Page 324 - Pipeline Risk Management Manual Ideas, Techniques, and Resources

Failure prediction 14/301
Another way to view this is that unanticipated (undesigned tribution-number of failures versus time-with an average.
for) stresses to a pipeline system are being evaluated in the median, and standard deviation. This distribution describes the
third-par02 and incorrect operations indexes, and also in the likely failures that would accompany any pipeline section with
design index variables of surge and land movements. that particular score.
Unanticipated weaknesses are being evaluated in the corrosion For third-parv damage and incorrect operations, the rela-
index and in the design index variable of fatigue. The design tionship can perhaps be assumed to be a direct proportionality:
index variables of safep factor and integrit), verification are Failures increase proportionally with the index scores and are
mitigation measures to address all unanticipated stresses and constant over time for a fixed index score. Therefore, if the con-
weaknesses. In a conversion to absolute failure probability esti- ditionsremain constant. as evidenced by a constant index score.
mates, the role of the mitigation measures in all failure mecha- then the failure rate will be constant. This is supported by the
nism might need to be more directly measured than is shown in belief that the underlying causes for third-party damage and
the design index. See the discussion of load and resistance human error are largely random (assuming that all other factors
curves in Chapter 5. are equal). A corrosion failure probability score index may be
theorized to be related to corrosion rate by an exponential rela-
tionship. Note that the corrosion index measures the potential
VI. Failure prediction for and aggressiveness of corrosion, but not the time to failure
from corrosion. The latter requires incorporation of pipe wall
A good risk assessment always produces some estimate of fail- thickness, pipe stress levels, age, and other considerations. In
ure probability. Given the relationship between probability and the design index, both random forces (earth movements) and
frequency of future events, this estimate can also be seen as a some time-dependent mechanisms (fatigue) are at work.
predictive tool. As failure probability changes over time, so Therefore, this index could also be representing a non-constant
would a predicted leakibreak rate. In most transmission failure rate over time. The design index also plays a large role in
pipelines, insufficient system-specific information exists to determining the time to failure since it measures remaining
build a meaningful leakibreak prediction model--events are so wall thickness and pipe strength.
rare that any such prediction will have very large uncertainty To begin building a predictive model, a baseline deteriora-
bounds. Possible exceptions include situations in which time- tion rate can be established for time-dependent failure mecha-
dependent failure mechanisms can be more reliably tracked and nisms that assumes two end conditions: (I) There is no
behave in more predictable fashions. Distribution systems, probability of failure immediately after installation and (2)
where leaks are precursors to “failures,” are often more viable there is a 100% probability of failure after some selected time
candidates for prediction models. Where the evaluator believes in service. For example, a service life on the order of 200 years
that leakhreak predictions of useful accuracy are possible, she might be selected.
may wish to incorporate results from the risk assessment as dis- It is well understood that a pipe can fail immediately after
cussed below. The following paragraphs generally describe the installation; many pipes will fail long before 200 years have
philosophy for how the relative risk scores can support a future passed; and some pipes may survive beyond 200 years. Any
leak prediction model. service life can be selected as long as it has some plausibility. A
A leakibreak rate assessment should capture both time- straight-line deterioration rate of time-dependent index risk
dependent failure mechanisms such as corrosion and fatigue scores (such as corrosion) can initially be assumed for simplic-
and more random failure mechanisms such as third-party dam- ity-change in risk score is directly (or inversely, depending on
ages and seismic events. The random events will normally scoring regime) proportional to changes in leakhreak rate for
occur at a relatively constant rate over time for a constant set of time-dependent mechanisms. Note that more complex relation-
conditions. See the classic bathtub curve shape discussed in an ships such as a theorized exponential increase in leaks over
earlier chapter. time can also be used. The assumptions are initially for model-
A current risk state for each length of pipe is determined via ing convenience only. The assumptions can be readily changed
the risk model. This current state is based on all available infor- when better information becomes available or shoilld abandon-
mation. The risk scores can also represent a theoretical ment ofthe simplifications be warranted.
leakibreak rate for the pipe. This rate is called a “deterioration” Table 14.13 is a very generalized example of the type of
rate by some. but that phrase seems to be best applied to time- analysis described. In this example. a risk model as described in
dependent failure mechanisms only (corrosion and fatigue). Chapters 3 through 6 has been used to create index scores. An
This linkage between risk scores and a predicted leakibreak index score of 100 is theorized to mean a 100% chance of sur-
rate is logical because both are appropriately based on all vival for 200 years; a score of 40 indicates a 40% chance of sur-
known conditions. recent inspection (if available), design, vival for 200 years. etc. Theoretically, information flow is
operations, maintenance, and environmental considerations. continuous and complete so that the index scores are continu-
Both are assessed from indirect evidence unless and until a ously changing with changing conditions. The failure mecha-
physical inspection is performed to better establish actual con- nisms of third-party damage and incorrect operations (human
ditions. The physical inspection then supersedes the previous error) are assumed to be constant. The mechanisms of corro-
estimates and the risk scores are adjusted in light of the new sion and design (due to the fatigue component and increased
information. A series of inspections and changing risk scores at uncertainty of pipe integrity over time) are assumed to cause
the same location over a period of time verifies or corrects esti- deterioration (straight-line rate) in index scores representing
mated deterioration predictions. those failure mechanisms.
Even though they are expressed as a single value, each fail- On initial installation (day 1 ). index sum scores in this exam-
ure probability estimate or score really represents a failure dis- ple total almost 400 on a scale where 400 represents the highest