A.2 Frequency Analysis
A.2.1 Need for failure rate data
One result of risk being defined in mathematical terms of frequency or probability is the necessity of having (or being able to estimate) failure rate data for the cause of the originally specified undesired event. Ideally, historical data for the exact causes of events encountered would produce the most accurate risk predictions for an activity. However, assembling such data in sufficient detail is notoriously difficult except in specialised industries or situations (e.g. the nuclear industry or aircraft crash investigation), where the value of such data to the owners or operators is clear. Recourse often has to be made to more generalised sources of data in order to carry out risk calculations.
Examples of data sources that could be pertinent to the separation, compression and transmission of CO2 are given in A.2.2.
Although such data may not represent the exact failure mechanisms likely to be encountered for carbon dioxide, data can be adjusted to account for known variations in physical conditions. They can also be, used to calculate the likelihoods of 'worst case' outcomes. Normally, acceptability of risk criteria are defined in terms of orders of magnitude or conservatively-set bands of tolerability. Thus, generic data may be adequate to screen out which risks are likely to be unacceptable, without further detailed consideration of avoidance, protection or mitigation; for this more accurate use of data may be required. In the UK, the concept of having to demonstrate that the detrimental effects are being kept ALARP, means that additional cost-effective improvements to reduce risk should be considered under the UK health and safety regime.
A.2.2 Introduction to failure frequency data sources
Failure rate data may be obtained in a number of ways:
- By sample testing - usually of mechanical or electrical components in a specific test environment.
- From plant experience – by companies from reliability based data collection, or by organisations collating and analysing industry or nationwide incident reports.
- From data banks and literature sources – much of the plant experience data and component information (as above) is reported in this way.
- By predictive techniques – appropriately combining component data on constituent parts of a complex system, e.g. by using fault tree analysis.
Generally, the usable working data for risk assessment will be from data banks and literature sources. However, there are potential drawbacks with their use in that much of the information may come from sources such as the nuclear, aerospace or defence industries. Here the necessary high quality of installation, maintenance and testing may require a 'negative' allowance when used in conjunction with data collected from less stringently controlled industrial areas. Essentially, it may be optimistic to suggest that all industrial areas have the same approach to maintaining machinery, and are bound by the same regulations.
In addition, literature sources will inevitably be historical compilations and can suffer from being incomplete or out of date for modern applications. For example, Lees' Loss Prevention in the Process Industries89 although currently published as the third edition in 2005 contains pipeline event and failure data references (Table 23.1) for only up to the end of the 20th century. Lees' list of principal reliability data sources (Table A14.1) cites Davison90 published in 1994. (Note: Tables 23.1 and A14.1 are not reproduced in this document.)
Published data considered appropriate as sources forof CO2 separation, compression and pipeline transmission systems are given in the following sections. However, care will be required even in the use of such selective data because of the potential effects of differences between CCS systems and the source industries.
A.2.3 Example data sources for pipeline failure rates
There have been only a small number of pipeline failures in total (relative to the lengths installed worldwide), and the number of failures of pipelines of the diameters that are likely to be associated with CCS applications is much smaller. Relevant data for gas pipelines are a subset of this and, given that CCS is an emergent technology, the data on CO2 pipelines are less still. In assembling a data set of pipeline failures, the whole sample size of both onshore and offshore is relevant, as it provides a context. One general conclusion is that by far the most common cause of failure is third party damage. Caution is urged when projecting the statistics for onshore pipelines to offshore pipelines, as this main failure mechanism is much less likely under water, even if it might remain the most likely cause of failure.
A.2.3.1 PARLOC 2001
There are a number of well-established international failure databases for gas pipelines. Probably the most comprehensive database for offshore gas pipelines is available in the report published by UK HSE entitled PARLOC 2001 (Pipeline And Riser Loss Of Containment). The most recent version of this database91 covers incidents from the 1960s until 2000. The information in this database is based on data obtained from the regulatory authorities in the UK, Norway, the Netherlands, and Germany, operators in the UK, Dutch and Danish sectors of the North Sea, and published sources.
The total number of pipelines (steel and flexible) is over 1 500, and the total length is almost 25 000km to the end of 2000.
The main causes of pipeline failure, as identified from a review of the PARLOC 2001 data, are listed in Table A.1. The PARLOC data suggest that anchor/impact, followed by internal corrosion, are the main contributors to subsea pipeline failures. Since internal corrosion is usually considered negligible for these pipelines, the likelihood of a loss of containment from subsea pipelines can be assessed primarily based on the potential for external or third party damage (TPD) along the pipeline route.
This factor has been considered in drawing up and allocating likelihood and severity to the list of failure scenarios in A.8.
Table A.1 Causes of subsea pipeline incidents from PARLOC 2001
A.2.3.2 European Gas Pipeline Incident Data Group (EGIG)
EGIG is a cooperation of 12 major European gas transmission system operators and is the owner of an extensive data base of pipeline incident information collected since 1970.
EGIG has maintained and expanded the European gas pipeline data base. The transmission companies now collect data on more than 122 000 km of pipeline each year. The total exposure, which expresses the length of a pipeline and its period of operation, is 2,77 million km.yr.
The statistics of all incidents collected in the database give failure frequencies. The seventh report92 gives an overall incident frequency equal to 0,37 incidents per year per 1 000 km over the period 1970 to 2007. The five-year moving average, which represents the average incident frequency over the last five years reported equals 0,14 per year per 1 000 km. This frequency is almost six times lower that that reported in the first years of the data base. Failure frequencies have been reducing regularly year by year although the rate of change has fallen in recent years.
The reported major cause of incidents remains external interference (third party damage) (50 % of all incidents), followed by construction defects/material failures (16 %) and corrosion (15 %).
The latest UKOPA93 report presents collaborative pipeline and product loss incident data from onshore Major Accident Hazard Pipelines (MAHPs) operated by National Grid, Scotia Gas Network, Northern Gas Network, Wales and West Utilities, UK, B P, Huntsman and UK, covering operating experience up to the end of 2006. The data cover reported incidents on unfenced pipelines (i.e. not within a compound), where there was an unintentional loss of product from the pipeline. Unlike the Europe-wide EGIG, this UKOPA database contains extensive data on pipeline failures and on part-wall damage, allowing prediction of failure frequencies for pipelines for which inadequate failure data exist. It is worth noting that these data are in the 'public domain' as the assets were deemed accessible to the public.
The overall failure frequency over the period 1962 to 2006 is 0,248 incidents per 1 000 km.year, whilst for the EGIG data in the previous section this figure was 0,263 incidents per 1 000km.year (covering the period from 1962 to 2004).
The failure frequency over the five-year period up to the end of 2006 is 0,028 incidents per 1 000 km.year, which remains unchanged from the figure in the previous report (covering the five-year period up to the end of 2004).
At over 35 000 km the inventory covered currently includes the vast majority of such pipelines in Europe, transporting around 800 million m3 per year of crude oil and refined products. The latest CONCAWE report94 covers the performance of these pipelines in 2006 and a full historical perspective since 1971. The performance over the whole 36 years is analysed in various ways including gross and net spillage volumes The causes are grouped into five main categories: mechanical failure, operational, corrosion, natural hazard and third party.
Twelve spillage incidents were reported in 2006, corresponding to 0,34 spillages per 1 000 km of line, slightly above the five-year average but well below the long-term running average of 0,56, which has been steadily decreasing over the years from a value of 1,2 in the mid 1970s.
Half the incidents were related to mechanical failures, four incidents to third party activities and two to corrosion. Over the long term, third party activities remain the main cause of spillage incidents.
A.2.3.5 Pipeline and Hazardous Material Administration (U.S. Department of Transport)
Statistics on pipeline incidents in thecan be found at the Office of Pipeline Safety (OPS) within the U.S. Department of Transportation, Pipeline and Hazardous Materials Safety Administration.
CO2 pipeline failure data95 are contained within the hazardous liquid accident data despite CO2 being both a gas when released at ambient conditions and classed as non-hazardous under DOT regulations. These data are the only specifics related to transmission of compressed supercritical CO2. The CO2 is used for through a system of onshore pipelines over approximately 5 000 km of network.
Det Norske Veritas (DNV) has analysed the data96 for CO2, and reports corrosion to be the major single cause of failure for the US system through the period 1986-2008. A separate analysis97 of the same data through to 2002 reported an incident rate of 0,33 per 1 000 km.year, which is higher than pipeline failure data reported from the US hydrocarbon pipeline transmission system. However, the authors caution on drawing conclusions from such a comparison because the CO2 system sample size is small.
A.2.3.6 Overall summary of failure data
A summary of the pipeline failure data in incidents per 1 000 km.year is given in Table A.2. Of these, the PARLOC 2001 data are probably the most relevant, as they relate solely to offshore incidents.
|PARLOC 2001||EGIG||UKOPA||CONCAWE||US DoT|
|Latest five-year rolling average||Not provided||0,14||0,028||0,34||NA*|
* Evaluated data are for early period of operation only.
Reported failures of past five years indicate that the rolling average will be higher than the overall value given.
Table A.2 Pipeline failure data summary (incidents per 1 000 km.year)
A.2.4 Selecting appropriate failure rate data for pipelines
Whilst there is a small body of failure rate data for CO2 pipelines and a larger body of data for other pipelines, it is important to understand the likely failure modes for CO2 versus other pipelines. This will help to ensure that appropriate comparisons are being drawn.
A.2.4.1 Third party interference
For CO2 pipelines, failure by third party interference is likely to be comparable with all pipeline types. It remains possible that future pipelines of this nature, constructed in Europe, could follow routes similar to those of existing buried petrochemical pipelines. On this basis it can be suggested that third party interference is of a similar likelihood.
This suggestion has supporting data that can be found from the EGIG, which cites third party interference as the largest failure mechanism. Where pipelines are running over ground it may be better to draw upon US data, as North America has more uncovered pipelines carrying CO2. Offshore pipelines are different, but the frequency of outside interference in relation to failure remains the same.
Corrosion is a common failure mode for pipelines and is the largest single cause cited for US CO2 pipelines. The applicability of corrosion data will depend on the design and operation of the pipeline. Corrosion risk should be mitigated through the rigorous control of moisture, ensuring that there is no free water and insufficient dissolved water to reduce hydrate formation across the range of operating conditions within the pipeline. Control and inspection in the event of water ingress should be included in the operating procedures.
The design of the pipeline, such as the material selected and methods of construction, has a direct bearing on the likelihood of failure by corrosion mechanisms.
When screening projects in the early design phase, using a range of failure frequency data should develop an understanding of howand operating techniques may impact on the likelihood and severity of a failure.
A.2.4.3 Other failure modes
Other failure modes for pipelines can be designed out or considered specifically. This includes appropriate selection of seals and valves for CO2 service and in particular the careful use of elastomers designed for carbon dioxide service. These are designed against failure from explosive decompression, and will mitigate against failure at valve locations alongside an appropriate maintenance regime. The design should take into account the possibility of different chemical compositions which theproduct may contain. The pipeline should also be able to withstand these differences whilst operating in abnormal conditions. Secondary effects of a failure should also be considered. These could include: further brittle failure through local cold temperature effects; ground movement giving rise to movement of the pipeline or displacement of its supports; secondary damage to surrounding equipment from the effects of a rapidly cooling high pressure release, and movement of any debris during the initial moments of release.
A.2.5 Plant equipment failure data
Lees' Loss Prevention in the Process Industries98, and the Offshore Reliability Data Handbook 4th Edition99, (OREDA) contain references to plant equipment failure rate data from numerous sources. As previously discussed component data from such sources could be synthesised into appropriate plant failure data if the nature and size of the plant (process) was available.
A convenient ,though dated, compilation of plant equipment leak data is given in Appendix 8 of Cox, Lees and Ang100. Cox et al demonstrate how to put together leak frequency and size data for a defined 'standard plant'. This is dealing specifically with the problem of estimating frequencies of ignition of flammable leaks for classification. However, it has relevance to QRA for the risks from loss of containment of any physical process, such as for CO2, even though flammability is not an issue.
A.2.6 Selecting appropriate failure rate data for plant
There are less data available for failure rates at plants, although existing data sources provide a valuable starting point. There would be benefit in gathering further data on CO2/CCS specific equipment failure rates (for equipment in service in CCS and other applications); this should be gathered in a systematic way to improve failure rate estimations.
89 Lees' Loss Prevention in the Process Industries 3rd edition Mannon S. (ed), 2005 (Elsevier Butterworth – Heinemann, Oxford UK)
90 Reliability of Mechanical Systems, Davidson J.(ed), 1994 (IMechE, London UK)
91 PARLOC 2001: The Update of Loss of Containment Data for Offshore Pipelines
92 Seventh report of the European Gas Pipeline Incident Data Group Gas Pipeline Incidents 1970-2007 Published Dec 2008
93 Document 07/0050 - UKOPA Pipeline Fault Database - Pipeline Product Loss Incidents (1962 - 2006)
94 CONCAWE Report No. 7/08 Performance of European cross-country oil pipelines - statistical summary of reported spillages in 2006 and since 1971
96 Mapping of potential HSE issues related to large-scale capture, transport and storage of CO2, Johnson,K. et al, DNV report number 2008-1993, 2009 (Det Norske Veritas , Horvik. Norway)
97 Gale, J. and Davidson, J. Transmission of CO2—safety and economic considerations, Energy, 29, 2004: pp.13191328
98 Loss Prevention in the Process Industries, Frank P Lees, Third Edition edited by Sam Mannan, 2005
99 Offshore Reliability Data Handbook 5th Edition, Volume 1 – Topside Equipment, Volume 2 – Subsea Equipment (OREDA 2009)
100 Classification of Hazardous Locations published 1990, Cox, Lees and Ang, (Institution of Chemical Engineers, Rugby UK)