Page 49 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

2/28 Risk Assessment Process
operations area. This should not be universally assumed, detail and complexity. Appendix E shows some samples of risk
however, during the data-gathering step. algorithms. Readers will find a review of some database design
concepts to be useful (see Chapter 8).
Persistence of segments
Data first or framework first?
Another decision to make is how often segment boundaries
will be changed. Under a dynamic segmentation strategy, There are two possible scenarios for beginning a relative risk
segments are subject to change with each change of data. assessment. In one, a risk model (or at least a framework for a
This results in the best risk assessments, but may create prob- model) has already been developed, and the evaluator takes
lems when tracking changes in risk over time. Difficulties can this model and begins collecting data to populate her model’s
be readily overcome by calculating cumulative risks (see variables. In the second possibility, the modeler compiles a list
Chapter 15) or tracking specific points rather than tracking of all available information and then puts this information into a
segments. framework from which risk patterns emerge and risk-based
decisions can be made. The difference between these two
Results roll-ups approaches can be summarized in a question: Does the model
drive data collection or does data availability drive model
The pipeline risk scores represent the relative level of risk development? Ideally, each will be the driver at various stages
that each point along the pipeline presents to its surround- of the process.
ings. It is insensitive to length. If two pipeline segments, One of the primary intents of risk assessment is to capture
say, 100 and 2600 ft, respectively, have the same risk score, and use all available information and identify information gaps.
then each point along the 100-ft segment presents the Having data drive the process ensures complete usage of all
same risk as does each point along the 2600-ft length. data, while having a predetermined model allows data gaps to
Of course, the 2600-ft length presents more overall risk be easily identified. A blend of both is therefore recommended,
than does the 100-ft length because it has many more risk- especially considering possible pitfalls of taking either exclu-
producing points. A cumulative risk calculation adds the sively. Although a predefined set of risk algorithms defining
length aspect so that a 100-ft length of pipeline with one how every piece of data is to be used is attractive, it has the
risk score can be compared against a 2600-ft length with a potential to cause problems, such as:
different risk score.
As noted earlier, dividing the pipeline into segments 0 Rigidity of approach. Difficulty is experienced in accepting
based on any criteria other than all risk variables will lead to new data or data in and unexpected format or information
inefficiencies in risk assessment. However, it is common that is loosely structured.
practice to report risk results in terms of fixed lengths such as Relative scoring. Weightings are set in relation to types of
“per mile” or “between valve stations,” even if a dynamic information to be used. Weightings would need to be
segmentation protocol has been applied. This “rolling up” of adjusted if unexpected data become available.
risk assessment results is often thought to be necessary for
summarization and perhaps linking to other administrative On the other hand, a pure custom development approach
systems such as accounting. To minimize the masking effect (building a model exclusively from available data) suffers
that such roll-ups might create, it is recommended that several from lack of consistency and inefficiency. An experienced
measures be simultaneously examined to ensure a more com- evaluator or a checklist is required to ensure that significant
plete use of information. For instance, when an average risk aspects of the evaluation are not omitted as a result of lack of
value is reported, a worst-case risk value, reflecting the information.
worst length of pipe in the section, can be simultaneously Therefore, the recommendation is to begin with lists of
reported. Length-weighted averages can also be used to better standard higher level variables that comprise all of the critical
capture information, but those too must be used with caution. A aspects of risk. Chapters 3 through 7 provide such lists for
very short, but very risky stretch of pipe is still of concern, even common pipeline components, and Chapters 9 through 13 list
if the rest of the pipeline shows low risks. In Chapter 15, a sys- additional variables that might be appropriate for special situ-
tem of calculating cumulative risk is offered. This system takes ations. Then, use all available information to evaluate each
into account the varying section lengths and offers a way to variable. For example, the higher level variable of activity (as
examine and compare the effects of various risk mitigation one measure of third-party damage potential) might be cre-
efforts. Other aspects of data roll-ups are discussed in Chapters ated from data such as number ofone-call reports, population
8 and 15. density, previous thirdparty damages, and so on. So, higher
level variable selection is standardized and consistent, yet the
model is flexible enough to incorporate any and all informa-
IV. Designing a risk assessment model tion that is available or becomes available in the future. The
experienced evaluator, or any evaluator armed with a compre-
A good risk model will be firmly rooted in engineering con- hensive list of higher level variables, will quickly find many
cepts and be consistent with experience and intuition. This useful pieces of information that provide evidence on many
leads to the many similarities in the efforts of many different variables. She may also see risk variables for which no infor-
modelers examining many different systems at many different mation is available. Similar to piecing together a puzzle, a
times. Beyond compatibility with engineering and experience, picture will emerge that readily displays all knowledge and
a model can take many forms, especially in differing levels of knowledge gaps.