Page 44 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

Risk assessment models 2/23
either might be more appropriate for specific applications. Highest
Desired accuracy, achievable accuracy, intended use, and avail- risk
ability of resources are considerations in choosing an approach.
Most pipeline risk efforts generally fall into the “model” cate- High
gory-seeking to gain risk understanding in the most efficient
manner.
Although not always apparent, the most simple to the most Consequence ’I
complex models all make use of probability theory and statis-
tics. In a very simple application, these manifest themselves in
experience factors and engineering judgments that are them-
selves based on past observations and inductive reasoning; that
is, they are the underlying basis of sound judgments. In the
more mathematically rigorous models, historical failure data
may drive the model almost exclusively.
Especially in the fields of toxicology and medical research,
risk assessments incorporate dose-response and exposure -
assessments into the overall risk evaluation. Dose-response
assessment deals with the relationship between quantities of Il Low
exposure and probabilities of adverse health effects in exposed
populations. Exposure assessment deals with the possible Lowest Low<I Likelihood =>High
pathways, the intensity of exposure, and the amount of time a risk
receptor could be vulnerable. In the case of hazardous materi- Figure 2.1 Simple risk matrix.
als pipelines, the exposure agents of concern are both chemical
(contamination scenarios) and thermal (fire related hazards) in
nature. These issues are discussed in Chapters 7 and 14. analyses. Initiating events such as equipment failure and safety
system malfunction are flowcharted forward to all possible
Three general approaches concluding events, with probabilities being assigned to each
branch along the way. Failures are backward flowcharted to all
Three general types of models, from simplest to most complex, possible initiating events, again with probabilities assigned to
are matrix, probabilistic, and indexing models. Each has all branches. All possible paths can then be quantified based on
strengths and weaknesses, as discussed below. the branch probabilities along the way. Final accident probabil-
ities are achieved by chaining the estimated probabilities of
Matrix models individual events.
This technique is very data intensive. It yields absolute risk
One of the simplest risk assessment structures is a decision- assessments of all possible failure events. These more elaborate
analysis matrix. It ranks pipeline risks according to the likeli- models are generally more costly than other risk assessments.
hood and the potential consequences of an event by a simple They are technologically more demanding to develop, require
scale, such as high, medium, or low, or a numerical scale; from 1 trained operators, and need extensive data. A detailed PRA is
to 5, for example. Each threat is assigned to a cell of the matrix usually the most expensive of the risk assessment techniques.
based on its perceived likelihood and perceived consequence. The output of a PRA is usually in a form whereby its out-
Events with both a high likelihood and a high consequence put can be directly compared to other risks such as motor vehi-
appear higher on the resulting prioritized list. This approach cle fatalities or tornado damages. However, in rare-event
may simply use expert opinion or a more complicated applica- occurrences, historical data present an arguably blurred view.
tion might use quantitative information to rank risks. Figure 2.1 The PRA methodology was first popularized through oppo-
shows a matrix model. While this approach cannot consider all sition to various controversial facilities, such as large chemical
pertinent factors and their relationships, it does help to crystal- plants andnuclear reactors [88]. In addressing the concerns, the
lize thinking by at least breaking the problem into two parts intent was to obtain objective assessments of risk that were
(probability and consequence) for separate examination. grounded in indisputable scientific facts and rigorous engi-
neering analyses. The technique therefore makes extensive use
Probabilistic models of failure statistics of components as foundations for estimates
of future failure probabilities. However, statistics paints an
The most rigorous and complex risk assessment model is a incomplete picture at best, and many probabilities must still be
modeling approach commonly referred to as probabilistic risk based on expertjudgment. In attempts to minimize subjectivity,
assessment (PRA) and sometimes also called quantitative applications of this technique became increasingly comprehen-
risk assessment (QRA) or numerical risk assessment (NRA). sive and complex, requiring thousands of probability estimates
Note that these terms carry implications that are not necessarily and like numbers ofpages to document. Nevertheless, variation
appropriate as discussed elsewhere. This technique is used in in probability estimates remains, and the complexity and cost
the nuclear, chemical, and aerospace industries and, to some of this method does not seem to yield commensurate increases
extent, in the petrochemical industry. in accuracy or applicability [MI. In addition to sometimes
PRA is a rigorous mathematical and statistical technique that widely differing results from “duplicate” PRAs performed
relies heavily on historical failure data and event-treelfault-tree on the same system by different evaluators, another criticism