installations

Both of these tables raise significant problems, which is a pity because the data provided are
virtually the only high quality data available for process plant ignition probabilities. The main
problem which arises is that ignition probability actually falls with increasing release size.
Additionally, two phase flammable releases, generally regarded as the most dangerous,

© J.R.Taylor 2012
2010 8
8
 QRAQ 7 Ignition Probability

actually have the recorded ignition percentage of zero. This cannot be explained by
inadequacies in the data collection, since the amount of data gives a high statistical
significance in all categories.

The only way in which we have been able to explain the results is the observation that the
locations where there are the largest pipe sizes, the highest pressures, and the largest
inventories, are also the places where the greatest care is taken to eliminate ignition sources.

One additional observation, which is similar to the general one, is that a large fraction of the
ignitions are from utilities and rotating equipment, compressors and generators, which are
generally located in such a way that ignition sources cannot be avoided.

The summary of types of ignition is shown in table 2.3. A large fraction of the ignitions are
ones in which the release and ignition are closely related, such as internal leaks in a burner.
Those ignitions which are truly independent of the release mechanism, such as ignition of a
passing gas cloud, are marked “Yes”.

Table 2.3 Ignition sources for offshore data

Independent
Ignition cause %
Ignition source
Gas turbine hot exhaust lube oil 22.1 Yes
Gas turbine hot exhaust gas 1.5 Yes
Gas turbine exhaust internal 7.6
Gas turbine start 0.8
Gas turbine other 3.8 Yes
Hot work 20.6 Yes in most cases
Hot pipe, flange 1.5 Yes
Flare carry over 3.8
Flare external 1.5
Flare internal 3.8
Static electricity 1.5 Yes
Electrical earth failure 1.5
Lightning 4.6
Pump seal 1.5
Pump hot surface 1.5 Yes
Diesel generator hot surface 7.6 Yes
Diesel generator exhaust
(internal) 2.3
Lagging fire 1.5
Fired glycol regenerator 3.8
Burner hot surface 0.8
Burner start 0.8
Filter change 0.8
Condensate in fuel gas to burner 1.5
Heat tracing damage 0.8 Yes
Unknown 2.3

© J.R.Taylor 2012
2010 9
9
 QRAQ 7 Ignition Probability

2.2 Onshore release

The offshore data in section 2.1 is a well controlled data set. The relative frequencies for
ignition sources can be relied on to a large degree. The plants which are the basis for the data
are also a relatively homogeneous set. However the data applies to one particular type of
plant, and several ignition types are missing from the data set because they are not applicable
to offshore installations.

Table 2.4 shows ignition causes for 325 records extracted from the MHIDAS database using
the key word IGNITED. MHIDAS is a case history database, not a statistical database. The
records are taken from company, journal and press reports, so the completeness of the data
set and the population of plants from which the data set is drawn is unknown. Nevertheless,
the data set is one of the largest, and for relative judgments can probably be relied on. I gives
at least the important ignition sources which must be taken into account in any ignition
probability model.

Table 2.4 Ignition sources for events in the MHIDAS data base
Cause % %
Uncontrolled chemical reaction 1.226994
External fire at tank 0.613497
Chemical reaction in spill 0.613497
Flare discharge 1.533742
Lightning 3.067485
Hot steel 0.306748
Cooling tower electrostatic discharge 1.533742
Stove, paraffin heater 0.613497
Autoignition, chemical 6.441718
Auroignition, temperature 0.306748
Polymerisation in piping (C2H4, HCN) 0.920245
Catalyst auto igniion 0.306748
Total auto ignition 0 7.97546
Reaction with moisture 0.613497
Seal friction 0.613497
Housing 0.306748
Sand friction, blown sand 3.067485
Fired heater, furnace, boiler 11.04294
Control room 0.613497
Sabotage, terrorism 0.920245
Vandalism, arson 1.533742
Naked flame in plant 0.613497
Naked flame in neighbourhood 1.226994
Naked flame in laboratory 0.306748
Total naked flame 0 2.147239
Ship 1.533742

© J.R.Taylor 2012
2010 10
10
 QRAQ 7 Ignition Probability

Cause % %
Fork lift truck 1.840491
Waste dump, trash 0.613497
Elecrostatic in liquid transfer 0.920245
Electrostatic C2S 0.613497
Electrostatic, dust 3.374233
Electrostatic, particles 0.306748
Electrostatic, hydrogen jet 0.613497
Electrostatic, venting 0.613497
Elecrostatic, gas jet 0.920245
Cooling tower electrostatic discharge 1.533742
Total electrostatic 0 8.895706
Friction, sensitive material 0.920245
Electrical equipment 0.306748
Electrical process heater 0.306748
Electrical fire 3.067485
Severed cable 0.613497
Spark in electrical system 0.613497
Electrical transmission lines 0.306748
Transformer 0.613497
Total, electrical 0 5.828221
Cutting torch 1.226994
Workshop spark 0.920245
Welding 6.441718
Blow torch 0.306748
Friction, maintenance 0.920245
Sanding, grinding, shot blasting 0.920245
Demolition 0.920245
Total, hot work 0 11.65644
Molten metal 0.306748
Lagging fire 1.840491
Hot surface 0.306748
Car engine 0.306748
Pump 1.533742
Motor, Non Ex 0.920245
Equipment, blower 0.613497
Equipment, air compressor 0.306748
Conveyor 0.306748
Machinery 0.306748
Centrifuge friction 0.306748
Gas engine 0.306748
Diesel engine exhaust 0.306748
Spark from machine 0.920245

© J.R.Taylor 2012
2010 11
11
 QRAQ 7 Ignition Probability

Cause % %
Hot spots in equipment 0.306748
Total, equipment 0 6.441718
Hot brakes 0.613497
Passing train 1.226994
Truck, tractor etc. 3.98773
Truck fire 0.306748
Total, transport ignitions 6.134969
Rail tank waggon crash 7.668712
Tank truck crash 6.748466
Crash into pipes 0.306748
Crash into vessel 0.613497
Total, transport crash and ignition 15.33742

© J.R.Taylor 2012
2010 12
12
 QRAQ 7 Ignition Probability

3. Ignition probabilities for specific

ignition sources
Before discussing specific ignition sources, the actual form and dependency of the
probabilities must be discussed. Most of the cases of concern involve a scenario in which gas
or liquid is released, and gas or vapour disperses as a cloud or plume. If the cloud or plume
encounters a potential ignition source which is active at the time, ignition will occur. Many of
the ignition sources covered by this model are static i.e. remain in the same place. Others
though are mobile, such as traffic (on roads, or truck, fork lift, crane traffic etc. inside a
plant). Persons such as operators and maintenance tradesmen also constitute sources of
ignition, for example from electro static discharges from clothing. For these cases, a factor is
needed not only for the probability of ignition but also for the presence of the person.

In other cases, the flammable substance is permanently present, such as in storage tanks
which are not blanketed, at floating roof tank rim seals, API separators, and at drains. In these
cases, a probability of ignition is inappropriate, what is needed is a frequency of ignition. For
some plants, especially refineries, oil terminals, and petrochemical plants, it is this last group
which dominates. The actual time of ignition is in many cases critical. If a gas or vapour
flame is ignited immediately the consequence will be a jet fire, followed possibly by a pool
fire. The consequences will be local unless domino ensure. If the vapour is ignited late, an
explosion may occur in the gas/vapour plume or cloud.

These data are not complete enough to be able to form the basis for a full model for ignition
probability. Among the difficulties are the limited range of ignition sources frequency of hot
work. Nevertheless the data are some of the most complete available from published sources.
They provide an important test set for other models, which should at least be able to predict
the frequency of releases, ignition, and the ignition source profile for a typical platform.

3.1.1 Self ignition

Some releases will ignite immediately on release. Examples in petroleum and petrochemical
plants are:

- liquids above their auto ignition point

- lithium aluminium hydride and aluminium alky used as catalysts
- liquids containing catalysts such as Kaney nichel which are pyrophoric.

In fire chemicals plants, in addition there are reagents which are pyropheric and some which
react violently with water, causing ignition for example:

- lithium aluminium hydride

- grignard reagents
- sodium borohydride
- sodium mehylate

© J.R.Taylor 2012
2010 13
13
 QRAQ 7 Ignition Probability

- sodium azide

For all these cases, the probability of ignition on release may be regarded as 1.0. generally,
the ignition will be immediate.

In some cases, air entering equipment causes fires. Examples are vacuum distillation columns
for crude oil, asphalt tanks and tanks containing sour oil or sour water. Asphalt ignites
because it is slowly pyrophoric. Sour oil may ignite from Ferrous sulphide which forms when
hydrogen sulphide reacts with steel. Ferrous sulphide is pyrophoric.

For asphalt tanks, the observed frequency of ignition is estimated to be 8*10-3 per year, based
on 3 incidents. For sour water and sour oil tanks, the frequency will depend on whether the
tanks are blanketed, (usually with fuel gas). If the tanks are blanketed the observed frequency
is very low (no known cases out of 1985 tank years for operating, blanketed tanks). There
will be possibilities for ignition when tanks are emptied and aerated for maintenance, but
these ignitions should not cause major accidents, since the tanks should be freed of oil and
vapour before ignition can occur. Four cases are known of ignitions in scale from the tanks
listed above, and there probably occurred many more, but in only one of the cases were the
consequences significant, with consequences limited to tank wall damage. Lack of care in
aerating a tank could lead to a tank explosion.

3.1.2 Jet ignition

Gas jets can ignite simply by generating static electricity. Hydrogen at high pressure has
ignited apparently spontaneously in several reported accidents. Several possible mechanisms
have been investigated. Possible causes are shock boundaries and electrostatic charging of
dust particles.

Sand particles blown by natural gas jet releases can definitely cause ignition, as has been
demonstrated by the author. In experiments using calcareous sand under desert conditions,
blown sand particles could reproducibly cause ignition.

3.1.3 Furnaces, fired heaters, and boilers

A furnace, fired heater, or boiler is an obvious source of ignition. If flammable gas or vapour
passes into the air intake to such a furnace. Cases are known in which gas has passed through
a boiler, with a release of block smoke from the chimney as the only consequence. These are
regarded as extremely lucky cases however.

In some cases the actual location of the air intake will be important, especially if the fire box
is elevated at level 1 (typically 4 to 5 m) or level 2. Cases are known in which heavy gas
clouds have passed beneath the fire box, without ignition.

Locations at which air is taken into the fire box are forced draft fans. Seals around separate
sections of furnace, burner front shield, and pilot burner air intakes. In some cases, viewing
point covers also can leak, or be left open. Note that gas will always be drawn into any
opening in a fire box, either due to the vacuum from induced draft fans or from the vacuum
induced by the chimney.

© J.R.Taylor 2012
2010 14
14
 QRAQ 7 Ignition Probability

3.1.4 Flares

Flares will obviously ignite flammable gas or vapour, if the vapour reaches the flare location.
This is extremely unlikely for most releases, because flares are usually positioned remotely
from the petrochemical or petroleum plant (at least 50 m safety zone) and partly because they
are elevated far above the height of a typical vapour plume or cloud. Ground flares are much
more likely to ignite a vapour plume, and many cases of such ignitions are known. In general
the ignition probability should be regarded as 1.0 if the plume or cloud reaches the flare.

Elevated flares present an additional ignition hazard by ejecting burning liquid or soot
particles. (The Milford Haven tank fire and boilover was considered to have been ignited by
falling glowing soot from a flare). The cause is either too high a level in the flare knock out
drum, or release of liquid into the flare due to a high level in the process level. (A few cases
are also known of liquid gathering at low points in flare lines, and then being ejected when
gas flow reaches a high level. Flare lines should not have low points where liquid can collect,
however).

The frequency of such burning liquid releases was observed to be 4 cases out of 28 flares
with known histories for a total of 455 flare years. This gives a frequency of 0.009 per year.

Such releases from flares have been known to cause severe injury, and at least one fatality is
known. More usually, the consequence is a limited fire and limited equipment damage.
Consequences become large if the burning liquid reaches floating roof tanks, API separators,
drains, or process equipment which is dirty with oil.

3.1.5 Hot work

Within process plant, hot work is one of the more frequent causes of ignition. Hot work
includes, welding, flam cutting, abrasive wheel cutting and grinding.

Hot work permit systems are intended to provide a formal check that an area is free of
flammables, before the work is started. Such systems do not always function, as for example
the cases of the large terminal fire at Thessaloniki and the fire at the Port Eduard Heriot
terminal. Reasons for failure of permit to work systems are:

- The safety inspector sometimes (illegally) signs the permit to work without
actually making an inspection. This happens especially if the permitting work
load is higher than be carried out by the available staff.
- The safety inspector overlooks a location with flammables (such as a sewer or
drain trench).
- The safety inspector makes an over optimistic evaluation of the hazard zone.
Sparks from abrasive cutting can easily be thrown 20 m. hot slag from welding
and flame cutting is usually thrown only a few meters horizontally but can fall
tens of meters to the ground, and still be hot enough to cause ignition. Safety
inspectors should ensure that the area is free of flammables before checked is
sometimes as little as 5 m. Also inspectors sometimes forget the problem of
falling sparks.

© J.R.Taylor 2012
2010 15
15
 QRAQ 7 Ignition Probability

- Workmen and foremen sometimes proceed with work without a hot work
permit. This applies especially for contractors.

The frequency for ignition of existing flammables from hot work is virtually impossible to
determine. Many such fires occur, and are immediately extinguished with hand extinguishers.
Which of these cases are significant is difficult to define, point and insulation fires may be
impossible to avoid, and instead are controlled.

From inspection of near miss records, one case per year of ignition in a refinery can be
typical. Such fires are usually controlled immediately.

If a large release occurs while hot work is taking place, the safety zone established for the
work will generally be completely inadequate. If the hot work activity is actually in progress
when the flammable vapour reaches the area, ignition will be virtually certain. The
probability of ignition depends on whether:

a) There is any maintenance or construction activity in progress requiring hot

work.
b) The location of the activity. In risk assessment this translates to the density
of the activity per unit area.
c) The fraction of time during which welding and cutting actually proceeds.

The frequency of major maintenance and construction work requiring hot work varies very
much from plant to plant. From experience, in a refinery or petrochemical plant there is
usually at least one maintenance or construction at work. Also, there will be a complete unit
turn round every year to every four years, which requires hot work nearly all the time.
Sometimes others continue operating. More often, a complete plant, or part of a plant is shut
down for a few weeks.

For fine chemicals plant, the frequency of hot work is more closely related to the frequency
of plant modifications.

A simple approach to determining probability of hot work ignition is to estimate first the
fraction of time with such activity. For refineries this is about 1.0 at any one time. For gas
plants it is about 0.1, and can by careful design be reduced to as little as 0.01 For fine
chemicals, it is the frequency of plant changes per year times the period of work, (about 2
months for the hot work part of new construction).

More difficult is to determine the probability of hot work taking place within range of a
release i.e. in the range where the gas plume is still above the lower explosion limit, or, more
precisely, where vortices within the plume contain some gas which is above the lower
explosion limit (LEL). Figure 3.1 shows a gas release calculated using the large eddy
simulation (LES) technique. The length of the plume where there is some gas above the LEL
is about twice the length of the plume where most of the gas is above the LEL.

The next stage involves determining the process area, and dividing the area of hot work (5m
diameter circle in 90% of cases, 20m. diameter circle in 10% of cases, based on photographic
survey).

© J.R.Taylor 2012
2010 16
16
 QRAQ 7 Ignition Probability

Finally, even when work activities are taking place which require hot work, the ignition
source will not be present all the time. Setting up, fitting, and erection requires a large
fraction of the time. In order to determine this fraction, direct observations were made for a
number of maintenance and new construction activities. These observations are given in table
3.1.

For risk analysis purposes it is important to take into account that:

• When ignition occurs due to hot work, the probability of personnel exposure will
be 1.0
• Hot work ignition is likely to occur when maintenance or plant upgrades are in
progress, and the frequency of releases at this time will be increased.

Activity Percentage of time with ignition source

active
Piping replacement 4%
Tank construction welded plates 86%
Structural steel construction 12%
Concrete construction, rebar cutting 6%
Installation of rotating equipment 0.4%
Tank repair 2%
Piping workshop 8%

Table 3.2. Fraction of time during which hot work ignition sources were observed to be
present. From 34 observations. Percentages apply during working hours, usually 8 hours per
day. Sometimes activities continue 24 hours per day, esp. during turn round maintenance, but
hot work is rarely carried out at night.

Workshops are usually located outside process areas, but will nevertheless constitute an
ignition source for the largest releases.

Machinery in a workshop will also be able to ignite gases and vapour, but the contribution of
these sources will usually be insignificant compared with welding, cutting etc.

3.1.6 Lagging fires

When oil or some kinds of solvent seek into insulation/lagging, the surface area of the liquid
is increased enormously. Natural tendency to oxidation is increased by the more intimate
contact between liquid and air. The heat generated by the oxidation is retained by the
insulation, and oxidation rate from the piping or vessel under the insulation speeds the
process.

Liquids which are subject to this kind of ignition are:

- crude oil
- residual oil
- resins, paints
- peroxides (such as benzoyl peroxide used as catalyst in polymerization)

© J.R.Taylor 2012
2010 17
17
 QRAQ 7 Ignition Probability

- many solutions of fine chemicals

- food oils such as soya oil

Fires typically break out when insulation is removed, allowing access of air to the hot liquid.
Flames can arise in the insulation itself, and can cause significant consequences if, for
example, flange leaks are ignited. (the initial flow from a seeping flange will not in itself be
significant, but the flange will probably open in response to even a small fire. The size of the
resulting fire will depend on whether bolts lengthen or break, whether the flange gasket
material is damaged by the heat (some gaskets are asbestos loaded, some are made of
graphite, and some are steel), and on the pressure in the pipe.

A search of equipment integrity audit records was made for 6 refinery, petrochemical, and
paint resins plants, with an estimated 8.5 km of susceptible piping. A total of 34 cases were
found which were considered significant potential locations for lagging fire because of leaks
with oil or solvent soaking lagging. This gives a frequency of between 1.3 and 3.9 potential
incidences per km year of insulated piping. The uncertainty depends on how long the cases
are regarded as having existed. Some will have been present for many years, others were
observed to be relatively new. As a basis for calculation a value of 2 releases per lagged pipe
km year is assumed with oil or solvent seeped into lagging.

For these plants, a total of 5 lagging fires has occurred over a period varying from 4 to 20
years, with a total of 65 plant years. This leads to an estimate of frequency of lagging fires in
susceptible piping of 0.009 per pipe km year. In only one case did the ignition lead to a large
fire, giving a frequency for large fire of 0.0018 per insulated pipe km. year.

3.1.7 Hot surfaces

As can be seen from the UK HSE offshore statistics, hot surfaces can cause ignition,
especially if there are droplets of flammable liquid which fall on the surface. Especially
heavier oils from highly flammable smoke and vapours as they crack on the hot surfaces.

3.1.8 Pump and fan bearings and seals

Many pumps have continuous releases from seals – A small leakage is necessary to ensure
that the seal surfaces are lubricated. The liquid may be piped away, but often there is a funnel
or similar opening to catch the liquid. Sometimes the liquid just drains to a sewer.
Overheating of bearings can ignite flammable liquid released in this way. Overheating due to
damage at the seal itself can both cause ignition and cause increased release. In the limit,
destruction of the pump may cause a fire.

Case story:

A large (2.5 meter rotor) blower lost a fan blade. The fan imbalance caused vibration which
ripped causing bolts (1.5” diameter) out of the foundation and stripped 24 bearing lousing
bolts (1.2”) out of the threads. This released lubrication oil. The still rotating fan and shaft
caused friction and beating which ignited the oil. The resulting pool fire was about 10 m in
diameter, and the flames 14m high. The frequency of fires initiated by pump bearing, seal,
and rotor failures. The frequency of such ignitions is drawn from experience from plant

© J.R.Taylor 2012
2010 18
18
 QRAQ 7 Ignition Probability

audits and incident records for refinery and petrochemical plant. The frequency is 1.8*10-4
per equipment year, based on 11 pump fires and on one large air blower fire..

3.1.9 Motors and machinery as ignition sources

Pump motors, conveyor motors and machinery can constitute ignition sources. In plants
which process or make use of flammable liquids or dusts, motors are explosion safe. It is
almost unheard of that such motors should cause ignitions. In plants which have only limited
use of flammable liquids, motors may be selected to be “non sparking”. These will only
rarely be able to function as ignition sources.

Other motors, for example in parts of plants processing non flammable materials and in
workshops, may be a permanent ignition source, due to brush arcing, or, more frequently,
during motor starting and stopping.

Case story:

Hexane had been released from an oil seed processing plant for about an hour. Vapour had
spread through the factory area and into the canteen kitchen. It was decided to shut off the
electrical power to the plant. Vapour ignited in the kitchen and this caused an unconfined
vapour cloud explosion. It was concluded that the most likely source of ignition was the
motor on one of the refrigerator compressors.

Spencer, Daycock and Rew, (ref. 6) estimated ignition probabilities in industrial areas
depending on the “level” of equipment (heavy, medium and light machinery) and the density
of machinery. Ignition probabilities vary from 0.15 to 0.5 for active machinery, and the rate
of activation of the source varying from 0.056 to 0.027.

The background for these values is given in terms of “covered” and “uncovered” spark
sources, with p = 0.6 for an uncovered large motor and p = 0.4 for an uncovered small motor.

Ex approved motors very rarely give spark ignitions. Damage junction boxes are nearly the
only cause. Ex approved motors can cause ignitions or electrical fires by short circuit due to
overheating, or due to bearing overheating.

3.1.10 Switch rooms and control rooms

Switching equipment is used to control motors and other electrical equipment. It consists of
circuit breakers and relays, with associated drive circuitry, arranged in racks. Generally, the
equipment is kept in an air conditioned room, often close to the control room.

Switching equipment is not usually explosion safe and never in the case if equipment in a
switching room. Switching equipment is operated frequently, so that the ignition probability,
if gas should enter, is close to 1.0. usually though, switch room ventilation is protected, so
that ventilation fans are stopped and ventilation ducts closed when gas detectors activate.

The probability of ignition is therefore determined largely by the probability that gas
detectors are in a failed state, or (from experience) erroneously installed.

© J.R.Taylor 2012
2010 19
19
 QRAQ 7 Ignition Probability

Similar consideration apply to control rooms, which generally contain computers coffee
machines, printers, and may have smoking rooms.

Operator rest rooms and shelters within process units in modern plants are generally
constructed in the same way, with automatic closure of ventilation in the case of gas or fire
alarms.

The best control rooms have airlocks with two sets of doors, and with alarms to indicate
when both sets of doors are open.

Analyzer houses have similar ignition properties to control rooms, but in addition may
accumulate gases from sampling if not properly ventilated.

3.1.11 Cabling
Cables running through plant are a potential source of ignition. The probability of such
ignition has been studied in depth for nuclear power plant.

The probability of ignition arises considerably if there are splices (joins) in cable runs,
because any failures in the splices, or corrosion, causes resistance and heat increases. For this
reason it is usual to insist in plant installation that cable runs are free from splices. Cables are
joined in junction boxes. These should be sealed because inside the junction box there are
live contacts. In actual practice, sealing is often deficient. This is especially a problem if there
is heavy rain, or if there is a leak of water in the plant. Cabling will cause an ignition if
damaged, for example by falling objects, crane or truck crashes. Cables in conduits may be
damaged when conduit corrodes, and begins to hang from the cable itself.

One of the highest hazards from cable arises if they are cut during the accident. Pipe ruptures
in particular can cause pipe ends to whip and cut cables.

In all though, cabling is one of the least probable causes of ignition within a plant.

3.1.12 Traffic

Traffic is one of the more probable cause of ignition of the larger releases of gas or vapour.
One precondition for ignition though is that the release is large enough to reach roadway.

There are reports of vehicles being able to drive through gas clouds. There are far more,
however, of vehicles igniting clouds and plumes. These include cases of vehicles stalling in
the cloud due to lack of air, and then igniting the cloud as the driver attempts to restart the
engine. Spencer, Daycock, and Rew assumed an ignition probability of 0.1 for all vehicles,
based on observations by Simmons and by Jeffreys. In view of the data from MHIDAS, with
3% of ignitions being by trucks (table 3.4) this value seems somewhat optimistic. A value of
0.25 seems more suitable.

The actual probability of ignition then depends on the density of traffic, i.e. the probability
that a vehicle will enter the cloud. This is given approximately by multiplying the time
duration of the cloud by the frequency of traffic along the route.

© J.R.Taylor 2012
2010 20
20
 QRAQ 7 Ignition Probability

For in plant vehicles, the actual ignition frequency will depend on the type of vehicle. Fork
lift trucks, especially, are often provided which are explosion safe. They at least will have
flame arrestors in the exhaust, if they are to operate in plants where highly flammable.

3.1.13 Lights

Lights in plants are a potential source of ignition. In plants processing flammables, lights will
usually be selected to be explosion safe. This will always be the cases in classified areas.
However, the seals on lights may break, and mountings may be misadjusted. More serious
problems can arise with broken lights.

The most likely cause of ignition from lamps arises however, if the lamp is broken by the
release itself. This has been observed to happen for very high pressure releases of liquids.

3.1.14 High voltage equipment

High tension cables, high voltage circuit breakers, and transformers are a likely cause
ignition if gas or vapour clouds pass over them. High tension equipment gives discharges
especially in damp, foggy weather which give an ignition probability close to 1.0. The
probability will be less in dry weather. The probability will be very close to 1.0 if liquid is
sprayed onto the high voltage equipment.

© J.R.Taylor 2012
2010 21
21
 QRAQ 7 Ignition Probability

4. Model Evaluation
Reviewing the methods, four groups can be found:

• those which give a simple ignition probability, dependent at most on installation type
• those which allow the probability to depend on release size
• those which allow the probability to depend on release size and on estimate of the
density of ignition sources
• those which allow the probability to depend on release size and on the actual location
of ignition sources.

Following the general intention of this study, one aim of any method should be to give credit
for risk reduction measures as far as possible. To motivate further risk reduction, the two last
categories of method are preferred. The methods require computer program implementation
to be tractable however.

Models which rely on fixed probabilities are hardly likely to be particularly accurate,
considering the wide range of probabilities arising in practice. The best that can be said for
such models is that they are convenient.

Models which take account of release rate only at least are responsive to the size of the area
covered by the gas plume. They do not take into account the layout of plant, and, for
example, layouts which segregate boilers and fired heaters from process plants. This can in
extreme cases cause errors of about an order of magnitude.

Both the methods which take into account density of ignition sources according to area type,
and those which record explicit single point ignition locations have the advantage that they
are responsive to good layout practices such as:

• Locating boilers and fired heaters remotely.

• Locating power distribution substations at a distance.
• Locating boilers, heaters and substations upwind to the prevailing wind direction,
when the prevailing wind probability is significantly higher than the average wind
direction probability.
• Changing heating technology (steam from a remote boiler for heating rather than local
fired heaters.)
• Using water curtains to isolate releases from items of equipment such as cracking
furnaces and FCC regenerator boilers.
• Providing rapid shutoff, such as inventory isolation ESD valves, at locations where
releases would give a high probability of ignition.

One of the problems with this line of thought is that such calculations are only really relevant
at the conceptual stage of design, yet they require quite detailed data. In order to overcome
this difficulty, it is useful to be able to transfer appropriate data from earlier analyses, so that
reasonably realistic assumptions can be made.

© J.R.Taylor 2012
2010 22
22
 QRAQ 7 Ignition Probability

In order to test the difference which choice of ignition probability model makes, fire and
explosion risk calculations were made for a large refinery crude distillation unit, including
vacuum distillation and a light ends (propane, butane) unit. The same data for release
frequencies (unit process approach, ref. 8) was used for all calculations.

Evaluation criteria are ideally:

• Accuracy on average when compared with known ignition frequency situations.

• Sensitivity to special cases
• Ability to support layout planning and risk reduction.
• Ease of use

For determining accuracy, the only data set which could be found which gives both release
and ignition statistics was the UK HSE HRDC database. To be able to reasonably check the
ignition probabilities, a risk analysis for an offshore platform is required. In order to get a
reasonable comparison for ignition of gases, only the ignition sources independent of the
release were taken into account. For the tests described below, the UK HSE HRDC release
frequency database was used.

Concerning ease of use, the methods based on release rate of flammables are the easiest to
use provided appropriate software is available. Current commercial software does not support
these methods however, so that some effort is required, first to calculate the release rates,
then to calculate the ignition probabilities, and finally to input the ignition frequencies to the
risk calculation program. Also, fixed probability methods are relatively easy to use, although
some effort is required because most methods of this kind give different probabilities for
different flammable materials.

Methods which use discrete ignition source mapping or ignition source density mapping
require the largest effort. With suitable software support though the effort is not large.

All calculations were carried out using the QRA Open software, which was extended to
implement all the methods described here.

4.1 Risk analysis for the refinery crude unit

4.2 Risk analysis for an oil production platform

A typical oil shallow water oil production platform was taken as the object for this
assessment.

© J.R.Taylor 2012
2010 23
23
 QRAQ 7 Ignition Probability

5. References
1. Dutch Purple Book
2. IFAL
3. Cox, Lees and Ang, Classification of Hazardous Locations, IChemE 1990
4. Simmons, Daycock and Rew
5. Simmons, Daycock and Rew
6. Daycock and Rew
7. UK HSE
8. J.R.Taylor, QRAQ report 4, Release frequency data for QRA
9. Mr J Gummer & Dr S Hawksworth, Spontaneous ignition of hydrogen, Literature
Review Prepared by the Health and Safety Laboratory for the Health and Safety
Executive 2008
10. J.R.Taylor, QRA Open, an open software tool for QRA teaching and research, 2012,
ITSA.DK

© J.R.Taylor 2012
2010 24
24