Working together

for a safer world

Process leak for offshore

installations frequency assessment
model - PLOFAM(2)
Main report

Report for:
Equinor ASA

Report no: 107566/R1 Rev: Final

Date: 6 December 2018
Document history
Revision Date Description/changes Changes made by

Draft A 15.11.2018 This draft version is an update of the report Ingar Fossan and Are
105586/R1 released 18.03.2016, ref /1/ Opstad Sæbø

Final 06.12.2018 Comments from Equinor, Safetec and DNV GL, Ingar Fossan and Are
and Lilleaker Consulting AS are implemented Opstad Sæbø

Preface
Extensive work has been carried out during the recent years regarding models for estimating leak
frequencies and ignition probabilities for offshore facilities at the Norwegian continental shelf (NCS). This
has resulted in the PLOFAM (Process leak for offshore installations frequency assessment model) and
MISOF (Modelling of Ignition Sources on Offshore oil and gas Facilities) 2018 models.

The developed models seek to give a realistic and unbiased prediction of hydrocarbon process leaks and
ignitions for an average facility on the NCS for the coming years. Users of the models and their results
should however be aware of the following aspects:

• PLOFAM (2) is tuned to give the same number of leaks >0.1 kg/s as observed in historical data for
NCS in the period 2006 – 2017, and predicts significantly fewer leaks than previous models
• The MISOF (2) model will for most modules give higher ignition probabilities than previous models. It
builds on few ignited events, and the statistical uncertainty is therefore relatively high. The
contribution from external ignition may be essential in such regard

For some analysed offshore modules, the combined use of these models may result in no dimensioning
loads (ref. PSA’s Facility regulation §11). Each risk owner needs to decide how these aspects shall be
considered in their risk management.

Executive summary
Hydrocarbon process leaks are a major contributor to offshore risk. The last decade the industry has used
a model denoted “Offshore QRA - Standardised Hydrocarbon Leak Frequencies” (SHLFM) to estimate leak
frequencies for these incidents. This model originates from the JIP project “Standardised Hydrocarbon
Leak Frequencies”, which was first reported in final version in 2005. Based on experience from use of the
model, Equinor has appreciated the need for a thorough revision of the methodology, and initiated a
project where the purpose has been to create an updated leak frequency model that can be accepted as
an industry standard for the Norwegian Continental Shelf by consultancy companies and operators.

To achieve this, Equinor contracted Lloyd’s Register Consulting (LRC), DNV GL, Safetec and Lilleaker
Consulting AS to work together. In addition to the four consultancy companies, the operators
ConocoPhillips and Lundin were invited to the project. LRC has been the lead contractor while the others
have contributed as advisors through workshops, document review and discussions in meetings. The
project has been run in two phases during the periods March – December 2015 and June – December
2018. Personnel participating in workshops (in one or both phases) are listed in Table 1.1. Also other
subject matter experts have been involved in video conferences and discussions.

Report no: 107566/R1 Rev: Final Page ii

Date: 6 December 2018 ©Lloyd’s Register 2018
This report with technical notes documents the resulting leak frequency model, denoted PLOFAM (Process
leak for offshore installations frequency assessment model) that for most situations is expected to be the
preferred model by all above mentioned project participants. It is expected that this model will be used for
most QRA’s for Equinor, ConocoPhillips and Lundin.

PLOFAM is designed to be a tool for estimation of future leak frequencies for use in QRAs. Overall the
model is built on a combination of the explanatory variable that shows the strongest correlations with
experienced number of leaks, and rational explanations and causalities reflecting known failure modes.
The number of equipment (for each equipment type) is concluded to be the best single explanatory
variable to build the model on. However, as only one explanatory variable is chosen for the model, there
are many factors influencing the leak frequency that are not captured by the model, which will give rise to
stochastic effects. The historical leak frequency per installation at the NCS can vary significantly from the
NCS average and from the model prediction, as a result of the stochastic effects, and also if the conditions
at a particular installation deviates from the normal conditions at installations on NCS.

The leak frequency model covers process leaks occurring during all operation phases, and topside leaks
from the well system occurring during normal production. The leak scenarios may have a leak point
associated with well, process system (including fuel gas system) or utility systems. The leak frequency for
process leaks estimated by the model accounts for leaks occurring both in the process system and utility
system fed from the process system. The model does however not give separate leak frequencies for
process releases through utility systems and through process system. Three main leak scenarios are
defined for the leak frequency model. That is Process leak, Producing well leak and Gas lift well leak.
Furthermore, the model distinguishes between leak scenarios where the total released amount of
hydrocarbons is ≤10 kg, and >10 kg. These leaks are classified as Marginal leaks and Significant leaks,
respectively. Only the Significant leak scenario is relevant for detailed modelling of consequences and
dimensioning accidental loads in a formal QRA. The Marginal leak scenario is only relevant with regard to
immediate exposure of personnel in the close vicinity to scene of the leak to accidental loads or for small
poorly ventilated enclosures.

The model itself consists of mathematical equations for the frequency hole size distribution per standard
equipment type per equipment dimension. Thus, the model is equipment size dependent. A significant
effort has been made to build a model where both the total leak frequency and the frequency for
ruptures are equipment size dependent, unique for every standard equipment type, and as good as
possible reflects the most common failure modes. The model includes the following new equipment types
not included in SHLFM; compact flanges, steel piping, flexible piping, gas lift well, producing well and a
model for leaks from hoses used in temporary operations.

The strategy has been to build a model that gives a best estimate for future leak frequencies, i.e. to create
an unbiased model without built in conservatism. It is observed a significant decreasing trend in historical
leak frequency with time for installations on the NCS in the period after year 2000 (actually since 1992).
The number of historical leaks in the period 2006-2017 is used as target for the total leak frequency while
leak data from the period 2001 – 2017 is used as target for the relative leak rate distribution. Targeting
this frequency level would imply that the model will estimate about 30% lower leak frequency than the
average leak frequency in the period 2001 – 2017, but also 30% higher leak frequency than seen for any
years after 2011, i.e. the chosen target level for the model account for uncertainty in the data material
and shifts in underlying causal factors (e.g. emerging unknown degradation mechanisms due to age or
changing operational conditions) affecting the future trend in leaks occurring on installations on the NCS.
In total the combination of the targeted total leak frequency and the fraction of large leaks will decide the
targeted leak frequency for large leaks, and is regarded reasonable and as a best estimate, slightly
approached from the conservative side. Note also that conservatism is embedded in the guideline for use
of PLOFAM in QRAs.

The stochastic uncertainty has been quantified and is larger for large leak rates than for small leak rates.
This is important to consider when evaluating the accuracy of a QRA model based on PLOFAM. For leak
rates above about 30 kg/s, the relative stochastic uncertainty constitutes a factor in the range 1.5 to 2.5.
i.e. based on the historical data it can be argued that the target value used for parametrization of the
model can be both a factor 1.5 – 2.5 higher and lower than the target values used in PLOFAM (PLOFAM
targets the most likely value). As a consequence it is shown that if two leaks >100 kg/s where one of
them is larger than 300kg/s occur tomorrow, the model will still be valid.

Report no: 107566/R1 Rev: Final Page iii

Date: 6 December 2018 ©Lloyd’s Register 2018
The model validation is performed by applying the model to all installations on NCS being in operation in
the period 2006 – 2017. The results shows that PLOFAM is able to:

(1) Reproduce the total number of leaks at NCS in the period 2006 – 2017
(2) Reproduce the total cumulative leak rate frequency distribution (i.e. the leak rate distribution) seen in
historical data from NCS in the period 2001 – 2017, which is the defined target for the model,
when applied to all installations on NCS.
(3) Reproduce the observed contribution to leaks originating from the different equipment types. The
model does also reproduce the observed frequency distribution of leaks with respect to initial leak
rate for the most dominating equipment types at NCS (i.e. valves, flanges, instruments and steel
pipes).

The model is mainly validated towards available data of leaks that has occurred at installations on the
Norwegian Continental Shelf (NCS), but also data from the United Kingdom Continental Shelf (UKCS) has
been utilized where the data material for NCS is scarce. A main overall conclusion is that the underlying
hole size frequency distribution for equipment at installations located on the NCS is similar to the
distribution for equipment located on UK installations. The differences may be explained by uncertainty
related to the datasets (both the leaks and the population data, and the way equipment is counted and
leaks are assigned to equipment types). Furthermore, also the total frequency and time trend in the leak
frequency at UKCS is similar to the total leak frequency and time trend seen on NCS. The model is
therefore regarded as valid for both sectors.

Table 1.1 – Personnel participating in one or more workshops in both project phases. Also other persons
have been involved in video conferences, project meetings and discussions
Name Company Role
Phase 1 (2015)
Eli Bech Equinor Equinor project manager
Unni Nord Samdal Equinor Technical point of contact
Espen Fyhn Nilsen Equinor Technical point of contact
Marie Saltkjel ConocoPhillips Participant
Espen Skilhagen Lundin Participant
Robert Schumacher Lundin Participant
Are Opstad Sæbø Lloyds's Register Consulting Project manager/participant
Ingar Fossan Lloyds's Register Consulting Technical responsible
Erik Odgaard Lloyds's Register Consulting Quality assurer
Jan Pappas Lloyds's Register Consulting Participant
Jens Garstad DNV GL Participant
Andreas Falck DNV GL Participant
Jo Wiklund Lilleaker Consulting AS Participant
Jens Morten Nilsen Lilleaker Consulting AS Participant
Jon Andreas Rismyhr Safetec Participant
Geir Drage Berentsen Safetec Participant
Morten Skjong Safetec Participant
Ole Magnus Nyheim Safetec Participant

Report no: 107566/R1 Rev: Final Page iv

Date: 6 December 2018 ©Lloyd’s Register 2018
 Name Company Role
Phase 2 (2018)
Eli Bech Equinor Equinor project manager
Marie Saltkjel ConocoPhillips Participant
Are Opstad Sæbø Lloyds's Register Consulting Project manager/participant
Ingar Fossan Lloyds's Register Consulting Technical responsible
Jan Pappas Lloyds's Register Consulting Quality assurance
Jens Garstad DNV GL Participant
Jo Wiklund Lilleaker Consulting AS Participant
Jon Andreas Rismyhr Safetec Participant

Report no: 107566/R1 Rev: Final Page v

Date: 6 December 2018 ©Lloyd’s Register 2018
Glossary/abbreviations
Abbreviations and expressions used in the main report and all technical notes are given in TN-1.
Abbreviations relevant for the main report are repeated in Table 1.2. An important expression, frequently
used in the model, is the Complementary cumulative hole size frequency distribution. This expression
denotes frequency distributions F(hole size > d), where d is a specific hole size. This expression is
throughout the report denoted 𝐹𝐹, and for simplicity it is referred to as the hole size frequency distribution.
The complementary cumulative hole size probability distribution for an equipment type multiplied by the
total leak frequency for that equipment type, gives the complementary cumulative hole size frequency
distribution.

Table 1.2- Abbreviations used in main report and technical notes

Abbreviation Description
ACH Air change per hour
ASCV Annulus safety check valve
ASV Annulus safety valve
DHSV Downhole safety valve
ESD Emergency shut down
HCRD Hydrocarbon release database
HSE Health and safety executive
LRC Lloyd’s Register Consulting
Modelling of Ignition Sources on Offshore oil and gas
MISOF
Facilities
NCS Norwegian continental shelf
Process Leak for Offshore installations Frequency
PLOFAM
Assessment Model
P&ID Piping and instrumentation diagram/drawing
Ptil Petroleumstilsynet (Petroleum safety authority)
PWV Production wing valve
QRA Quantitative risk analysis
Risikonivå i norsk petroleumsvirksomhet (Risk level in
RNNP
Norwegian petroleum industry)
SHLFM Standardised hydrocarbon leak frequencies model
TN Technical note
UKCS United kingdom continental shelf

Report no: 107566/R1 Rev: Final Page vi

Date: 6 December 2018 ©Lloyd’s Register 2018
Table of contents Page

1 Introduction .....................................................................................................................................1
1.1 Report structure ....................................................................................................................1
1.2 Objective ...............................................................................................................................2
2 Philosophy for model development and expectations to the model ..................................................2
3 Leak scenarios covered by the model ...............................................................................................2
4 Model summary and application of the model .................................................................................6
4.1 Mathematical formulation .....................................................................................................6
4.2 Equipment types covered by the model .................................................................................8
4.3 Application of the model .......................................................................................................9
4.4 Example of application of the model ...................................................................................10
5 Data basis ......................................................................................................................................11
5.1 NCS data .............................................................................................................................11
5.2 UKCS data ...........................................................................................................................12
6 Model parameterisation and validation ..........................................................................................13
6.1 Parametrization methodology..............................................................................................13
6.2 Overall validation .................................................................................................................15
7 PLOFAM parameters ......................................................................................................................20
8 Comparison of PLOFAM and SHLFM ..............................................................................................21
9 Robustness of PLOFAM ..................................................................................................................23
10 Summary and concluding remarks .................................................................................................26
11 References .....................................................................................................................................28

Report no: 107566/R1 Rev: Final Page vii

Date: 6 December 2018 ©Lloyd’s Register 2018
1 Introduction
This report describes the leak frequency model, denoted PLOFAM (Process leak for offshore
installations frequency assessment model), and used for estimation of topside process leak
frequencies for use in Quantitative Risk Analysis of fire and explosion at installations located on
the Norwegian Continental Shelf.
The model is in general fully documented in the technical notes (TN) listed in Chapter 1.1. This
main report presents the most important aspects of the model, without presenting all details, but
with a sufficient level of detail to gain an overview of the model. For further details it is referred
to the TNs throughout the report.
The project has been run in two phases during the periods March – December 2015 and June –
December 2018. The first version was issued in March 2016, Ref. /1/, and documents the
concluded model in the first phase. This report is the second version of PLOFAM and documents
the model after the revision in the second phase. The main changes from the first phase are:
• Data for the period 2015 – 2017 included in the database
• Relative leak rate distribution reassessed based on leaks on NCS in the period 2001 – 2017 as
opposed to 2001 – 2015 In the first version. Relative leak rate distributions including data
from the period 1992 - 2000 has also been assessed.
• Population data base used for validation/parametrization increased from 62 to 109
installations, including all installations that have been in operation on NCS.
• Failure modes for valves, flanges and instruments are discussed together with experts in
Equinor. This resulted in an updated mathematical model for leak frequency distribution and
reduced rupture fraction for valves and flanges, and updated guidelines for instruments.
• The model for hose leaks is re-assessed giving reduced leak frequency for leaks giving large
released quantities
• The data base and hence the model is now fully aligned with the MISOF model, Ref. /2/. The
models are based on the same assessment of the historical leak scenarios and are therefore
interlinked. In particular, the number of large leaks has a significant effect on both models.
Note however that if the number of large leaks is misinterpreted in the data material and
should have been higher (i.e. that the leak frequency for large leaks should have been higher)
then the ignition probability would have been lower in MISOF. This demonstrates why the
two models should be used together and not combined with other models. Using the models
together ensures consistent interpretation of scenarios having impact on both models,
ensuring a best estimate for the fire and explosion frequency as well as a consistent estimate
the uncertainties in these frequencies

1.1 Report structure

The report consists of the following technical notes:
• TN-1 Expressions and abbreviations
• TN-2 NCS data
• TN-3 UKCS data
• TN-4 Leak scenarios
• TN-5 Leak frequency model and Guideline for use of PLOFAM in QRAs
• TN-6 Model parametrization and validation

Report no: 107566/R1 Rev: Final Page 1

Date: 6 December 2018 ©Lloyd’s Register 2018
 1.2 Objective
The objective of the leak frequency model is to serve as a tool for prediction of the future leak
frequency for topside process leaks at installations located on the Norwegian Continental Shelf
(NCS) for use in QRAs. The model should be unbiased, i.e. it should aim at a best estimate.
However the best estimate should be approached slightly from the conservative side.

2 Philosophy for model development and expectations to

the model
The reasons for leaks occurring from process systems at offshore installations are diverse and
many, and hence there is a large number of factors that influence the leak frequency. Such
factors may be the components that the process system consists of, the equipment size
distribution, the process conditions, the environment around the process system, the
maintenance scheme, training of personnel, work culture and time and cost requirements. Many
of these factors will be different from installation to installation and some will strongly influence
the leak frequency, while other will only to some extent have implications on the leak frequency.
When building a model serving as a tool for prediction of future leak frequency for topside
process leaks in QRAs, it is obvious that all factors influencing the leak frequency cannot be
included. Building a model for such a complex phenomenon will be a trade-off between model
complexity, user friendliness of the model, and the model’s ability to predict good overall
estimates for single installations. The model should therefore capture the “most important”
contributing factors to topside process leaks in order to reflect the most important differences
between the installations. The “less important” contributing factors, not included in the model,
will give rise to stochastic effects, i.e. comparing the predicted number of leaks (by the model)
and historical leaks for every single installation must be expected to show stochastic behaviour.
The reasons for leaks occurring are many and normally all factors that resulted in an observed
leak cannot be fully understood. However, some failure modes can be understood, and in such
cases these known failure modes should be aimed reflected in the model.
Based on the above, and a more thorough discussion given in TN-5, the following important
philosophy for building the model is established: Overall the model should be built on a
combination of the parameter that shows the strongest correlations with experienced number of
leaks, and rational explanations and causalities reflecting known failure modes.
In PLOFAM the number of equipment (for each equipment type) is the only explanatory variable
assumed in the model. In addition, known failure modes are reflected in the parametrization of
the model. Note that this does not mean that it is concluded that the number of equipment is
the only factor having implications on the leak frequency, but it is concluded to be the best single
explanatory variable. However, as only one explanatory variable is chosen for the model, there
will be many factors influencing the leak frequency that are not captured by the model, which
will give rise to stochastic effects.

3 Leak scenarios covered by the model

The leak frequency model covers process leaks and topside leaks from the well system occurring
during normal production. A detailed description of system boundaries and scenarios covered by
the model is given in TN-4.
The leak scenarios covered by the model may have a leak point associated with well, process
system (including fuel gas system) or utility systems, and are described in Table 3.1. Other leak
scenarios, such as leaks from utility systems fed from utility systems (for example diesel from
diesel tanks and MEG from MEG-system) are not included.

Report no: 107566/R1 Rev: Final Page 2

Date: 6 December 2018 ©Lloyd’s Register 2018
 Note that the leak frequency for process leaks estimated by the model does also account for
leaks occurring in the utility system, but being fed from the process system. This is done by
including process leaks fed through utility systems, but not equipment counts from utility systems
as basis for the model validation. This implies that utility equipment should not be counted as
basis for estimation of process leak frequencies. Furthermore the model does not give separate
leak frequencies for process releases through utility systems and through process system. This
means that a QRA based on PLOFAM will not reflect the potential location of the leak sources in
utility systems. Furthermore, the leak frequency contribution from utility systems will scale with
the number of equipment counts for process system. This contribution will in practice vary
somewhat with the system at hand, but this cannot be quantified based on PLOFAM. A detailed
risk assessment of leaks in utility systems, if found required, should hence be covered by special
evaluations. Figure 3.2 gives an illustration of leak scenarios normally considered in a QRA. The
figure shows which scenarios that are covered by the model and which that are not.
Incidents occurring during well interventions/operations, such as wire line and coiled tubing, are
defined as blowouts or well releases, and are covered by Ref. /3/ that is based on the SINTEF
Offshore Blowout Database. These incidents are not covered by the model.

Table 3.1 - Leak scenarios covered by the model. They occur in well system, process system or utility
system (process leaks fed through utility systems). Scenarios that are not listed in this table are not
covered by the model
Leak point in well system Leak point in process Leak point in utility system
system
1. Producing well/Injection 4. Leak point in 5. Leak point in flare system (low
well: Topside well release process system pressure or high pressure flare
where the inventory bet- between PWV and system)
ween DHSV and PWV is topside riser ESDV/- 6. Excessive releases through flare
released during normal storage ESDV. The tips and atmospheric vents that
production. fuel system is exceed the design specification
2. Gas lift well: Topside well regarded as part of and pose a fire and explosion
release where the inven- the process system. hazard to equipment, structures
tory between the ASV or personnel. Such leaks are de-
and the barrier towards noted vent leaks.
the process system is 7. Leak point in utility systems that
released. In cases where is fed by hydrocarbons stemm-
no ASV is present, the ing from process system.
entire inventory in the gas Systems covered by the model
lift annulus to the ASCV are:
may be released. Assu-
a. Open drain system
ming that the check valve
ASCV is functioning, b. Closed drain system
otherwise there is no c. Chemical injection systems.
barrier towards the reser- d. Produced water
voir.
3. Release of hydrocarbon
fluid from annuli that are
not used for gas lift.

Three main leak scenarios for modelling in QRAs are defined in PLOFAM. That is Process leak,
Producing well leak and Gas lift well leak.

Report no: 107566/R1 Rev: Final Page 3

Date: 6 December 2018 ©Lloyd’s Register 2018
 For all leak scenarios, 0.1 kg/s is recommended as the general leak rate threshold for estimation
of leak duration (both in terms of calculation of fluid dispersion and fire duration) in a QRA, for
3
all leak scenarios in open areas and leaks in enclosures having a net volume more than 1,000 m
and with ventilation rate of 12 ach or higher (see TN-4). The lower leak rate threshold is put as
basis for the lower boundary with regard to aggregated released amount of hydrocarbons (10
kg). The model distinguishes between leak scenarios (rate > 0.1 kg/s) where the total released
amount of hydrocarbons is ≤10 kg, and >10 kg. These leaks are classified as Marginal leaks and
Significant leaks, respectively.
In a QRA, the risk in terms of fire- and explosion load exposure to vulnerable equipment and
structures such as safety systems, pressurized equipment, load carrying structures and main
safety functions, associated with Marginal leaks can normally be neglected. However, the risk to
personnel associated with Marginal leaks should not be neglected.
The three main leak scenarios for modelling in QRAs are summarized in Table 3.2, and in Figure
3.1. The table shows how the three main leak scenarios for modelling in QRAs relate to the leak
scenarios in Table 3.1. The figure shows how the leak scenarios in PLOFAM relate to the leak
scenarios in SHLFM, Ref. /4/.

Table 3.2 - Leak scenarios suggested for QRAs at NCS

Modelled leak scenario Leak scenarios included
Significant Scenario 4-7 in Table 3.1, released quantity >10 kg
Process leak
Marginal Scenario 4-7 in Table 3.1, released quantity ≤10 kg
Production well Significant Scenario 1 and 3 in Table 3.1, released quantity >10 kg
1
leak Marginal Scenario 1 and 3 in Table 3.1, released quantity ≤10 kg
Gas lift well leak Significant Scenario 2 and 3 in Table 3.1, released quantity >10 kg
Marginal Scenario 2 and 3 in Table 3.1, released quantity ≤10 kg

Figure 3.1 - Illustration and summary of the leak scenarios to be modelled in a QRA based on
PLOFAM, together with the leak scenarios Full pressure leaks (Limited leaks and Full leaks) and
Zero pressure leaks defined in the SHLFM Ref. /4/

1
The frequency for production wells and injection wells are considered to be identical. The leak scenario is
denoted production well only

Report no: 107566/R1 Rev: Final Page 4

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 3.2 - Illustration of leak scenarios normally considered in a QRA. The figure shows which scenarios that are covered by the model and which that
are not

Report no: 107566/R1 Rev: Final Page 5

Date: 6 December 2018 ©Lloyd’s Register 2018
4 Model summary and application of the model
PLOFAM is based on the assumption that the leak frequency is proportional to the number of
each type of equipment. This assumption is justified in TN-5, where it is also assessed to what
extent the level of operational activity on an installation contributes to leaks. However, for
reasons presented in TN-5 (Chapter 3 and 4), the number of equipment (for each equipment
type) is the only explanatory variable implemented in the model.
This chapter summarizes the concluded mathematical formulation of the model. Furthermore,
equipment types included in the model and a description of how to use the model, including an
example is given. Further details including the rationale and detailed description of the model is
given in Chapter 2 in TN-5.

4.1 Mathematical formulation

The general formulation of the mathematical equations for the complementary cumulative hole
size frequency distribution 𝐹𝐹 (i.e. the frequency for hole diameter equal to or larger than 𝑑𝑑
millimetres, given equipment diameter 𝐷𝐷 in millimetres, see also TN-1 for definition) valid for a
unique equipment type, which for simplicity is referred to as the hole size frequency distribution,
is:

[𝐹𝐹 (𝐷𝐷) − 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷)] ∙ 𝑑𝑑 𝑚𝑚(𝐷𝐷) + 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷) , 1 ≤ 𝑑𝑑 ≤ 𝐷𝐷 (1)

𝐹𝐹(𝑑𝑑, 𝐷𝐷) = � 0
0 , 𝑑𝑑 > 𝐷𝐷

𝐹𝐹0 (𝐷𝐷) = 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 ∙ (𝐴𝐴0 ∙ 𝐷𝐷 𝑀𝑀0 ) (2)

𝐹𝐹𝐷𝐷 (𝐷𝐷) = 𝐹𝐹0 (𝐷𝐷) ∙ (𝐴𝐴𝐷𝐷 ∙ 𝐷𝐷 𝑀𝑀𝐷𝐷 + 𝐵𝐵𝐷𝐷 ) (3)

log(𝐹𝐹D − 𝛼𝛼 ∙ 𝐹𝐹D ) − log(𝐹𝐹0 − 𝛼𝛼 ∙ 𝐹𝐹D ) (4)

𝑚𝑚 (𝐷𝐷) =
log(𝐷𝐷)

The parameters in the equations above are described in Table 4.1. Note that except for the
parameters 𝑑𝑑 and 𝐷𝐷, all parameters are in general unique for every equipment type, even though
this is not reflected in the mathematical formulation above. A list of unique parameter values for
𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 , 𝐴𝐴0 , 𝑀𝑀0 , 𝐴𝐴𝐷𝐷 , 𝑀𝑀𝐷𝐷 , 𝐵𝐵𝐷𝐷 and 𝛼𝛼 necessary to estimate leak frequencies for every equipment type
included in the model are given in Table 7.1. 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 is the historical leak frequency (given in leaks
per year per piece of equipment), for the relevant equipment. The other parameters are
dimensionless model parameters. The subscript “0” is used to indicate the total leak frequency
for an equipment and hence the “starting point” on the y-axis. The subscript D is used to
indicate the frequency for getting a hole diameter equal to the equipment diameter 𝐷𝐷. Both 𝐹𝐹0
and 𝐹𝐹𝐷𝐷 are in general dependent on the equipment diameter 𝐷𝐷, which is indicated in the
parenthesis:

𝐹𝐹0 (𝐷𝐷) = 𝐹𝐹(𝑑𝑑 = 1, 𝐷𝐷) (5)

𝐹𝐹𝐷𝐷 (𝐷𝐷) = 𝐹𝐹(𝑑𝑑 = 𝐷𝐷, 𝐷𝐷) (6)

Report no: 107566/R1 Rev: Final Page 6

Date: 6 December 2018 ©Lloyd’s Register 2018
 In short the model described above can be summarized to be built up of the following parts
1. Modelling of the total leak frequency per equipment, 𝐹𝐹0 . To model the equipment diameter
dependency of 𝐹𝐹0 , the parameters 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 , 𝐴𝐴0 and 𝑀𝑀0 are used.
2. Modelling of the full bore hole frequency 𝐹𝐹𝐷𝐷 . To model the equipment diameter dependency
of 𝐹𝐹𝐷𝐷 , the parameters 𝐴𝐴𝐷𝐷 , 𝑀𝑀𝐷𝐷 and 𝐵𝐵𝐷𝐷 are used.
3. Modelling of the cumulative frequency for hole diameters 𝑑𝑑 in the interval 1 < 𝑑𝑑 < 𝐷𝐷. This
model is built up of a power law modelling the hole size dependency, and an additional
frequency for full bore hole leaks:
a. The model assumes that the hole size dependent part of the hole size frequency
distribution follows a power law, that “starts” at (𝐹𝐹0 (𝐷𝐷) − 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷)) for 𝑑𝑑=1and
“ends” at (𝐹𝐹D (𝐷𝐷) − 𝛼𝛼 ∙ 𝐹𝐹D (𝐷𝐷)) for 𝑑𝑑 = 𝐷𝐷. The formula for the slope parameter 𝑚𝑚(𝐷𝐷) in
Eq. (4) follows from the assumption that the hole size dependent part of the hole size
frequency distribution (becoming the first part in Eq. (1)) follows a power law with
“start” and” end” points as described.
b. The last term in Eq. (1), which is the product of 𝐹𝐹𝐷𝐷 and the parameter 𝛼𝛼 ∈ [0, 1⟩, is
introduced to capture the effect that the frequency for hole diameters close to the
equipment diameter is expected to be even lower than estimated by the normal power
law for some failure modes. The parameter 𝛼𝛼 is the fraction of the full bore hole
frequency that is added in the second term in Eq. (1). This parameter only influences the
frequency for hole diameter in the interval 1 < 𝑑𝑑 < 𝐷𝐷, while the total leak frequency 𝐹𝐹0
and the full bore hole frequency 𝐹𝐹𝐷𝐷 are unaffected. The net effect of a non-zero 𝛼𝛼 is to
shift more of the leak frequency towards smaller holes compared to 𝛼𝛼 = 0.
In addition to the equations described above, the parameter 𝐹𝐹1 is introduced and expressed as a
function of 𝐹𝐹𝐷𝐷 as follows, and can be substituted in the equations above when convenient:

𝐹𝐹1 (𝐷𝐷) = 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷) , 𝛼𝛼 ∈ [0, 1⟩ (7)

𝐹𝐹1 is useful both when implementing the model and when describing the rationale for the model
(see TN-5 Appendix A).
For a further detailed description of the rationale for the model, derivation of the expression for
the slope parameter and illustrations, it is referred to TN-5 Appendix A. The appendix also
compares the model with the previous leak frequency model used in the industry, denoted
SHLFM, Ref. /4/, and explains the difference.

Table 4.1 - Summary of all parameters used for each equipment type in the model. Except for the
parameters 𝑑𝑑 and 𝐷𝐷, all parameters are in general unique for every equipment type. Note that not
all parameters are included in the above equations. Some are introduced later in the report.

Parameter Description

𝐹𝐹(𝑑𝑑, 𝐷𝐷) Hole size frequency distribution (see TN-1) [year-1 equipment-1].

𝐹𝐹0 Total leak frequency [year-1 equipment-1]. The subscript 0 is used to

indicate the total leak frequency for an equipment and hence the “starting
point” on the y-axis.
𝐹𝐹0 = 𝐹𝐹(𝑑𝑑 = 1, 𝐷𝐷).

𝐹𝐹𝐷𝐷 The total full bore hole frequency [year-1 equipment-1]. The subscript D is
used to indicate the frequency for getting a hole equal to the equipment
diameter 𝐷𝐷.
𝐹𝐹𝐷𝐷 = 𝐹𝐹(𝑑𝑑 = 𝐷𝐷, 𝐷𝐷).

Report no: 107566/R1 Rev: Final Page 7

Date: 6 December 2018 ©Lloyd’s Register 2018
 Parameter Description

𝜃𝜃(𝐷𝐷) The full bore fraction of total leak frequency

𝐹𝐹𝐷𝐷 (𝐷𝐷)
𝜃𝜃(𝐷𝐷) =
𝐹𝐹0 (𝐷𝐷)

𝑑𝑑 Hole diameter in millimetres

𝐷𝐷 Equipment diameter in millimetres

𝑚𝑚 Slope parameter

𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 The average leak frequency (independent of equipment diameter) for the
-1 -1
relevant equipment type [year equipment ]

𝐴𝐴0 Parameter in equation for total leak frequency, 𝐹𝐹0

𝑀𝑀0 Parameter in equation for total leak frequency, 𝐹𝐹0

𝐴𝐴𝐷𝐷 Parameter in equation for full bore hole frequency, 𝐹𝐹𝐷𝐷

𝑀𝑀𝐷𝐷 Parameter in equation for full bore hole frequency, 𝐹𝐹𝐷𝐷

𝐵𝐵𝐷𝐷 Parameter in equation for full bore hole frequency, 𝐹𝐹𝐷𝐷

Dimensionless parameter, independent of equipment diameter 𝐷𝐷, 𝛼𝛼 ∈ [0, 1⟩.

𝛼𝛼 The fraction of the full bore frequency that comes from the second term in
Eq. (1)
-1 -1
𝐹𝐹1 Additional full bore hole frequency [year equipment ]

4.2 Equipment types covered by the model

In total 20 different equipment types are covered by the model, including Gas lift well and
Production well, which belongs to the well system. The other equipment types included in the
model are the most common process equipment types at offshore installations. All equipment
types covered by the model are given in Table 4.2.

Table 4.2 - Equipment types included in the model

Equipment type Additional description

Air-cooled heat exchanger
Atmospheric vessel Vessels with atmospheric pressure
Centrifugal compressor
Centrifugal pump
Compact flange
Filter
Flexible pipe Permanently installed hose
Hose Temporary hoses
Instrument

Report no: 107566/R1 Rev: Final Page 8

Date: 6 December 2018 ©Lloyd’s Register 2018
 Equipment type Additional description
Pig trap Pig launchers and pig receivers
Plate heat exchanger
Process vessel Pressurized process vessels
Reciprocating compressor
Reciprocating pump
Shell and tube side heat Includes equipment where the hydrocarbon is on the shell
exchanger side and/or tube side of the heat exchanger
Standard flange Includes all flange types, except compact flanges
Steel pipe Process steel pipe
Valve Includes all types of valves
Gas lift well Well head with gas lift.
Producing well Well head with or without gas lift

4.3 Application of the model

When applying the model on a specific installation the first step is to define the desired leak rate
intervals defined by the leak rates 𝑞𝑞1 < 𝑞𝑞2 < ⋯ < 𝑞𝑞𝑁𝑁 , or the desired hole size intervals defined by
𝑑𝑑1 < 𝑑𝑑2 < ⋯ < 𝑑𝑑𝑁𝑁 . Next the following procedure is suggested for all equipment types on the
installation:
1. Calculate 𝐹𝐹0 (𝐷𝐷) for the relevant equipment types and dimensions using equation (2)
2. Calculate 𝐹𝐹𝐷𝐷 (𝐷𝐷) for the relevant equipment types and dimensions using equation (3)
3. Calculate 𝐹𝐹1 (𝐷𝐷) = 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷) for the relevant equipment using equation (7). This step is not
necessary, but may be convenient.
4. Calculate 𝑚𝑚(𝐷𝐷) for the relevant equipment using equation (4)
5. If leak frequencies are calculated for leak rate intervals: For each piece of equipment (or
group of equipment with the same process conditions) calculate the hole size intervals
defined by 𝑑𝑑1 < 𝑑𝑑2 < ⋯ < 𝑑𝑑𝑁𝑁 corresponding to the defined leak rate intervals 𝑞𝑞1 < 𝑞𝑞2 < ⋯ <
𝑞𝑞𝑁𝑁 based on appropriate equations for modelling of leak rate. The models used for
estimating release rates should be carefully chosen based on fluid composition and process
conditions (e.g. pressure, composition and temperature). TN-5 Appendix B gives relations for
gas and liquid leak rate estimations. Since the leak frequency model is defined for hole
diameters >1 mm only, it is recommended to set 𝑑𝑑𝑖𝑖 to 1 if the calculated hole size is < 1mm.
Thus leak frequencies for hole diameters less than 1 mm is not included. This will in general
not affect results in most QRA’s as leaks around 1 mm will produce small release rates (< 0.1
kg/s). In some cases, leaks having a release rate less than 0.1 kg/s ought to be assessed in the
QRA to model the risk picture with adequate precision (e.g. enclosures with poor ventilation,
and release of poisonous gases). A special assessment of leaks with an initial leak rate less
than 0.1 kg/s has to be performed in such cases.
6. Calculate leak frequencies for all hole diameters 𝑑𝑑1 , 𝑑𝑑2 … 𝑑𝑑𝑁𝑁 , for the relevant equipment
based on the equation (1) given in Chapter 4.1.
7. Calculate the frequency 𝐹𝐹(𝑑𝑑𝑖𝑖 < 𝑑𝑑 < 𝑑𝑑𝑖𝑖+1 ) = 𝐹𝐹(𝑑𝑑𝑖𝑖 ) − 𝐹𝐹(𝑑𝑑𝑖𝑖+1 ) for the relevant equipment for
the hole size intervals and/or leak rate intervals

Report no: 107566/R1 Rev: Final Page 9

Date: 6 December 2018 ©Lloyd’s Register 2018
 8. Multiply the leak frequencies with the number of equipment for the relevant equipment type
and dimension with the same process conditions. The number of equipment on an
installation should be estimated based on equipment counting on P&ID’s or similar. The
exception is hoses where the frequency is multiplied with the number of hose operations,
and steel pipe and flexible pipe where the frequency is multiplied with the number of steel
pipe meters/flexible pipe meters (see also item 9 below). The number of hose operations
must be clarified with the operator of the installation. A guideline for use of PLOFAM in
QRAs is given in TN-5 Appendix B.
9. In cases where the contribution from steel pipes is not assessed based on the length of steel
pipes in the process system, but rather on a general assessment of the expected fraction of
leaks stemming from steel pipes, this fraction must be added to the estimated leak
frequency. See Appendix B for guidance. It is also referred to TN-2 for an assessment of the
fraction of leaks at NCS stemming from steel pipes.

4.4 Example of application of the model

This example is given to illustrate the recipe given in Chapter 4.3. The frequencies for a
“Significant leak” (see Chapter 2) from a 4” standard flange, containing gas with density 132
kg/m3 at pressure 156 bara is calculated. Steps 1-7 are followed to estimate the leak frequency
distribution for this piece of equipment.
The model parameters for Standard flange are given in Table 4.3. Following step 1- 4 in Chapter
4.3, gives 𝐹𝐹0 (𝐷𝐷), 𝐹𝐹𝐷𝐷 (𝐷𝐷), 𝐹𝐹1 (𝐷𝐷) and 𝑚𝑚(𝐷𝐷) as given in Table 4.4.
By using the relation between hole size and leak rate the given in TN-5 Appendix B, the hole
diameters corresponding to 0.1, 0.5, 1, 5, 10 and 30 kg/s can be calculated as in step 5 in
Chapter 4.3. The results are given in Table 4.5 together with the corresponding cumulative leak
frequency calculated following step 6 in Chapter 4.3.
Next the leak frequency for the leak rate intervals and corresponding hole size intervals can be
calculated following step 7 in Chapter 4.3. The results are given in Table 4.6.

Table 4.3 – Model parameters for Standard flange

Equipment Fhist,sign
A0 M0 AD MD BD α
type
Standard
1 0 18 -1.45 0.005 0.5 2.50E-05
flange

Table 4.4 – Calculated total leak frequency 𝐹𝐹0 (𝐷𝐷 = 101.6), rupture frequency 𝐹𝐹𝐷𝐷 (𝐷𝐷 = 101.6),
𝐹𝐹1 (𝐷𝐷 = 101.6) = 𝛼𝛼 ∙ 𝐹𝐹𝐷𝐷 (𝐷𝐷 = 101.6) and slope parameter 𝑚𝑚(𝐷𝐷 = 101.6) for the 4” standard flange.
Equipment
𝐹𝐹0 𝐹𝐹𝐷𝐷 𝐹𝐹1 𝑚𝑚
type
Standard
2.50E-05 6.79E-07 3.39E-07 -0.93
flange

Report no: 107566/R1 Rev: Final Page 10

Date: 6 December 2018 ©Lloyd’s Register 2018
 Table 4.5 – Leak rates, corresponding hole sizes and cumulative leak frequency
Leak rate [kg/s]
Parameter 0.1 0.5 1 5 10 30
Hole size [mm] 2.22 4.97 7.02 15.71 22.21 38.47
Cumulative leak
frequency, F(d> hole
1.21E-05 5.92E-06 4.38E-06 2.26E-06 1.73E-06 1.17E-06
size) [per year per
equipment]

Table 4.6 - Leak rate intervals, corresponding hole size intervals and leak frequency

Leak rate interval [kg/s]

Parameter 0.1 - 0.5 0.5 - 1 1-5 5 - 10 10 - 30 >30
Hole size interval 2.22 - 4.97 - 7.02 - 15.71 - 22.21 -
>38.47
[mm] 4.97 7.02 15.71 22.21 38.47
Leak frequency [per
6.19E-06 1.53E-06 2.13E-06 5.27E-07 5.55E-07 1.17E-06
year per equipment]

5 Data basis
The model has been developed, parameterised and validated towards data gathered from two
sources of data:
• NCS data: 254 incidents recorded at all installations located on the NCS in the period
01.01.2001 – 31.12.2017
• UKCS data: 4561 incidents at installations on the UKCS recorded in HCR database in the
period Q3 1992 - Q1 2015
The NCS and UKCS databases are described in detail in TN-2 and TN-3, respectively. A short
review is given in Chapter 5.1 and 5.2.

5.1 NCS data

Population data has been collated for 85 installations based on equipment counts extracted from
the QRAs for the installations. 6 out of the 85 installations had not been set in operation by
31.12.2017. Hence, population data is available for 79 installations being in operation in the peri-
od 01.01.2001 – 31.12.2017 (full period or part of it).
For the remaining 25 installations, where equipment counts have not been available, the
population data (i.e. equipment counts) has been estimated by defining an equivalent installation
in the NCS population dataset. The equivalent installation has been based on an overall
evaluation of the installation characteristics. Only 11 out of the 25 installations have been in
operation in the period 01.01.2001 – 31.12.2017, while 14 installations have either been
decommissioned before 2001 (13 installations) or not been set in operation yet (1 installation). In
total the population data set consist of 109 installations where 90 of them have been in
operation in the period 2001 – 2017, and is denoted “NCS population dataset”.
The data basis of recorded leaks at the NCS has been established based on the following data
sources:
1. RNNP dataset collated by Petroleumstilsynet (Ptil) and Safetec
2. Review of accident investigation reports. Accident investigation reports have been available
for the major fraction of the incidents

Report no: 107566/R1 Rev: Final Page 11

Date: 6 December 2018 ©Lloyd’s Register 2018
 Recorded leaks at NCS have all an initial hydrocarbon leak rate of 0.1 kg/s or larger.
The total number of leaks reported in RNNP in the period 01.01.2001 – 31.12.2017 is 260. After
review of the incidents, it has been concluded that 43 of those incidents are not relevant for the
leak scenarios to be modelled by PLOFAM, i.e. they are not process leaks or topside well leaks
during normal operation (see Chapter 2). Typical properties of disregarded incidents are as
follows:
• The leak is a release through a vent or a dump line where the rate is not considered to
exceed the design specification for the vent or dump line
• The leak is originating from a piece of equipment not being covered by the model, such as a
pipeline or a riser
• The leak is occurring in the well system during a drilling operation or intervention
Out of the remaining 217 leaks (260 - 43), 210 leaks have occurred on the 85 installations in the
“NCS population dataset” where equipment counts have been performed. The remaining 7 leaks
have occurred on the 11 installations where equipment counts are established based on
equivalent installations. Detailed information about all 217 relevant leaks is given in TN-2
Appendix A.

5.2 UKCS data

Information about offshore releases of hydrocarbons at United Kingdom Continental Shelf
(UKCS), are collected in Hydrocarbon Release Database (HCRD). The database is operated by
Health and Safety Executive (HSE).
In total 4863 events occurring in the period Q3 1992 - Q1 2015 are recorded in HCRD. Not all of
the incidents are relevant for the defined leak scenarios (see TN-4). A thorough analysis has been
necessary to extract the relevant incidents for the model. The resulting databasis consist of 2855
recorded incidents from the period Q3 1992 - Q1 2015, and 1597 recorded incidents from the
period Q1 2001 - Q1 2015 that are fed through process systems. By also taking process leaks fed
through utility systems and topside leaks from well systems during normal operation into account,
the total number of leaks considered is 3208 for the period ( Q3 1992 - Q1 2015. The number of
relevant leaks at UKCS installations in the period Q2 2015- Q4 2017 is 210. The total number of
leaks at UKCS installations thus becomes 3318.
The UKCS historical data extracted from the HCR database has not been used directly when
setting the leak frequency model parameters. The UKCS data, with its uncertainties, does
(however) nevertheless, constitute an important data basis when evaluating certain aspects on a
higher level, such as:
• the relative distribution of leaks on the various types of equipment
• the relative distribution of leaks in terms of the initial leak rate, e.g. the fraction large vs.
small leaks
• the relative distribution of leaks equivalent with the leak scenario modelled in QRA’s (leak
from a fully pressurized isolatable process segment during normal operation) and leaks from
initially isolated and/or depressurized segments (in PLOFAM denoted ‘Significant’ and
‘Marginal’ leaks respectively)
• the time trend of observed leaks at UKCS demonstrating a downward trend from the initial
years levelling out around year 2010 to around 10 leaks per year
The UKCS data is also important for our confidence in the performance of the PLOFAM model
based on NCS data. The PLOFAM parameters derived based on NCS data generate a good fit to
the UKCS data when accounting for the uncertainties in the UKCS data. The observed deviations
are very likely to be explained by differences in counting rules and in the general quality of the
UKCS data (such as lack of consistency in the way incidents are logged, inconsistency in the
logged hole size/leak rate and higher uncertainties related to the population data). Taking this
into account, the underlying leak frequency at installations located on the UKCS appears to be
the same as the underlying frequency at installations on the NCS (see also TN-3).

Report no: 107566/R1 Rev: Final Page 12

Date: 6 December 2018 ©Lloyd’s Register 2018
6 Model parameterisation and validation
The parametrisation and validation process is interlinked, as the target for the model is defined
based model performance when applying PLOFAM to all installations being in operation at NCS
in period 2006-2017, and the parametrization is performed as an iterative process where these
results are assessed towards the model targets and adjusted to give satisfactory results.
However, the overall validation of the model also includes other assessments than defined by the
model targets. Both the parametrization process and the overall model validation is described in
this chapter.

6.1 Parametrization methodology

The starting point for the parametrization of the model is the model parameters established in
the first version of PLOFAM, Ref. /1/. Next the parametrization of the model is performed as an
iterative process consisting of the following step:
• Knowledge on failure modes driving the occurrence of leaks for the various equipment types
are applied and reflected in the parameters. This consist of reflecting known failure modes
for specific equipment types in the total leak frequency 𝐹𝐹0 (𝐷𝐷), the full bore hole frequency
𝐹𝐹 (𝐷𝐷)
𝐹𝐹𝐷𝐷 (𝐷𝐷) and the associated full bore hole fraction (rupture fraction) 𝜃𝜃(𝐷𝐷) = 𝐷𝐷 and 𝛼𝛼 (see
𝐹𝐹0 (𝐷𝐷)
also TN-5). An example is given below for valves, where 𝐹𝐹0 (𝐷𝐷), 𝐹𝐹𝐷𝐷 (𝐷𝐷) and 𝜃𝜃(𝐷𝐷) are plotted
for different values of the equipment diameter. The failure modes for the most important
equipment types (valve, standard flange and instrument) were discussed together with
subject matter experts in Equinor. Some equipment types defined in the model consist of
several equipment types with different failure modes. This is for example the case for the
equipment type standard flange which includes clamp connectors (e.g. Grayloc) and flanges
with different sealing designs (ASME ring joint and raised face). Due to variability within the
equipment category known failure modes may be more difficult to reflect in the joint model
parameters for all types within the category. For more homogenous equipment categories,
such as compact flanges, parameterisation based on failure modes is more straight forward.
• The model is applied to all installations being in operation on NCS in the period 2006 – 2017
(86 installations), and model performance is compared with the defined targets for the
model. The model is also applied to all installations being in operation on NCS in the period
2001 – 2017 (90 installations). The defined targets for the model are described in TN-5 and
TN-6. The most important targets are:
1. The historical leak frequency on NCS in 2006 – 2017 is regarded as a reasonable
estimate for future leak frequencies (see TN-2). Hence, the model should be able to
reproduce the total number of leaks observed for all installations at NCS being in
operation in the period 2006-2017
2. The model should be able to reproduce the total cumulative leak rate frequency
distribution seen in historical data from NCS in the period 2001 - 2017 when applied to
all installations on NCS (see TN-2)
3. The model should be able to reproduce the relative leak rate frequency distribution per
equipment type seen in the experienced data from NCS (and UKCS). Stochastic effects
are expected to be prominent in this regard as the number of incidents will be few for
some equipment types
Some types of equipment have been subjected to special evaluations where the methodology for
parameterization deviates slightly from the general methodology described above. This is either
due to lack of data in either of the datasets or uncertainties/shortcomings in the available data.
The types of equipment subjected to special evaluations and the reason for the alternative
approach is summarized in Table 6.1.
The concluded model parameters are given in Chapter 7.

Report no: 107566/R1 Rev: Final Page 13

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 6.1 – Total frequency, rupture fraction (full bore hole fraction) and rupture frequency for
valve

Table 6.1 - Equipment types subjected to separate parameterisation process

Equipment Description
type

Hoses No data available in HCRD on leaks occurring under temporary operations

involving use of hoses. The defined model parameters are therefore solely
based on data gathered from installations on the NCS

Steel pipe The quality of the population data in HCRD for steel pipes is judged to be
poor. Hence, the model is parameterized based on a subset of the NCS
population dataset where equipment counts of length steel pipe are
available. However, available data in HCRD have been used to model the
effect of equipment size on the hole size distribution for steel piping

Compact No data is available from UKCS and only limited data available from
flanges installations on NCS. A separate assessment is performed to set the model
parameters

Air-cooled heat No units registered at NCS. UKCS data applied to set parameters.
exchanger

Flexible piping No population data available at NCS. UKCS data applied to set parameters.

Atmospheric No data available at NCS nor UKCS. Recommended to use process vessel
vessel model presuming that the vessel is slightly over-pressurized.

Report no: 107566/R1 Rev: Final Page 14

Date: 6 December 2018 ©Lloyd’s Register 2018
6.2 Overall validation
The validation model has demonstrated that PLOFAM is able to:
• Reproduce the total number of significant leaks (104) observed at NCS in the period 2006 –
2017 when applied to all installations on NCS being in operation in this period. Details are
presented in TN-6
• Reproduce the total cumulative leak rate frequency distribution seen in historical data from
NCS in the period 2001 - 2017 when applied to all installations on NCS. See Figure 6.2 ,
which shows good agreement between the resulting relative leak rate distribution from
applying PLOFAM to all installations in operation in the period 2006 – 2017 and historical
leak rate distributions recorded both at UKCS in the period 1992 – 2017 and NCS in the
period 2001 – 2017
• Reproduce the total cumulative leak rate frequency distribution for every equipment type
seen in historical data from NCS in the period 2001 - 2017 when applied to all installations
on NCS. TN-6 shows the resulting cumulative leak rate frequency distribution for every
component when applying the model to all installations being in operation in the period
2006 – 2017, and compares it with the historical data both from both NCS and UKCS. The
results shows good fit with historical data
• Generate the split on the number of leaks being significant and marginal, when applied to all
installations on NCS being in operation in the period 2006 - 2017. These results are
presented in TN-6
• Reproduce the relative distribution between equipment types seen based on experienced
data from NCS and/or UKCS. See further details below
The average number of leaks per installation is given in Figure 6.3. The average number of leaks
>0.1 kg/s predicted by the model is 0.1 leaks per installation per year. The figure shows the
average cumulative number of leaks per installation (i.e. the average number of leaks larger than
the value on the x-axis) for the main equipment types. Other equipment types are grouped
together. The historical distribution is also given both for the period 2001 – 2017 and for the
period 2006 – 2017. As the average number of leaks in the period 2001 – 2017 is higher than
for the period 2006 – 2017, the historical number of leak in 2001 – 2017 has been adjusted to
give the same total number of leaks as predicted in the period 2006 – 2017. It is however
included to demonstrate that the model is able to reproduce the model target defined by the
relative leak rate distribution for the full period.
To assess how the model performs for each installation in operation in the period 2006 – 2017,
Figure 6.4 shows the number of estimated leak versus the number of historical leaks for each
installation (each data point represents one specific installation). The number of data points
above (overestimated) and below (underestimated) the grey line is about the same, indicating
that the overall prediction of the model is good, even though the model is obviously not able to
predict the correct number of leaks for every installation. As discussed in TN-5 and TN-6, this is
not expected as there are many factors influencing on the leak frequency that are not
implemented in the model. However, there is a clear correlation in the data points indicating that
the model is able to capture important factors influencing the leak frequency.
It must be noted that this plot is prone to stochastic effects. The model will predict leaks on
installations where no leak has occurred yet. Since the model is targeting the total number of
observed leaks for all installations, the model will lead to an average underprediction of the
number of leaks on all installations with 1 or more leaks. This stochastic effect will diminish with
time as the number of installations with zero leaks becomes fewer and fewer. At some point in
time it is expected that the scatter plot for all installations will follow an average linear trend with
slope 1:1.
The model is mainly validated towards available data of leaks that has occurred at installations on
the Norwegian Continental Shelf (NCS), but the data from the United Kingdom Continental Shelf
(UKCS) indicate that the underlying hole size frequency distribution for equipment at installations
located on the NCS is similar to the distribution for equipment located on UK installations. The
differences may be explained by uncertainty related to the datasets (both the leaks and the
population data, and the way equipment are counted and leaks are tagged to equipment types).

Report no: 107566/R1 Rev: Final Page 15

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 6.5 display the number of leaks (significant + marginal leaks) per equipment year (all types
of pipes excluded) per year for NCS and UKCS. The plot show that the leak frequency per
equipment year and time trend in the leak frequency at UKCS is similar to the time trend seen on
NCS. The average frequency appears to be slightly less at UKCS (about 20%), but that may for
instance be due to uncertainty in the UKCS population data. The result presented in Figure 6.2
demonstrates that the relative distribution with regards to initial leak rate is the equivalent for the
two data sets.
No causal arguments have been found that supports a difference in the underlying leak
frequency between NCS and UKCS installations. This does not mean that such a difference does
not exist, only that the PLOFAM project has not identified any justification for such a difference.
The same conclusion is established in the MISOF project in terms of probabilistic modelling of
ignition sources. A hypothesis claiming that the underlying leak frequency and ignition probability
is the same for the two domains cannot be rejected based on the available data. Both the
PLOFAM model and MISOF model is therefore concluded to be valid for both sectors.
Figure 6.6 display the resulting overall distribution of significant leaks per equipment for all
installations operating at NCS and UKCS compared with the PLOFAM prediction. The results
show that PLOFAM reproduce the distribution observed at NCS. A significant difference with
regards to the data from UKCS installations appear. This may be due to randomness, in particular
for equipment with a low number of equipment years, but is also clearly due to differences in the
way leaks are logged in HCRD relative to the NCS dataset. In some situations, it is not straight
forward to allocate a leak to a specific equipment, leading to uncertainty related to the tagged
equipment. However, for NCS leaks extra quality assurance has been performed together with
Equinor, Safetec and ConocoPhillips to make sure that the leaks are tagged to the right
equipment. A few leaks are tagged to a different equipment than in the previous version of
PLOFAM. Typically this was related to small leaks from instrument tubing that previously was
tagged to steel pipe that are now tagged to instruments.

Figure 6.2 – Overall significant leaks: complementary cumulative relative leak rate distribution for
all installations operating at NCS in the period 2006-2017

Report no: 107566/R1 Rev: Final Page 16

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 6.3 – The average number of leaks per installation for the main equipment types predicted
by the model for the period 2006 – 2017. Other equipment types than standard flange, valve,
instrument and steel pipe are grouped into “Other”. The historical number of leak in 2001 – 2017
has been adjusted to give the same total number of leaks as predicted in the period 2006 – 2017. It
is however included to demonstrate that the model is able to reproduce the model target defined
by the relative leak rate distribution for the full period. The total number of installation years (for
installations in operation) for the NCS population dataset is 899 for the period 2006-2017 and 1237
for the period 2001 – 2017

Figure 6.4 – Overall significant leaks: observed number of leaks vs. predicted number of leaks for
all installations operating at NCS in the period 2006-2017. Datapoints below the grey line indicate
underprediction. Datapoints above the grey line indicate overprediction

Report no: 107566/R1 Rev: Final Page 17

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 6.5 – Annual frequency for leaks ≥ 0.1 kg/s per equipment (includes all types of equipment
except steel pipe) both for UKCS and NCS. For NCS the columns giving the leak frequency after
2001 are filled to indicate that that there is a shift in the uncertainty related to the data. Note
however that the uncertainty related to the overall frequency presented in the figure is regarded
low also before 2001. No shift in data quality is known for UKCS data. The correct exponent
belonging to the figures in the table must be read from the second axis (the font size is maximized
to enhance readability of the figures)

Report no: 107566/R1 Rev: Final Page 18

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 6.6 – Historical (both for NCS and UKCS) and modelled equipment type distribution for Significant leaks. Both the historical equipment type
distribution for the period 2001 – 2017 and 2006 – 2017 are given. The modelled result is achieved by applying PLOFAM to all installations at NCS being
in operation in the period 2006 – 2017

Report no: 107566/R1 Rev: Final Page 19

Date: 6 December 2018 ©Lloyd’s Register 2018
7 PLOFAM parameters
A list of PLOFAM parameter values for 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 , 𝐴𝐴0 , 𝑀𝑀0 , 𝐴𝐴𝐷𝐷 , 𝑀𝑀𝐷𝐷 , 𝐵𝐵𝐷𝐷 and α necessary to estimate leak frequencies for all equipment types are given in
Table 7.1. 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 is given both for Significant and Marginal leaks. The parameters are documented in TN-6. Note that for the equipment types marked
with a star (*), the data basis is scarce and the model parameters are related with higher uncertainty than for the remaining equipment types (see
TN-6).

Table 7.1 – PLOFAM model parameters. 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖 is given both for Significant and Marginal leaks.
Equipment type 𝐴𝐴0 𝑀𝑀0 𝐴𝐴𝐷𝐷 𝑀𝑀𝐷𝐷 𝐵𝐵𝐷𝐷 𝛼𝛼 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖,Significant 𝐹𝐹ℎ𝑖𝑖𝑖𝑖𝑖𝑖,Marginal
Air-cooled heat exchanger* 1 0 0 0 3.0E-02 0 5.0E-04 0
Atmospheric vessel* 1 0 0 0 1.0E-01 0 5.0E-04 0
Centrifugal compressor 1 0 0 0 6.0E-03 0 1.3E-03 0
Centrifugal pump 1 0 0 0 3.0E-05 0 3.0E-03 0
Compact flange 1 0 0 0 1.0E-03 0.90 3.0E-06 0
Filter 1 0 0 0 8.0E-04 0 2.3E-03 0
Flexible pipe* 1 0 0 0 4.0E-01 0.75 1.4E-04 0
Gas lift well 1 0 0 0 2.5E-02 0 1.0E-04 1.0E-04
Hose 1 0 0 0 4.0E-01 0.75 6.0E-05 1.5E-05
Instrument 1 0 0 0 1.5E-01 0 1.3E-04 0
Pig trap 1 0 0 0 2.0E-02 0 1.7E-03 0
Plate heat exchanger 1 0 0 0 1.0E-03 0 3.5E-04 0
Process vessel 1 0 0 0 6.0E-04 0 5.0E-04 0
Producing well 1 0 0 0 2.0E-02 0 2.0E-05 1.3E-04
Reciprocating compressor 1 0 0 0 1.0E-02 0 5.0E-03 -
Reciprocating pump 1 0 0 0 3.0E-05 0 3.0E-03 -
Shell and tube heat exchanger 1 0 0 0 7.5E-03 0 3.3E-04 -
Standard flange 1 0 18.0 -1.45 5.0E-03 0.50 2.5E-05 5.0E-06
Steel pipe 4.20 -0.3 17.6 -1.75 1.0E-03 0.90 1.4E-05 2.0E-06
Valve 1.11 -0.1 16.0 -1.70 1.0E-03 0.50 2.15E-04 3.5E-05

Report no: 107566/R1 Rev: Final Page 20

Date: 6 December 2018 ©Lloyd’s Register 2018
8 Comparison of PLOFAM and SHLFM
Frequency estimated using PLOFAM, is compared with the current frequency model commonly
used in the industry in Norway, denoted SHLFM (Ref. /4/).
The resulting frequency distributions obtained when applying PLOFAM and SHLFM on the NCS
population data set for the period 2006 – 2017 are shown in Figure 8.1. Figure 8.2 shows the
relative number of leaks estimated using PLOFAM relative to using SHLFM. The results show that
the difference between the two models is considerable. The difference is following from:
• PLOFAM is based on historical data of leaks occurring at installations on the NCS in the
period 2001-2017. The last version of SHLFM was solely based on data of leaks occurring at
installations on the UKCS in the period 1992-2010. There has been a considerable decrease
in historical leak frequency at installations on the UKCS over this period. It has been shown
that the underlying hole size frequency distribution on the UKCS and NCS after year 2000 is
similar, and most likely the same statement is valid for the total leak frequency as well
• Enhanced understanding of the quality of the data in the HCR database, which has provided
basis for implementing considerably less conservatism to account for uncertainty related to
the data basis. This has been made possible by the fact by that more data related to the leaks
in HCRD has been made available by HSE. Two items have been particularly important in this
regard. Firstly, the actual hole size for incidents where the hole size was larger than 100 mm
has been provided. Previously, it was stated that the hole size was > 100 mm in such cases.
The additional information on large hole sizes has provided confidence in estimation of a
more accurate frequency for large leaks. Secondly, it has been found that the population
data (i.e. number of equipment years) in HCRD has not been updated after 2005, which
means that the estimated frequency for leaks extracted from HCRD will lead to an excessive
estimate of the leak frequency even for installations located on the UKCS
• The mathematical formulation in PLOFAM enables an improved representation of the effect
of the equipment size on the hole size frequency distribution for the various equipment
types. In SHLFM, the capability in terms of capturing the shift in hole size distribution with
varying equipment size for a given equipment was less pronounced. In combination with
parameterisation of SHLFM outside the range of HCRD data for large holes led to estimation
of excessive frequency for large holes for all equipment sizes for all equipment type.
Moreover, as additional data on equipment size per incident in HCRD has been made
available to the project, and more insight has been gained related to failure modes and their
influence on different equipment sizes, it has been possible to develop and parameterize the
equipment size dependent model in PLOFAM
• PLOFAM is unbiased and parametrized to reproduce the number of historical leaks at NCS in
the period 2006 - 2017. Bias of frequency towards large hole sizes were included in SHLFM
to account for uncertainty
The resulting quantitative fire and explosion risk picture in a QRA for a typical installation on the
NCS will be very different based on PLOFAM opposed to SHLFM. The validation model has
demonstrated that the model denoted PLOFAM is able to predict the observed number of leaks
at installations located on the NCS in the period 2006-2017, whilst SHFLM will overpredict the
observed number of leaks in the same period. Hence, SHFLM is not recommended for prediction
of the frequency for leaks on oil and gas installations at NCS.
The total quantitative fire and explosion risk picture is determined by combining PLOFAM with
the MISOF ignition model. In the MISOF report, a more thorough comparison of PLOFAM and
SHLFM including the effect of modelling of ignition probabilities is included (Ref. /2/). The effect
on the fire and explosion frequency is prominent.

Report no: 107566/R1 Rev: Final Page 21

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 8.1 – Comparison of SHLFM and PLOFAM. The frequency is the sum of frequency for gas and
liquid leaks. For the SHLFM model, only Full pressure leaks and Limited leaks are included. In
PLOFAM, only significant leaks are included. Note that the ratio for leaks >1000 kg/s is infinite but
plotted as 1700% to illustrate in the figure that the value is high

Figure 8.2 – Comparison of SHLFM and PLOFAM. The bars display the ratio per leak category for
the total of gas and liquid leaks. For the SHLFM, only Full pressure leaks, i.e. Full leaks and Limited
leaks are included. In PLOFAM, only significant leaks are included

Report no: 107566/R1 Rev: Final Page 22

Date: 6 December 2018 ©Lloyd’s Register 2018
9 Robustness of PLOFAM
The quality and limitations of the data used as basis for the parameterisation of PLOFAM is
fundamental for the precision of the model. The quality of the data basis is discussed throughout
the report, both in the technical notes presenting the data basis (TN-2 and TN-3), but also in
discussion of the results from applying the model to all installations at NCS (see TN-6).
It is judged that the elements affecting the quality and limitations of the data are understood, but
some of them may be hard to quantify. On a high level, the frequency distributions based on
data extracted from the HCR database and the NCS database is similar. This is considered to be a
strong argument for that the PLOFAM model is based on a solid understanding of the data basis.
This can for instance be seen from Figure 6.2, where it is shown that the relative complementary
cumulative leak frequency distribution for both NCS and UKCS coincide with the PLOFAM
estimate.
The quality of the NCS leak database regarded as high because it is established based on review
of the accident investigation reports for the incidents.
The overall quality of the NCS population data is discussed in TN-2 and also as part of the
guidelines given in TN-5 Appendix B. The data are gathered from QRAs performed by 4 different
consultancy companies. In general, the variation between the different consultancy companies is
small, which indicates that the industry practice on counting of equipment is quite homogenous.
This is interpreted as an argument why the quality of the population data is regarded as good,
even though uncertainties do exists. The uncertainties are in general related to how equipment is
counted. By introducing guidelines for equipment counting the aim is to achieve an even more
unified way of counting equipment (see TN-5). However, as different strategies may have been
used as basis for the population database one may introduce a bias toward underpredicting or
overpredicting the leak frequencies by applying the guidelines. In such cases the guidelines have
in general been formed to rather overpredict than underpredict the leak frequency in future
QRAs. Note however that the expected overprediction is expected to be low and well within the
uncertainty related to other aspects. It should however be noted that uncertainties related to
how pumps are counted relative to how it is recommended counted in the guidelines may
introduce a uncertainty related to the fire frequency up to 40% for modules with pumps (see TN-
5 Appendix B). Another main uncertainty in the population database is related to the number of
wells at the installations. Leaks from wells are however not expected constitute a significant part
of leak frequency for most installations.
PLOFAM is designed to be a tool for estimation of future leak frequencies for use in QRAs. Hence
an important aspect with respect to robustness is the model target relative to the historical
observations. The model target, time trend and stochastic effects are discussed in detail in TN-2
and TN-6. In short PLOFAM is designed to predict the same total leak frequency as seen at NCS in
the period 2006 – 2017. Targeting this frequency level would imply that the model will estimate
about 30% lower leak frequency than the average leak frequency in the period 2001 – 2017, but
also 30% higher leak frequency than seen for any years after 2011, and about 50% higher leak
frequency than recorded in 2017. If the observed trend in time for leaks >0.1 kg/s continues,
PLOFAM will overestimate the total leak frequency even more, i.e. the chosen target level for the
model account for uncertainty in the data material and shifts in underlying causal factors (e.g.
emerging unknown degradation mechanisms due to age or changing operational conditions)
affecting the future trend in leaks occurring on installations on the NCS.
The targeted leak rate distribution is important for estimating the leak frequency for large leaks.
As the number of large leaks is few, the stochastic uncertainty of the relative leak rate
distribution is significant. Data periods do exist where both higher and lower fractions of large
leaks are seen. Considering only data after 2007 will give lower fraction of large leaks if used as
basis for the model, whilst considering only data from the period 2006 – 2017 gives a higher
fraction of large leaks. It should also be noted that data for the period 1992 – 2000 shows a
significant lower fraction of large leaks. The difference is large and cannot be explained solely by
stochastic variations. This is first of all a result of the clear time trend seen for leaks >0.1 kg/s, i.e.
that the leak frequency is significantly reduced. For leaks > 10 kg/s, it is also likely that there is a
time trend, i.e. that the leak frequency is being reduced, but it cannot be ruled out that the

Report no: 107566/R1 Rev: Final Page 23

Date: 6 December 2018 ©Lloyd’s Register 2018
 observed variations is due to stochastic variations (see TN-2). Hence the increased fraction of
large leaks is mainly a consequence of the reduced leak frequency for small leaks (this should be
carefully taken into account in future updates of the model).
The robustness of the target value is also discussed in TN-6, where the stochastic uncertainty is
shown to increase with increasing leak rate. Due to randomness, one cannot be sure about the
targeted leak rate distribution. The parameterisation process in PLOFAM targets the most likely
underlying distribution, but both significantly lower or higher target frequency cannot be rule out
(illustrated by the fortunate and unfortunate scenarios corresponding to 10% and 90%
probability of exceedance, i.e. 80% confidence interval). See TN-6 for further explanations of the
fortunate and unfortunate scenarios established based on the Poisson process. This aleatory
uncertainty should be taken into account in the decision-making process. The ratio between the
fortunate and unfortunate scenario relative to the observed/targeted number of leaks are given in
Figure 9.1 and summarizes the discussion of the stochastic uncertainty for the defined model
target. The increasing spread of relative distributions with respect to initial leak rate for the
unfortunate and fortunate scenario displayed in Figure 9.1 is important to consider when
evaluating the accuracy of a QRA model based on PLOFAM. For leak rates above about 30 kg/s,
the relative difference between the fortunate and unfortunate scenario constitutes a factor in the
range 1.5 to 2.5.
In total, the combination of the targeted total leak frequency and the targeted fraction of large
leaks will decide the targeted leak frequency for large leaks. The resulting target is regarded
reasonable and a best estimate, slightly approached from the conservative side.
Note also that the guidelines for use of PLOFAM in QRAs presented in TN-5 Appendix B are
established from the conservative side. First of all it is recommended to model all Significant leaks
as leaks occurring during normal operation where the full inventory is released (taking ESD and
blowdown into account). Secondly the counting rules takes uncertainty in the population
database used for model parametrization into account, by rather overestimate the leak frequency
than underestimating it. This adds robustness to the frequency and risk estimates generated in
QRAs using PLOFAM.
Although a strict update of the PLOFAM model parameters in case of the occurrence of one
future large leak at NCS would lead to a model that would predict a slightly higher frequency for
leaks, the current model cannot be disregarded if one (or two) large leak occurs in the near
future as the target value used for PLOFAM still would have a relatively high probability of
occurrence. Figure 9.2 shows the effect on the targeted relative leak rate distribution if two leaks
>100 kg/s where one of them is larger than 300kg/s potentially occurring tomorrow (for
comparison there has been three leaks >100 kg/s and one leak >300 kg/s in the period 2001 –
2017). For leaks <10 kg/s the fraction is relatively unchanged, while for leaks >10 kg/s, the
fraction is increased by up to a factor of 2. Comparing the updated model target (red curve in
Figure 9.2) with Figure 9.1 shows that the updated model target will still be within the limits
defined by the fortunate and unfortunate scenarios in Figure 9.1. Likewise, a long period (> 5
years) without observing any large leak in the future does not imply that the current model is
conservative in terms of estimation of the frequency for large leaks.

Report no: 107566/R1 Rev: Final Page 24

Date: 6 December 2018 ©Lloyd’s Register 2018
 Figure 9.1 – Ratio unfortunate and fortunate scenario (10% percentiles used) with respect to
observed data (NCS 20017-2017) versus initial leak rate. These ratios are derived directly from the
relative leak rate distributions shown in TN-6

Figure 9.2 – Relative leak rate distributions for the periods 2001 – 2017 (target for model), 2006 –
2017, 2007 – 2017, and 2001 – 2017 including two leaks >100 kg/s where one leak is larger than
300kg/s potentially occurring tomorrow

Report no: 107566/R1 Rev: Final Page 25

Date: 6 December 2018 ©Lloyd’s Register 2018
10 Summary and concluding remarks
The objective has been to build a leak frequency model that will serve as a tool for prediction of
the future leak frequency for topside process leaks at installations located on the Norwegian
Continental Shelf (NCS) for use in QRAs. The model should be unbiased, i.e. it should aim at a
best estimate.
Overall, the model is built on a combination of the explanatory variable that shows the strongest
correlations with experienced number of leaks, and rational explanations and causalities
reflecting known failure modes. The number of equipment (for each equipment type) is
concluded to be the best single predictor to build the model on. However, as only one
explanatory variable is chosen for the model, there are many factors influencing on the leak
frequency that are not captured by the model, which will give rise to stochastic effects and
deviations from the average for single installations.
It should be emphasised that PLOFAM is designed to serve as a tool for prediction of the future
leak frequency for topside process leaks at installations located on the Norwegian Continental
Shelf (NCS) for use in QRAs. Even if failure modes are aimed to be reflected as well as possible in
the model, the model should first and foremost be used for prediction of leak frequencies for
complete process systems. The model is to a much less degree valid for single components as a
range of different designs (for instance standard flange constitute flanges with different sealing
design) have formed the basis for the model. This should be addressed in further work (see TN-6),
i.e. improving the capability of PLOFAM to reflect the specific technical characteristics and failure
modes of the various types of components (i.e. ASME ring joint flange vs. Grayloc clamped
connection).
The model is mainly parametrized and validated towards NCS data. The quality of the data is
regarded as high. However, data from UKCS seem to show similar hole size frequency
distributions, time trends and total leak frequency. The model is therefore regarded as valid for
both sectors.
The strategy has been to build a model that gives a best estimate for future leak frequencies, i.e.
to create an unbiased model without built in conservatism. It is observed a significant decreasing
trend in historical leak frequency with time for installations on the NCS in the period after year
2000 (actually since 1992). The number of historical leaks in the period 2006-2017 is used as
target for the total leak frequency while leak data from the period 2001 – 2017 is used as target
for the relative leak rate distribution. Targeting this frequency level would imply that the model
will estimate about 30% lower leak frequency than the average leak frequency in the period
2001 – 2017, but also 30% higher leak frequency than seen for any years after 2011, i.e. the
chosen target level for the model account for uncertainty in the data material and shifts in
underlying causal factors (e.g. emerging unknown degradation mechanisms due to age or
changing operational conditions) affecting the future trend in leaks occurring on installations on
the NCS. In total the combination of the targeted total leak frequency and the fraction of large
leaks will decide the targeted leak frequency for large leaks, and is regarded reasonable and as a
best estimate, slightly approached from the conservative side. Note also that conservatism is
embedded in the guideline for use of PLOFAM in QRAs given in TN-5.
Stochastic uncertainty related to the targeted relative leak rate distribution is discussed in TN-2,
TN-6 and in the previous chapter. The stochastic uncertainty is larger for large leak rates than for
small leak rates. This is important to consider when evaluating the accuracy of a QRA model
based on PLOFAM. For leak rates above about 30 kg/s, the relative stochastic uncertainty
constitutes a factor in the range 1.5 to 2.5. i.e. based on the historical data it can be argued that
the target value used for parametrization of the model can be both a factor 1.5 – 2.5 higher and
lower than the target values used in PLOFAM (PLOFAM targets the most likely value). As a
consequence it is shown that if two leaks >100 kg/s where one of them is larger than 300kg/s
occur tomorrow, the model will still be valid.

Report no: 107566/R1 Rev: Final Page 26

Date: 6 December 2018 ©Lloyd’s Register 2018
 The models ability to estimate the target value is discussed in TN-6, and also in Chapter 6.2
above. It is concluded that the model predicts the targeted values well when applied to the all
installations in operation on NCS in the period 2006 – 2017, which is in accordance with the
requirement to the model. Hence the aimed robustness established when establishing the target
values is concluded to be implemented in the model.
The model will show stochastic variations for single installations (see also discussion in
Chapter 6.2). As described in Chapter 2 there will be many factors influencing on the leak
frequency (where some are installation specific) that are not captured by the model, which will
give rise to stochastic effects. The historical leak frequency per installation at the NCS can
therefore vary significantly from the NCS average, as illustrated in Chapter 6.2.
PLOFAM has been compared with the commonly used leak frequency model denoted SHLFM
(“Standardised Hydrocarbon Leak Frequencies Model”, Ref. /4/). The difference between the leak
frequencies generated by the two models is considerable. These differences are explained by
effects following from properties of the new data material being available for development of
PLOFAM as well as new features of the mathematical framework enabling improved
representation of the equipment size for the various equipment types. The resulting quantitative
fire and explosion risk picture in a QRA for a typical installation on the NCS will be significantly
different based on PLOFAM opposed to SHLFM. When applied to all installations being in
operations in the period 2006 – 2017, PLOFAM is able to predict the observed number of leaks in
this period, whilst SHLFM will overpredict the observed number of leaks for the same period
approximately by a factor 3 for small leaks and a factor 10 for large leaks. Hence, SHLFM is not
recommended for prediction of the frequency for leaks on oil and gas installations at NCS.
Although the model is based on releases of hydrocarbons from process equipment on North Sea
offshore facilities, it is found reasonable to argue that the model is applicable to platforms and
land based facilities in other domains. This should be based on a specific assessment to qualify
use of the model in the particular domain. The important element to evaluate is whether the
properties of the equipment and operation conditions can be considered equivalent with what
are found generally on installations located in the North Sea.
The data basis for PLOFAM and the ignition model MISOF are fully aligned, and should therefore
be used together. An important argument is that the historical data put as basis for the models
are implemented based on the same assessment and understanding of the leak scenarios. One of
the main uncertainties related to PLOFAM is the fraction of large leaks. Note however that if the
number of large leaks is misinterpreted in the data material and should have been higher (i.e.
that the leak frequency for large leaks should have been higher) then the ignition probability
would have been lower in MISOF. This demonstrated why the two models should be used
together and not combined with other models.
For further development of the model and future updates, a list of suggested focus areas has
been listed in TN-6.

Report no: 107566/R1 Rev: Final Page 27

Date: 6 December 2018 ©Lloyd’s Register 2018
11 References

/1/ Lloyd’s Register Consulting, “Process leak for offshore installations frequency assessment
model – PLOFAM”, report no: 105586/R1, Rev: Final B, Date: 18.03.2016

/2/ Lloyd’s Register Consulting, “Modelling of ignition sources on offshore oil and gas facilities
- MISOF”, Date: November 2018, Report No: 107566/R2, Rev: Final

/3/ Lloyd’s Register Consulting, “Blowout and well release frequencies based on SINTEF
offshore blowout database 2017”, 20 April 2018, Report No: 19101001-8/2018/R3 Rev:
Final

/4/ DNV, Offshore QRA – Standardised Hydrocarbon Leak Frequencies, report number 2009-
1768, rev. 1, 16.01.2009.

Report no: 107566/R1 Rev: Final Page 28

Date: 6 December 2018 ©Lloyd’s Register 2018