Page 26 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

Basic concepts 115
will often provide answers that are highly inappropriate data, however, are not generally available in sufficient quantity
estimates of probability. or quality for most event sequences. Furthermore, when data
Even in cases where past frequencies lead to more reliable are available, it is normally rare-event data4ne failure in
estimates of future events for populations, those estimates are many years of service on a specific pipeline, for instance.
often only poor estimates of individual events. It is relatively Extrapolating future failure probabilities from small amounts
easy to estimate the average adulthood height of a class of third of information can lead to significant errors. However, his-
graders, but more problematic when we try to predict the height torical data are very valuable when combined with all other
of a specific student solely on the basis of averages. Similarly, information available to the evaluator.
just because the national average ofpipeline failures might be 1 Another possible problem with using historical data is the
per 1,000 mile-years, the 1,000-mile-long ABC pipeline could assumption that the conditions remain constant. This is rarely
be failure free for 50 years or more. true, even for a particular pipeline. For example, when histori-
The point is that observed past occurrences are rarely suffi- cal data show a high occurrence of corrosion-related leaks, the
cient information on which to base probability estimates. Many operator presumably takes appropriate action to reduce those
other types of information can and should play an important leaks. His actions have changed the situation and previous
role in determining a probability. Weather forecasting is a good experience is now weaker evidence. History will foretell the
example of how various sources of information come together future only when no offsetting actions are taken. Although
to form the best models. The use of historical statistics (clima- important pieces of evidence, historical data alone are rarely
tological data-what has the weather been like historically on sufficient to properly estimate failure probabilities.
this date) turns out to be a fairly decent forecasting tool (pro-
ducing probability estimates), even in the absence of any mete- Failure rates
orological interpretations. However, a forecast based solely on
what has happened in previous years on certain dates would A failure rate is simply a count of failures over time. It is usually
ignore knowledge of frontal movements, pressure zones, cur- first a frequency observation of how often the pipeline has
rent conditions, and other information commonly available. failed over some previous period of time. A failure rate can also
The forecasts become much more accurate as meteorological be a prediction of the number of failures to be expected in a
information and expert judgment are used to adjust the base given future time period. The failure rate is normally divided
case climatological forecasts [88]. into rates of failure for each failure mechanism.
Underlying most of the complete definitions of probability is The ways in which a pipeline can fail can be loosely catego-
the concept of degree of belief: A probability expresses a degree rized according to the behavior of the failure rate over time.
of belief. This is the most compelling interpretation of proba- When the failure rate tends to vary only with a changing environ-
bility because it encompasses the statistical evidence as well as ment, the underlying mechanism is usually random and should
the interpretations and judgment. Ideally, the degree of belief exhibit a constant failure rate as long as the environment stays
could be determined in some consistent fashion so that any two constant. When the failure rate tends to increase with time and is
estimators would arrive at the same conclusion given the same logically linked with an aging effect, the underlying mechanism
evidence. It is a key purpose of this book to provide a frame- is time dependent. Some failure mechanisms and their respective
work by which a given set of evidence consistently leads to a categories are shown in Table 1.1. There is certainly an aspect of
specific degree of belief regarding the safety of a pipeline. randomness in the mechanisms labeled time dependent and the
(Note that the terms likelihood. probability, and chance are possibility of time dependency for some of the mechanisms
often used interchangeably in this text.) labeled random. The labels point to the probability estimation
protocol that seems to be most appropriate for the mechanism.
Frequency, statistics, and probability The historical rate of failures on a particular pipeline system
may tell an evaluator something about that system. Figure 1.1 is
As used in this book, frequency usually refers to a count of past a graph that illustrates the well-known “bathtub shape of fail-
observations; statistics refers to the analyses of the past obser- ure rate changes over time. This general shape represents the
vations; and the definition ofprobability is “degree of belief,” failure rate for many manufactured components and systems
which normally utilizes statistics but is rarely based entirely on over their lifetimes. Figure 1.2 is a theorized bathtub curve for
them. pipelines.
A statistic is not a probability. Statistics are only numbers or
methods of analyzing numbers. They are based on observa- Table 1.1 Failure rates vs. failure mechanisms
tions-past events. Statistics do not imply anything about
future events until inductive reasoning is employed. Therefore, Nature of Failure rate
a probabilistic analysis is not only a statistical analysis. As pre- Failure mechanism mechanism tendency
viously noted, probability is a degree of belief. It is influenced
by statistics (past observations), but only in rare cases do the Corrosion Time dependent Increase
statistics completely determine our belief. Such a rare case Cracking Time dependent Increase
would be where we have exactly the same situation as that from Third-party damage Random Constant
which the past observations were made and we are making esti- Laminationsiblistering Random Constant
Constant
mates for a population exactly like the one from which the past Earth movements Random (except for
slow-acting instabilities)
data arose-a very simple system. Material degradation Time dependent Increase
Historical failure frequencies-and the associated statistical Material defects Random Constant
values-are normally used in a risk assessment. Historical