Risk ID & Assessment Training Course DNV Consulting

Module 17: Frequency analysis using historical data

MODULE 17

FREQUENCY ANALYSIS USING HISTORICAL DATA

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page i

CONTENTS

17. FREQUENCY ANALYSIS USING HISTORICAL DATA ..........................................1

17.1 DEFINITIONS.......................................................................................................1
17.1.1 Frequency Analysis and Reliability Analysis.........................................................1
17.1.2 Frequencies and Probabilities ..............................................................................1
17.1.3 Presentation of Frequencies.................................................................................1

17.2 APPROACHES TO FREQUENCY ANALYSIS ....................................................3

17.3 FREQUENCY ESTIMATION USING HISTORICAL DATA ...............................4

17.4 METHOD ..............................................................................................................5

17.4.1 Step 1 - Define Context.......................................................................................5
17.4.2 Step 2 - Review Source Data ..............................................................................5
17.4.3 Step 3 - Check Data Applicability........................................................................6
17.4.4 Step 4 - Calculate Event Likelihood ....................................................................6
17.4.5 Step 5 - Validate Frequency ................................................................................6

17.5 EXAMPLE 1: PRESSURE VESSELS ...................................................................7

17.5.1 Define Context....................................................................................................7
17.5.2 Review Source Data ...........................................................................................7
17.5.3 Check Data Applicability.....................................................................................7
17.5.4 Calculate Event Likelihood .................................................................................7

17.6 EXAMPLE 2: PIPELINE FAILURE FREQUENCY ANALYSIS ........................10

17.6.1 Define Context..................................................................................................10
17.6.2 Review Source Data .........................................................................................10
17.6.3 Check Data Applicability...................................................................................10
17.6.4 Calculate Event Likelihood ...............................................................................11
17.6.5 Validate Frequency ...........................................................................................13

17.7 MEASURES OF EXPOSURE..............................................................................15

17.7.1 Frequency per Calendar Year ............................................................................15
17.7.2 Frequency per Installation Year.........................................................................15
17.7.3 Frequency per Equipment Item..........................................................................15

17.8 SOME HISTORICAL INCIDENT DATA SOURCES .........................................16

17.9 FAILURE CASE FREQUENCIES .......................................................................19

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 1

17. FREQUENCY ANALYSIS USING HISTORICAL DATA

17.1 DEFINITIONS

17.1.1 Frequency Analysis and Reliability Analysis

Frequency analysis involves estimating the likelihood of each of the failure cases which were defined in the
hazard identification stage. Typical requirements are frequencies of pipe leaks, collisions, structural failures,
etc. Approaches to this are described in this module.

Reliability analysis involves identifying failure modes and calculating the likelihood of failures of systems
which consist of numerous components. Failure frequencies for these are often used as inputs to a frequency
analysis. Reliability analysis includes techniques (such as fault tree analysis and event tree analysis) which
are widely used to present frequency calculations in a QRA.

17.1.2 Frequencies and Probabilities

There are two basic forms in which the likelihood of an event may be expressed:

• Frequency - the expected number of occurrences of the event per unit time, usually a year, but sometimes
an hour or a project lifetime. In reliability analysis, frequencies of failures are known as “failure rates”.

Probability - the probability of the event occurring in a given time period or the conditional probability of it
occurring given that a previous event has occurred.

The essential difference between the two forms is that probabilities cannot exceed 1.0, whereas frequencies
can be larger (e.g. where several events occur per year). In risk analysis, the frequency is often very small,
and the difference between a frequency per year and a probability per year may be negligible. However, in
reliability analysis (such as fault tree analysis) the conceptual difference is important, and a clear distinction
should be maintained.

The units of a frequency or probability should always be stated explicitly:

A frequency of 10-4 per year

A probability of 0.5 given a release of gas

In a risk assessment, where a frequency is given without the units being stated, it usually refers to a frequency
per year.

17.1.3 Presentation of Frequencies

In this module, frequency and risk numbers less than 0.01 are usually presented in scientific notation, e.g. 4.2
× 10-3. The equivalence of scientific notation to other forms is as follows, and is explained in Table Error!
No text of specified style in document..1:

4.2 × 10-3 per year

0.0042 per year
1 in 238 per year

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 2

Table Error! No text of specified style in document..1 Equivalence of Scientific Notation

Scientific Decimal Chance

10-0 1 1 in 1
10-1 0.1 1 in 10
10-2 0.01 1 in 100
10-3 0.001 1 in 1000
10-4 0.0001 1 in 10,000
10-5 0.00001 1 in 100,000
10-6 0.000001 1 in 1 million
10-7 0.0000001 1 in 10 million
10-8 0.00000001 1 in 100 million

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 3

17.2 APPROACHES TO FREQUENCY ANALYSIS

The main approaches to estimating accident frequencies are:

• Historical accident frequency data. This uses previous experience of major accidents. It is a simple
approach, relatively easy to understand, but is only applicable to existing technology with significant
experience of accidents. It is described in more detail in the following sections.

• Fault tree analysis. This involves breaking down an accident into its component causes, including human
error, and estimating the frequency of each component from a combination of generic historical data and
informed judgement.

• Theoretical modelling. The frequencies of some types of accidents can be predicted using theoretical
models of the accident situation. An example of this is ship collisions, where the ship movements can be
represented by a theoretical model and the frequency of collisions determined by simulation or analytical
solution.

• Event tree analysis. This is a means of showing the way an accident may develop from an initiating
event through several branches to one of several possible outcomes. The technique is usually used to
extend the initiating event frequency estimated by one of the above means into a failure case frequency
suitable for combining with the consequence models.

Judgmental evaluation. In some cases, it may be appropriate to select a frequency based on judgement of
experienced personnel. This may be for simple assessments, for frequent events, for events considered to
have minimal risk, or for events where no better approach is available.

In general, these techniques are used in combination.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 4

17.3 FREQUENCY ESTIMATION USING HISTORICAL DATA

Frequencies of failure events used in quantitative risk analysis are usually derived from relevant
experience based data, if this exists. Failure databases compiled from component failures provide
information on specific components and groups of components. If failure frequency data does not exist
one can use fault tree analysis to derive the appropriate failure frequency (see fault tree analysis
module).

The routes to estimating incident frequency are summarised in Figure Error! No text of specified style
in document..1.

Figure Error! No text of specified style in document..1 Frequency Estimation Methods

Common Cause Analysis

Population Historical Incident Fault Tree Analysis
Human Reliability Analysis
Data Records Event Tree Analysis
External Event Analysis

Likelihood
(frequency or probability)

The advantage of using data sources rather than fault tree and event tree analysis are:

Credibility The use of real data from previous incidents avoids the need for further justification of
the likelihood, provided appropriate and accurate data is used

Speed Fault tree and event tree analysis are time consuming techniques requiring considerable
expertise.

Experience and knowledge of failure causes, reporting systems, statistical methods and intended end use
is needed to develop and compile valid and useful databases. Of course, the cooperation of industry in
providing the failure data in useable format together with the necessary information on component
population and conditions of use, is equally important.

When using historical record data for frequency estimation you should be aware that:

• Rare incidents may not have occurred.

• Sufficient accurate and applicable data must be available.
• Newer designs reflect incident knowledge and thus may not be directly applicable.

Historical record data is particularly applicable in the following situations:

• Early in the design stage i.e. before plant systems have been defined (coarse study).
• For diverse or difficult to predict failure causes e.g. maritime transport.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 5

17.4 METHOD

The process can be described in 5 steps as follows:

Step 1 - Define Context

Step 2 - Review Source Data

Step 3 - Check Data Applicability

Step 4 - Calculate Event Likelihood

Step 5 - Validate Frequency

17.4.1 Step 1 - Define Context

The objective of the study and the design and operating conditions must be clearly understood. It may
also be useful to state the type of study planned (coarse or detailed QRA) and the degree of accuracy
required. For example:

Objective:

As part of a coarse QRA to determine the leakage frequency of a proposed pipeline.

Design and Operating Conditions:

Pipe diameter, pipe length, type of fluid transported, materials of construction, route, corrosion
protection, impact protection, design and construction codes and practices.

17.4.2 Step 2 - Review Source Data

All relevant historical data sources should be consulted.

A number of databases are available for the process industry, onshore and offshore. A list of databases
is provided at the end of this module in Section 17.8. In addition operating companies and consultants
may have their own databases

These data sources contain failure rates for a variety of process equipment used onshore and offshore.
In most cases, the failure rates for an offshore installation are kept separate since the equipment often
has different failure rates due to different environment and working conditions or because they only
exist offshore. However, some of these data sources present data based on failure modes enabling the
appropriate failure frequency to be evaluated for the particular environment and failure mode.

Some of these data sources give the failure rate for equipment without much explanation on where they
are taken from, how they are evaluated and what working environment existed at the time of failure or
during the life of operation of that equipment. Others, however, can provide a detailed description of
each failure. These are the sources that should be used since they may provide the most exact
information available and since they may provide the greatest flexibility of failure rate analysis.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 6

Wherever possible, data sources providing both population and exposure times should be used. If it is
necessary to estimate exposure times (plant years etc.) for some of this data then care is needed to avoid
significant errors which can occur when using incorrect exposure data.

17.4.3 Step 3 - Check Data Applicability

Once the database has been selected then the data within that database must be reviewed to check its
applicability to the situation. This requires a review of the incidents included and the failure modes
described to determine if they are credible for the conditions being studied. If the process has been
significantly modified compared to the system from which the data originated then the data may not be
applicable.

17.4.4 Step 4 - Calculate Event Likelihood

When suitable incidents, failure modes and populations have been identified the event frequency can be
determined. Event likelihood is calculated by using the following equation:

Number of Incidents
Historical Frequency =
Exposed Population

For example, assume we have 5 major leaks from pressurised ammonia tanks from a population of
2500 vessel years. The:

5
Leak Frequency = = 0.002 per vessel year
2500

It may be necessary to adjust incident frequencies where the plant on which the historical data is based
and the plant under review are not identical.

17.4.5 Step 5 - Validate Frequency

The calculated frequency should be checked to highlight obvious errors or that wildly inaccurate data
has been used. This can be achieved by checking the data against a known population of plant or
equipment not used for the data generation.

In this course, two sources of data have been chosen to illustrate the form in which data is usually
presented and to enable example calculations to be performed. These are as follows:

• Pressure Vessels - Source: Smith and Warwick

• Pipelines - Various DNV Technica Sources

The data in the databases may be modified to suit specific applications where there are recognised
differences between the database conditions and the application. This is an area requiring knowledge of
the database data, as well as the various failure mechanisms. Care should be taken in modifying data
on the basis of elimination of cause categories since there may be other causes or factors not apparent
to the analyst. Similarly some databases include unknown cause categories or may have incorporated
unknown causes into the other known cause categories.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 7

17.5 EXAMPLE 1: PRESSURE VESSELS

17.5.1 Define Context

The aim of the study is to obtain failure frequency data for pressure vessels in petrochemical facility.
This is being done as part of a detailed QRA study on the facility.

17.5.2 Review Source Data

Of the many sources of data available, that of Smith and Warwick (1981) was chosen as the most
useful for the following reasons:

1. The vessel population is precisely defined, being pressure vessels registered for insurance purposes
with the Associated Offices Technical Committee.

2. As a qualitative description is given of every failure, it is possible to make an analysis of the failures
and to form a judgment as to the relevance to the vessels to be found on a petrochemical site.

17.5.3 Check Data Applicability

In most risk analysis, the boundaries of the pressure vessel are defined as being up to the flanged joint
on the nozzles of each vessel. Smith and Warwick define the boundary of the pressure vessel to include
associated pipework built to the same standard as the pressure vessel. In order to avoid double
counting the pipework, which in most studies are covered by the piping analysis, the overall failure rate
quoted by Smith and Warwick was not used. Moreover, it can be desired to have a distribution of leak
sizes rather than an overall leak rate, as given by Smith and Warwick.

The following procedure was therefore adopted when analysing the Smith and Warwick data:

• Failures described as being "Catastrophic Failure" were examined to decide if they were in the
vessel or in the associated pipework.

• Failure described as being "Leaks" were screened to eliminate those in associated pipework and an
attempt was made to classify the leaks by equivalent hole diameter from the short description
available.

17.5.4 Calculate Event Likelihood

Smith and Warwick quoted 12 failures as being catastrophic ruptures and 76 as leaks. Of these,
however, only 2 catastrophic ruptures and 44 leaks can be attributed to the pressure vessel itself.

20,000 pressure vessels were covered by the study giving a total of 310,000 vessel-years, including
data analysed by Smith and Warwick (1974). Tabulations of data in the Smith and Warwick reports
show a number of failures versus material, operating pressure and temperature, but population data are
not given to estimate frequencies for these aspects.

Table Error! No text of specified style in document..2 gives failure rates for pressure vessels from the
screened Smith and Warwick data sorted by cause. This data is presented in graphical form in Figure
Error! No text of specified style in document..2. From this figure the failure rate for holes less than
50 mm has been split into the two categories 5 mm (0 to 10 mm) and 25 mm (10 to 50 mm) as 3.7 ×
10-5 per vessel-year and 9.6 × 10-5 per vessel-year, respectively. The breakdown by cause of the

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 8

failures can be used, where appropriate, to adjust the base failure frequencies to reflect the specific
design and use of equipment.

Figure Error! No text of specified style in document..2 Distribution of Hole Sizes for Pressure Vessel
Failures

100%
96% at 150mm
Percentage with hole diameter ≤ D

89% at 50mm

54% at 25mm
50%

0%
0 25 50 75 100 125 150 175 200

Hole Diameter

Table Error! No text of specified style in document..3 summarises the data evaluated in this section.
Please note that these frequency values are different than those shown in Table Error! No text of
specified style in document..2.

The failure rate data for pressure vessels are generally also considered valid for drums and the chemical
reactors, as well as storage tanks.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 9

Table Error! No text of specified style in document..2 Source Data for Pressure Vessel Failure Rates
(Analysis of Smith and Warwick, 1981)

Cause of Failure Number of Leaks of Size

Small Large Very Large Catastrophic
(Up to 25 mm) (25 to 50 mm) (50 to 150 mm) (Over
150 mm)
Design Error 2 1 - -
Material/Construction Defect 5 3 1 -
Mechanical Wear 1 - - -
Fatigue 5 3 - -
Corrosion 1 3 - -
Corrosion Fatigue 1 - - -
Thermal Fatigue 1 - - -
Creep 2 - 1 -
Operational Overload 1 2 - 2
Unknown 6 4 1 -
Total 25 16 3 2
Failure Frequency 8.1 × 10-5 5.2 × 10-5 9.7 × 10-6 6.5 × 10-6
(per vessel-year)

Table Error! No text of specified style in document..3 Leak Frequencies of Pressure Vessels

Equivalent Hole Size (mm) Leak Frequency

(per item year)
5 3.7 × 10-5
25 9.6 × 10-5
100 9.7 × 10-6
Rupture 6.5 × 10-6
Total 1.5 × 10 -4

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 10

17.6 EXAMPLE 2: PIPELINE FAILURE FREQUENCY ANALYSIS

17.6.1 Define Context

The aim of the study is to obtain failure frequency data for a natural gas transmission pipeline to be
built in the UK. This is being done as part of a detailed QRA study on the transmission line.

The pipeline is to be 250 mm diameter and will be buried. The pipeline will be of steel construction,
seamless and designed and tested to modern codes and standards. It will be wrapped and cathodically
protected for corrosion protection.

17.6.2 Review Source Data

There are several data sources containing varying degrees of information; some are not available in
their raw data form, but only as a summary which may not give all the necessary information. These
cover transmission pipelines from different countries that have been in operation over several years.

The gas line data base used in this analysis has by far the greatest detail, and enables great flexibility of
analysis because of its availability in computerised raw form. This allows the analyst to exclude old
pipelines, and those not built to a suitable design code, from the statistics where that is relevant to the
study.

17.6.3 Check Data Applicability

The gas line data has been analysed in order to eliminate incidents which are not applicable to a natural
gas transmission pipe line. The following screening process has been applied:

1. Eliminating certain incidents from the data base because of their inapplicability to the present
systems. These classes were:

a) Those incidents where the leak occurred in any part of a pipeline system other than the
pipeline itself.

b) Those incidents where the leak originated at a longitudinal weld.

c) Those incidents which occurred within buildings.

d) Corrosion failure which occurred on pipe which was either not cathodically protected
or was bare.

e) Those construction defects or material failures which occurred on pipe which had not
been hydrostatically tested.

2. Eliminating certain classes of pipeline and their associated incidents from the databases, but
then also subtracting the mileage associated with these pipelines from the total mileage. These
were:
a) Pipelines classified as anything other than transmission lines. This was to exclude
distribution lines, and lines which may be lower pressure and lower specification. This
did not change the data base substantially as transmission lines contribute greater than
90% of the total mileage.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 11

b) Pipelines which were not steel.

c) Pipelines where the year of installation was given as pre 1950. This was on the
grounds that they would have been built to lower specifications in terms of design,
materials or construction techniques, than would be the case today. In addition, in the
case of external impact, the fact that the position of older pipes would not be so well
documented or marked would be expected to increase the risk of damage.

d) Pipelines neither coated not wrapped, and not cathodically protected. This exclusion
was made because the length of pipeline involved was very small and there was
concern that these pipes might be sub-standard or untypical in other respects.

17.6.4 Calculate Event Likelihood

Because the database was in electronic form allowed great flexibility in investigating the parameters
that would affect each failure mode.

The parameters that could be investigated were limited in certain cases by lack of information about the
mileage of pipelines concerned. Pipeline mileage data was available by year of installation, diameter,
type of duty (i.e. transmission, gathering, etc.), pipeline material and degree of corrosion protection.

Some of the parameters found to be important, such as cathodic protection and age of pipeline, were
dealt with fairly simply by the exclusion procedure described previously. However, one parameter,
diameter, was found to need more detailed analysis before any exclusion could be made. The failure
frequencies were therefore expressed as a function of diameter by dividing the number of failures for
each diameter by the amount of pipe experience of that diameter over the 11 years of reporting.

Preliminary inspection of the variation of frequency with diameter for all the failure modes indicates
that only one failure mode, external impact, is affected significantly by diameter. As wall thickness
increases with diameter, it is this fact that increases resistance against external impact. This has a
significant effect on the overall failure rate because of the large contribution of external impact to the
failure rate.

The variation with diameter for external impact is so that there is a decreasing frequency with
increasing pipe diameter. The frequency for this mode of failure was therefore obtained by reading off
the appropriate graph to give the probability for that pipe diameter. This was not considered necessary
for the other modes.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 12

Table Error! No text of specified style in document..4 shows the base frequencies estimated for a
250 mm (10 inch) standard buried pipeline.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 13

Table Error! No text of specified style in document..4 Base Failure Frequencies used for a 250 mm
Pipeline

Failure Mode Failure Frequency

per 1000 km year
Material Defect 0.0225
Construction Defect 0.0242
External Corrosion 0.0141
Internal Corrosion 0.0191
External Impact 0.1400
Natural Hazard 0.0150
Other 0.0138
Total 0.2487

Each failure mode is broken according to specific hole size information. This data is summarised in
Table Error! No text of specified style in document..5 for standard buried pipelines of 250 mm in
diameter.

Table Error! No text of specified style in document..5 Failure Frequencies by Failure Mode and
Hole Size for a Standard 250 mm Pipeline (Buried)

Failure Mode Failure Frequency per 1000 km year

5 mm 20 mm 50 mm Full Bore Total
Defect 0.0234 0.0145 0.0060 0.0029 0.0467
Corrosion 0.0186 0.0088 0.0038 0.0021 0.0332
External Impact 0.0846 0.0429 0.0093 0.0031 0.1400
Natural Hazard 0.0014 0.0014 0.0062 0.0060 0.0150
Other 0.0082 0.0041 0.0011 0.0004 0.0138
Total 0.1363 0.0717 0.0263 0.0144 0.2487

17.6.5 Validate Frequency

In the US the Department of Transportation produced data for 250,000 pipe-miles of gas transmission
lines. This data gives a final leakage frequency of 0.4 per 1000 kilometre-years for all leak sizes. This
is in reasonable agreement with the value of 0.25 calculated above.

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 14

Figure Error! No text of specified style in document..3 Distribution of Hole Sizes for Pipeline Failure

100%
100% at 1.0
Percentage with hole diameter ≤ D

94% at 0.2
84% at 0.08

50% 54% at 0.02

0%
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Hole Diameter d/D ratio

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 15

17.7 MEASURES OF EXPOSURE

The choice of units for the exposure in a frequency analysis determines the units of the calculated frequency.
When the frequency is applied in a QRA, the choice of units can also affect the risk predictions, and often has
a strong influence on the predicted effectiveness of risk reduction measures.

Some of the possible options are described below.

17.7.1 Frequency per Calendar Year

For example, if there have been 10 gas explosions on production platforms in the UK Sector over a 12 year
period, the frequency could be expressed as:

10 explosions
= 0.83 explosions per calendar year
12 years

This form may be used to help evaluate an FN curve predicted for an entire population, but is otherwise of
little value in QRA.

17.7.2 Frequency per Installation Year

If there were an average of 100 production platforms in service, the frequency could be expressed as:

10 explosions
= 8.3 x 10 -3 explosions per platform year
12 years x 100 platforms

This form is the most commonly used in QRA, and has the advantage that it is easy to relate to the
conventional form of risk results per year for a single installation. Its main limitation is that it implies that the
frequency is the same for any installation, regardless of the equipment on it.

17.7.3 Frequency per Equipment Item

If the events can be associated with individual modules or items of equipment, and equipment counts can be
obtained for these, the frequency can be expressed per equipment item. For example, if 2 of the events
resulted from leaks from gas compressors, and there were on average 2 compressors per platform, a partial
frequency could be expressed as:

2 explosions
= 8.3 × 10 -4 explosions per compressor year
12 years × 100 platforms × 2 compressor s

This form is desirable for QRA, and has the advantage that it takes account of the amount of equipment on
the installation. Its main limitation is that too much confidence may be given to the implicit assumption that
frequency is proportional the number of equipment items (rather than, say, equipment layout, the quality of
maintenance or the time in operation).

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 16

17.8 SOME HISTORICAL INCIDENT DATA SOURCES

Source Description Title

M&M Protection Consultants, “One Hundred Largest Losses”
1221 Avenue of the Americas, Annual review of large losses in the hydrocarbon-chemical
New York, New York 10020 industries. Ninth edition published April 1986. Free
Lees (1980) “Loss Prevention in the Process Industries”
Appendix 3 contains some case studies of major chemical
incidents and a chronological listing of many more
V. C. Marshall (1987) “Major Chemical Hazards”
Contains 40 case studies of major incidents.
Loss Prevention Bulletin Annual survey of chemical industry accidents (worldwide),
IChemE., UK. covering a wide range of accidents and with accident descriptions
J. H. Sorensen (1986) “Evacuations due to Chemical Accidents: Experience from 1980
to 1984”
Office of Radiation Programs “The Consequences and Frequency of Selected Man-Originated
US Environmental Protection Agency Accidents Events” NTIS PB80-211303
B. J. Robinson (1987) “A Three Year Survey of Accidents and World Dangerous
Occurrences in the UK Chemical Industry” Covers 1982 to 1984/5
J. A. Davenport (1983) “A Study of Vapour Cloud Incidents - An Update”
Lists UVCE incidents, cross-referenced to Gugan (see below)
K. Gugan (1979) “Unconfined Vapour Cloud Explosions”
Lists UVCEs; includes some BLEVEs and partially confined
explosions
P. Field (1982) Dust Explosions
Lists major incidents of this type
N. C Harris (1978) “Analysis of Chlorine Accident Reports”
Chlorine Institute, Washington, DC
U.S. Department of Transportation Pipeline Leak Reports for Onshore Gas Transmission and
Research and Special Programs Gathering Lines, and Liquid Lines (see also 3.1)
Administration
Office of Pipeline Safety Washington, DC
CONCAWE Annual reports of leaks from cross country pipelines in Europe
The Oil Companies European
Organization for Environmental and
Health Protection, The Hague,
Netherlands
R.F. De la Mare and O. Andersen “Pipeline Reliability Report”
Det Norske Veritas Analyses and compares onshore and offshore pipeline failure data
Oslo, Norway
Report No. 80-0572, Aug. 1980
Minerals Management Service Collects data on pipeline leaks in the Gulf of Mexico
Gulf of Mexico OCS Region,
1201 Elmond Park Blvd.,
New Orleans, LA70123
Office of Technology Assessment “Data on Transportation of Hazardous Materials: state and local
Washington, D.C. activities”
Marsh 1986 OTA -SET-301
Office of Technology Assessment “Data on Transportation of Hazardous Materials”
Washington, D.C.
July 1986 OTA-SET-340
M. Abkowitz and J. Galarraga (1985) “Tanker Accident Rates and Expected Consequences in US Ports
and High Seas Regions”

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 17

Source Description Title

National Transportation Safety Board Accident Reports
(NTSB) A detailed report is produced for transportation accidents
US, DOT, Washington, DC involving hazardous materials.
Hazardous Materials Accident Spill Maps
These give a map showing the location of the spill, any airborne
plume, site of fatalities, and injured people, at one or more times
after the start of the incident.
American Petroleum Institute “Summary of motor vehicle accidents in the June 1983 Petroleum
Industry for 1982”
Association of American Railroads Railroad Facts (Annual Editions)
Federal Railroad Administration Accident/Incident Bulletins (Annual)
US DOT Annual Reports on Hazardous Materials Transportation Incidents
Materials Transportation Bureau
Washington, DC
FACTS Computerised major incident data base (worldwide). Incidents
TNO Division of Technology for Society must have had potential for off-site impact to be included
P O Box 342
7300 AH Apeldoorn
Netherlands
SONATA (Summary of Notable Computerised data bank for incidents with hazardous materials
Accidents in Technical Activities)
TEMA
Via Medical de Vascello
26-Milano, Italy
WOAD (World Offshore Accident Data) Computerised data bank for World Offshore Data
DNV
Oslo, Norway
Hazardous Cargo Bulletin Annual Survey
ITACA Industry and Transportation accident Catalogue.
TER, TESCA, Risk and Reliability Specialises in fires and explosions
Italy Case histories
Bockholts et al (1986)
Fire Brigades, Ministry of the Interior, Fire Incidents
Italy Raw, unprocessed data
Bockholts et al (1986)
ISPESL Pressure vessel incident data
Central Institute for Health and Safety
Administration, Italy
Bockholts et al (1986)
GESIP Industrial accident case histories
France
Bockholts et al (1986)
IFP (Institute Francois du Petrol) Marine transport and offshore operation incidents
France
Bockholts et al (1986)
Berufsgenossinchaften Industrial accident data bases
Germany
Bockholts et al (1986)
CDMR (Central Environmental Control Industrial accident data base
Agency Rijnmond)
The Netherlands
Bockholts et al (1986)
BLS (Bureau of Labour Statistics) Industrial accident data
Bockholts et al (1986)

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 18

Source Description Title

OSHA (Occupational Safety and Health Industrial accident data
Administration)
Bockholts et al (1986)
NIOSH (National Institute for Industrial accident data
Occupational Safety and Health)
Bockholts et al (1986)
NFPA (national Fire Protection Fire incident data
Association)
Bockholts et al (1986)
CMA (Chemical Manufacturers Accident case histories in the chemical industry to 1979
Association)
Bockholts et al (1986)
IRI (Industrial Risk Insurers) Fire incident data
Bockholts et al (1986)
IOI (Loss Data Base) Information on over 1500 claims in excess of $10,000 in oil, gas
CWA Information and Research Ltd and petrochemical industries
18 Grosvenorblau, London

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved
Risk ID & Assessment Training Course DNV Consulting
Module 17: Frequency analysis using historical data Page 19

17.9 FAILURE CASE FREQUENCIES

Failure cases (such as those defined in Module 5) often group together failures of several individual
items. For example, a failure case for a line would generally include failures of flanges and valves
along the line. Hence the failure case frequency modelled needs to be the total frequency for all items
modelled by the case, so as not to ‘lose’ any failure frequency. (It is important to realise that, in a
QRA, although we model cases which are representative of all the possible releases which could occur,
these representative cases must nevertheless reflect the total frequency of all the possible releases.)

The methodology described in the preceding sections deals with estimating the generic frequencies of
failures of single components such as a pressure vessel or pipeline. These frequencies provide one set
of inputs to the failure frequency calculation in the study. The other set of inputs is a count of the
number of each component represented by the failure case. For example, along a line there may be 2
valves, each with 2 flanges. Hence the total failure case frequency will be the sum of frequencies for
the line + 2 valves + 4 flanges.

In general the failure case (incident) frequency is given by:

n
F= ∑n f
i=i
i i

where: F is the frequency of incident.

ni is the number of components i.
fi is the frequency of failure for component i.

This is illustrated (for a different combination of items) in Figure Error! No text of specified style in
document..4.

Figure Error! No text of specified style in document..4 Failure Case Frequency Calculation Method

Generic Frequencies
Leak Rupture
Pipe 1× 10-5 2 × 10-6
Valve 1 × 10-5 2 × 10-6
Case Frequencies
Leak Rupture
Hose 5 × 10-4 1 × 10-4
Pipe 1 × 10-5 2 × 10-6
Valve 7 × 10-5 1.4 × 10-5
Parts Count Hose 5 × 10-4 1 × 10-4
TOTAL5.8 × 10-4 1.2 × 10-4
Pipe 1
Valve 7
Hose 1

P:\2004 Contracts\21506545 PetroVietnam HAZOP+QRA Course\CD-ROM\Word files\Module17.Doc ©2004 DNV. All Rights Reserved