Page 45 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

2/24 Risk Assessment Process
includes the perception that underlying assumptions and of places where risks are relatively lower (fewer “bad” factors
input data can easily be adjusted to achieve some predeter- present) and where they are relatively higher (more “bad”
mined result. Of course, this latter criticism can be applied to factors are present).
any process involving much uncertainty and the need for An indexing approach to risk assessment is the emphasis of
assumptions. much ofthis book.
PRA-type techniques are required in order to obtain esti-
mates of absolute risk values, expressed in fatalities, injuries, Further discussion on scoring-type risk assessments
property damages, etc., per specific time period. This is the
subject of Chapter 14. Some guidance on evaluating the quality Scoring-type techniques are in common use in many applica-
of a PRA-type technique is also offered in Chapter 14. tions. They range from judging sports and beauty contests to
medical diagnosis and credit card fraud detection, as are dis-
Indexing models cussed later. Any time we need to consider many factors simul-
taneously and our knowledge is incomplete, a scoring system
Perhaps the most popular pipeline risk assessment technique in becomes practical. Done properly, it combines the best of all
current use is the index model or some similar scoring tech- other approaches because critical variables are identified from
nique. In this approach, numerical values (scores) are assigned scenario-based approaches and weightings are established
to important conditions and activities on the pipeline system from probabilistic concepts when possible.
that contribute to the risk picture. This includes both risk- The genesis of scoring-type approaches is readily illustrated
reducing and risk-increasing items, or variables. Weightings by the following example. As operators of motor vehicles, we
are assigned to each risk variable. The relative weight reflects generally know the hazards associated with driving as well as
the importance of the item in the risk assessment and is based the consequences of vehicle accidents. At one time or another,
on statistics where available and on engineering judgment most drivers have been exposed to driving accident statistics as
where data are not available. Each pipeline section is scored well as pictures or graphic commentary of the consequences of
based on all of its attributes. The various pipe segments may accidents. Were we to perform a scientific quantitative risk
then be ranked according to their relative risk scores in order to analysis, we might begin by investigating the accident statistics
prioritize repairs, inspections, and other risk mitigating efforts. of the particular make and model of the vehicle we operate. We
Among pipeline operators today, this technique is widely used would also want to know something about the crash survivabil-
and ranges from a simple one- or two-factor model (where only ity of the vehicle. Vehicle condition would also have to be
factors such as leak history and population density are consid- included in our analysis. We might then analyze various road-
ered) to models with hundreds of factors considering virtually ways for accident history including the accident severity. We
every item that impacts risk. would naturally have to compensate for newer roads that have
Although each risk assessment method discussed has its own had less opportunity to accumulate an accident frequency base.
strengths and weaknesses, the indexing approach is especially To be complete, we would have to analyze driver condition as it
appealing for several reasons: contributes to accident frequency or severity, as well as weather
and road conditions. Some of these variables would he quite
Provides immediate answers difficult to quantify scientifically.
Is a low-cost analysis (an intuitive approach using available After a great deal of research and using a number of critical
information) assumptions, we may be able to build a system model to give us
Is comprehensive (allows for incomplete knowledge and is an accident probability number for each combination of vari-
easily modified as new information becomes available) ables. For instance, we may conclude that, for vehicle type A,
Acts as a decision support tool for resource allocation driven by driver B, in condition C, on roadway D, during
modeling weather and road conditions E, the accident frequency for an
Identifies and places values on risk mitigation opportunities accident of severity F is once for every 200,000 miles driven.
This system could take the form of a scenario approach or a
An indexing-type model for pipeline risk assessment is a rec- scoring system.
ommended feature of a pipeline risk management program and Does this now mean that until 200,000 miles are driven, no
is hlly described in this book. It is a hybrid of several of the accidents should be expected? Does 600,000 miles driven guar-
methods listed previously. The great advantage of this tech- antee three accidents? Of course not. What we do believe from
nique is that a much broader spectrum of information can he our study of statistics is that, given a large enough data set, the
included; for example, near misses as well as actual failures accident frequency for this set of variables should tend to move
are considered. A drawback is the possible subjectivity of the toward once every 200,000 miles on average, if our underlying
scoring. Extra efforts must be employed to ensure consistency frequencies are representative of fume frequencies. This may
in the scoring and the use of weightings that fairly represent mean an accident every 10,000 miles for the first 100,000 miles
real-world risks. followed by no accidents for the next 1,900,000 miles-the
It is reasonable to assume that not all variable weightings will average is still once every 200,000 miles.
prove to be correct in any risk model. Actual research and fail- What we are perhaps most interested in, however, is the rela-
ure data will doubtlessly demonstrate that some were initially tive amount of risk to which we are exposing ourselves during a
set too high and some too low. This is the result of modelers single drive. Our study has told us little ahout the risk of this
misjudging the relative importance of some of the variables. drive until we compare this drive with other drives. Suppose we
However, even if the quantification of the risk factors is imper- change weather and road conditions to state G from state F and
fect, the results nonetheless will usually give a reliable picture find that the accident frequency is now once every 190,000