Page 354 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

QRAquality 14/329
of accidents that would affect one or more persons, on average, is less How to evaluate the quality of a QRA
than 6.OE-07 chancedyear, or one chance in 1.7 million.
The following are common errors seen in QRA studies.
Historical data on fatal accidents involving natural gas gathering A check for these can be used as a simple audit of a QRA
and transmission pipelines have been compiled by the Department of report.
Transportation (DOT). During the 14.5 years for which summary data
are available, the maximum number of fatalities due to any single 0 Failure to define clearly the scope and boundaries of the
accident was six, and only two accidents caused six fatalities. For the study
pipeline/well network involved in this study, the maximum expected 0 Failure to cover all relevant hazards
number of fatalities for any single accident is five on average.
Insufficient failure cases
To put this type of evaluation into perspective, it is instructive to 0 Screening of failure data, optimistic assumptions and other
look at the types of risks people are ordinarily exposed to during day- biases tending to produce low-risk results
to-day life. There are voluntary activities (driving a car) and involun- 0 Concentration of modeling and risk reduction effort on haz-
tary activities (being hit by lightning) that involve risks higher than ards that do not dominate the risks
those due to this pipeline/well network. 0 Lack of attention to escalation ofhydrocarbon events
0 Use of only one risk measure
0 Failure to define individual risk
Use of assumptions even though data are available
XII. QRA quality 0 Failure to provide references (or an auditable internal refer-
encing system) for quoted frequency and probability data
The above case studies are based on numerical analyses tech- 0 Use of insufficient sources for frequency and probability
niques, making them at least akin to QRA-type evaluations. It is data
useful to examine characteristics that might make one QRA- 0 Lack ofattention to risk reduction measures.
type approach generally more complete or more useful as a
valid measure of risk. The following lists are extracted from The following are features that may indicate a high-quality
checklists offered by a well-known practitioner of QRA, as a QRA, equal to the best QRAs currently being performed:
means of evaluating the QRA study itself.
Reference [91] states that from an analyst’s point of view, a 0 Use of formal hazard identification procedure linked to fail-
QRA study can be thought of as having three general stages: ure case generation
0 Use of intelligent failure case and accident scenario selection
Use of validated software for modeling
TOP Use of audited software for risk summation
Establishing the objectives and scope ofthe study 0 Documentation of all input data and modeling assumptions
Collecting all relevant information 0 Traceability of risk results through intermediate results to
Identifying what can go wrong input data
0 Quantitative uncertainty analysis, including identification of
MIDDLE the most critical assumptions and exploration of the effects
Estimating event frequencies of alternatives (Note that the existence and application of an
Performing consequence modeling uncertainty analysis is a better indicator of quality than the
Calculating risk results degree of uncertainty that is estimated.)
0 A smooth FN curve Experience with FN curves has sug-
TAIL gested that more detailed analyses, especially those with
Investigating risk reduction measures intelligently selected failure cases, tend to produce smoothly
Developing cost-effective solutions rounded F’N curves, whereas less detailed studies produce
Communicating the results FN curves with large discontinuities (unless some additional
smoothing is used).
From a client’s point of view, many QRAs seem to place too 0 Use of actual accident experience in developing accident
much emphasis on the technical details in the middle, at the scenarios and validating risk results [91].
expense of the top and, more notably, the tail. This is perhaps not
unexpected since the middle part is the most technically demand- However, many ofthese aspects might not be ofhigh value to
ing. However, the tail part is arguably the most valuable to the the user of the analysis, so not all of these ingredients is neces-
client since it is usually acritical aspect of decision making. sarily appropriate for every study.