Diane J. Hovey, Edward J. Farmer
EFA Technologies Inc.
Sacramento
The probability of a spill occurring along a pipeline lies at the core of risk management for pipeline operators. Thus, a look at historical accident trends may provide some insight into this probability.
Analyses of data for U.S. petroleum product pipelines operating between 1982 and 1991 indicate that such pipelines of short-to-moderate lengths (for example, 50 miles) are likely to have at least one reportable spill within a 20-year period. Longer lines (as much as 1,000 miles, for example) may suffer a reportable spill within 1 year.
These are major conclusions of analyses by EFA Technologies Inc., Sacramento, of statistics compiled by the U.S. Department of Transportation (DOT) on liquid pipelines operated under the Code of Federal Regulations (CFR) Title 49D, Part 195 Transportation of Hazardous Liquids by Pipeline.
These regulations are based on the Pipeline Safety Act of 1968 and amendments, the Hazardous Liquid Pipeline Safety Act of 1979 and amendments, and the Hazardous Materials Transportation Act of 1979 and amendments. 1
In 1992, moreover, the data show that 52.5% of the oil spilled in the U.S. in accidents of more than 10,000 gal each came out of pipelines. Worldwide, pipelines caused 51.2% of the leaks involving this magnitude.
DOT REGS
U.S. DOT regulations present legally minimum standards for design, construction, testing, operation, and maintenance of these lines.
These regulations require reporting to DOT of all pipeline accidents that involve explosion or fire, the loss of 50 bbl or more of liquid, the loss of 5 or more bbl of highly volatile liquid, the death of any person, bodily harm to any person, or estimated property damage exceeding $5,000 (Article 195.50).
The DOT compiles accident statistics from these reports.2
EFA Technologies studied these statistics for 1982 through 1991 because this 10-year period encompassed the time over which various safety programs were implemented (for example, one-call, hydrotesting, smart pigging) and because it was long enough to provide adequate data from which to calculate the statistics and reliably determine the trends.
In its published summary of these statistics, EFA addressed the frequency and cause of accidents but not their probability of occurrence. 3
Following completion of that summary, EFA obtained from the DOT information on the miles of line subject to regulation during the study period.
This information, taken with the accident frequency data, allows determination of meaningful failure rates. The available data cover several pipelines, many miles of pipe, many operators, a significant period of time, and many accidents.
Statistics can be used to estimate the behavior of systems for which the statistical data are relevant. In this case, the failure rates would be expected to be typical of all pipelines designed, constructed, tested, operated, and maintained in accordance with DOT Part 195.
FAILURE RATES
Failure-rate analysis is a component of system safety and reliability analysis. The methodology is widely used in industry and is increasingly appearing as an agency requirement for granting construction and operating permits.
The methodology used as the basis for all calculations here is described in Nuclear Regulatory Commission NUREG 0492.4
In essence, this methodology computes anticipated failure rates and the probability a system will fail over a period of time on the basis of observed failure rates over time.
As long as the underlying system is unchanged, the failure data should remain relevant. If the underlying system is changed, the effect would have to be determined by investigation and analysis.
A review of DOT accident data for the 10-year study period produced a compilation of the annual quantity of accidents that occurred in each category DOT uses to classify the cause of accidents.
Analysis disclosed no statistically significant trends, favorable or unfavorable, in any of the cause categories. 2 3 Accordingly, dealing with accident frequency using the 10-year average data is appropriate.
During the study period, the DOT regulated an average of 214,155 miles of liquids pipeline. Year-to-year variations are quite small, on the order of 1,000 miles. Accordingly, the failure-rate calculations are based on this average mileage.
So that individual pipeline operators can relate to the numbers more easily, pipeline failure rates are expressed as "per 1,000 miles per year." For this calculation, the 10-year-total number of accidents in each "cause" category was divided by 10 years (to get annual frequency) and the result divided by 214.155 (to place this frequency in terms of thousands of miles).
Therefore, the failure rate (r) = number of accidents in a category/(10 years x 214.155 miles).
Table 1 presents the results of these calculations. The cause categories are in the first column and the 10-year total number of accidents is in the second column. The third column shows the failure rate in failures per 1,000 miles per year.
The largest number of accidents (31%) appear in the "outside force" category. Consequently, in 1985 DOT began subdividing outside force accidents into subcategories.
Table 2 shows the failure rate data for each subcategory for the 6-year period comprising 1986 through 1991.
FAILURE CALCULATIONS
Failures do not occur exactly when failure rates might seem to indicate. That is, they occur randomly over the exposure time.
According to the methodology described in NUREG 0492, the occurrence of failures is assumed to exhibit a Poisson distribution. The Poisson distribution has been found to describe the behavior of many "rare event" occurrences, regardless of the underlying physical processes.
Fig. 1 shows the cumulative effect of exposure in a Poisson-distributed process. The horizontal axis represents the exposure in mile-years and the vertical axis shows the probability a failure will occur.
For example, a 100-mile pipeline with a 30-year life would have a lifetime exposure of 3,000 mile-years and a 93% probability of a failure during its lifetime.
Use of the Poisson distribution assumes these pipelines are under stable conditions in which start-up failures (the so called "infant mortality") and failures due to the lines being worn-out (at the ends of their service lives) are not a factor.
Because DOT Part 195 requires testing and certification before operation and requires that the lines be maintained in a specified serviceable condition, this assumption is warranted.
Under this methodology, the probability a failure will occur (F) during an exposure time (t) is expressed in the following:
F = 1 - e-rt
As previously discussed, r the failure rate. Because we computed the failure rate using years, t must be in years.
Because the failure rate is "per thousand miles," F = the probability a failure will occur in a 1,000-mile pipeline during t years. F is usually expressed as a percentage.
This calculation can be generalized to lines of different lengths by adjusting the failure rate proportionally. If a line has a length of miles (L), the adjusted value (r') would be calculated from the following:
r' = r(L/1,000)
The failure probability (F) for that pipeline would then be calculated as previously discussed, but using r'.
Assuming that the failure rate is directly proportional to length is not perfectly rigorous but without additional data not currently available there is no better or more defensible assumption.
OPERATOR EXPOSURE
Table 3 presents the probability that pipelines of different lengths will fail over various periods of time from any cause. Note that it is very probable most pipelines tabulated will have a reportable failure during their useful life.
Table 4 presents the probability that a line will fail in a specific DOT cause category. Three different length lines are presented: 25 miles, 100 miles, and 1,000 miles.
Table 5 presents the probability that the pipelines of Table 4 will fail in a specific "outside force" subcategory.
The failure probability tabulated by cause category (Tables 4 and 5) is useful for estimating the exposure of a particular pipeline. For instance, if the pipeline is not located in a region where frostheave can occur, it would be anticipated that the probability the line will fail in that mode will be lower than reported in the tables.
These failure rates and probabilities are not specific to line pipe or other components: They apply to entire pipelines. Because the accident data show only that a reportable accident occurred on a regulated pipeline, the resulting statistics pertain to regulated pipelines in their entirety.
In other words, the failure statistics represent failures that will occur on a "pipeline" as opposed to the individual components (for example, line pipe) from which the line is constructed. In that regard, this analysis is less specific than the "component based" method described in NUREG 0492 would envision.
This is not a serious limitation for the present purpose because the safety of pipeline systems as a whole is the item of interest. Nonetheless, component failure data would be useful in assessing where the next measure of mitigative effort should be spent.
DATA CORRELATION
The Oil Spill Intelligence Report, which identifies, investigates, and reports oil spills worldwide of more than 10,000 gal, has summarized worldwide oil spill statistics for 1992. 5
It reports that 52.5% of the oil that was spilled in the U.S. in 1992 (in spills of more than 10,000 gal each) came from pipelines. Worldwide, pipelines accounted for 85 out of a total of 166 spills, or 51.2%.
These failure rates are similar to the results of other investigations.
Notably, the Province of Alberta's Energy Resources Conservation Board investigated all pipeline failures in the province between 1975 and 1982 (Table 6). This study indicated pipeline failure rates exceeding by three times the rates indicated by DOT data. 6
During the period covered by the Canadian study, the miles of pipeline more than doubled. The number of accidents increased but less than proportionally.
This indicates safety improvements occurred that reduced the accident frequency from 8/year/1,000 km in 1975 to 5/year/1,000 km in 1982. The comparable failure statistic, as calculated above, for DOT Part 195 lines is 1.43 failure/km/year (0.89 failure/1,000 miles/year).
Because the DOT Part 195 regulated pipelines are a subset of classes of lines included in the Canadian study and because the study periods are different, the statistics cannot be directly compared.
ACKNOWLEDGMENT
The authors acknowledge the assistance provided by the U.S. Department of Transportation and particularly Sherry Boerner of the John A. Volpe Research Center, Cambridge, Mass., for assistance preparing of this article.
REFERENCES
- U. S. CFR49D Part 195.
- U.S. Department of Transportation, "1982-1991 Liquid Pipeline Accident Report," February 1992.
- Hovey, Diane J., and Farmer, Edward J., "Trends in the Incidence and Cost of Liquid Pipeline Accidents from 1982 to 1991," EFA Technologies Inc., Sacramento, 1992.
- Roberts, N.H., Vesely, W.E, Haasl, D.F., and Goldberg, F.F., NUREG 0492, Fault Tree Handbook, U.S. Nuclear Regulatory Commission, January 1986.
- Oil Spill Intelligence Report, Mar. 18, 1993 (Cutter Information Services Corp.).
- An Analysis of Pipeline Performance in Alberta (D83-G, Energy Resources Conservation Board, Calgary), 1983.
Copyright 1993 Oil & Gas Journal. All Rights Reserved.
