Page 38 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

Choosing a risk assessment technique 111 7
What data do you have? Comparisons can be made against fixed or floating “stan-
What is your confidence in the predictive value of the dards” or benchmarks
data?
What resources are available in terms of money, person- Finally, a view to the next step, risk management, should be
hours, and time? taken. A good risk assessment technique will allow a smooth
What benefits do you expect to accrue in terms of cost sav- transition into the management of the observed risks. This
ings, reduced regulatory burdens, improved public support, means that provisions for resource allocation modeling and the
and operational efficiency? evolution ofthe overall risk model must be made. The ideal risk
assessment will readily highlight specific deficiencies and
These questions should be kept in mind when selecting point to appropriate mitigation possibilities.
the specific risk assessment methodology, as discussed further in We noted previously that some risk assessment techniques
Chapter 2. Regardless ofthe specific approach, some properties are more appropriately considered to be “building blocks”
of the ideal risk assessment tool will include the following: while others are complete models. This distinction has to do
with the risk assessment’s ability to not only measure risks, but
Appropriate costs. The value or benefits derived from the also to directly support risk management. As it is used here. a
risk assessment process should clearly outweigh the costs of complete model is one that will measure the risks at all points
setting up, implementing, and maintaining the program. along a pipeline, readily show the accompanying variables
Ability to learn. Because risk is not constant over the length driving the risks, and thereby directly indicate specific system
of a pipeline or over a period of time, the model must be able vulnerabilities and consequences. A one-time risk analysis-a
to “learn” as information changes. This means that new data study to determine the risk level-may not need a complete
should be easy to incorporate into the model. model. For instance, an event-tree analysis can be used to esti-
Signal-to-noise ratio. Because the model is in effect a meas- mate overall risk levels or risks from a specific failure mode.
urement tool, it must have a suitable signal-to-noise ratio, as However, the risk assessment should not be considered to be
discussed previously. This means that the “noise,” the a complete model unless it is packaged in such a way that it
amount of uncertainty in the measurement (resulting from efficiently provides input for risk management.
numerous causes), must be low enough so that the “signal:’
the risk value of interest, can be read. This is similar to the Four tests
accuracy of the model, but involves additional considera-
tions that surround the high level of uncertainty associated Four informal tests are proposed here by which the difference
with risk management. between the building block and complete model can be seen.
The proposition is that any complete risk assessment model
should be able to pass the following four tests:
Model performance tests
1. The “I didn’t know that!” test
(See also Chapter 8 for discussion of model sensitivity analy- 2. The “Why is that?” test
ses.) In examining a proposed risk assessment effort, it may be 3. The “point to amap” test
wise to evaluate the risk assessment model to ensure the 4. The “What about -?’test
following:
Again, these tests are very informal but illustrate some key
All failure modes are considered characteristics that should be present in any methodology that
All risk elements are considered and the most critical ones purports to be a full risk assessment model. In keeping with the
included informality, the descriptions below are written in the familiar,
Failure modes are considered independently as well as in instructional voice used as if speaking directly to the operator
aggregate of a pipeline.
All available information is being appropriately utilized
Provisions exist for regular updates of information, includ- The “I didn ’t know that! ” test (new knowledge)
ing new types of data
Consequence factors are separable from probability factors The risk model should be able to do more than you can do in
Weightings, or other methods to recognize the relative your head or even with an informal gathering of your experts.
importance of factors, are established Most humans can simultaneously consider a handful of factors
The rationale behind weightings is well documented and in making a decision. The real-world situation might be influ-
consistent enced by dozens of variables simultaneously. Your model
A sensitivity analysis has been performed should be able to simultaneously consider dozens or even
The model reacts appropriately to failures of any type hundreds of pieces of information.
Risk elements are combined appropriately (“and” versus The model should tell you things you did not already know.
“or” combinations) Some scenario-based techniques only tend to document what is
’ Steps are taken to ensure consistency of evaluation already obvious. If there aren’t some surprises in the assess-
Risk assessment results form a reasonable statistical distri- ment results, you should be suspicious ofthe model’s complete-
bution (outliers?) ness. It is difficult to believe that simultaneous consideration of
There is adequate discrimination in the measured results many variables will not generate some combinations in certain
(signal-to-noise ratio) locations that were not otherwise intuitively obvious.