Sie sind auf Seite 1von 12

Proceedings of the 2012 9th International Pipeline Conference IPC2012 September 24-28, 2012, Calgary, Alberta, Canada

IPC2012-90539

USEFUL TRENDS FOR PREDICTING CORROSION GROWTH


Shahani Kariyawasam and Hong Wang TransCanada Pipelines Calgary, Alberta, Canada

ABSTRACT

The objective of an effective corrosion management program is to identify and mitigate corrosion anomalies before they reach critical limit states. Often as there are many anomalies on pipelines an optimized program will mitigate the few corrosion anomalies that may grow to a critical size within the next inspection interval, without excavating many of the anomalies that will not grow to a critical size. This optimization of the inspection interval and the selection of anomalies to mitigate depend on understanding of corrosion growth. Prediction of corrosion growth is challenging because growth with time is non linear and highly location specific. These characteristics make simplistic approaches such as using maximum growth rates for all defects impractical. Therefore it is important to understand the salient aspects of corrosion growth so that appropriate decisions on excavation and re-inspection can be made without compromising safety or undertaking undue amounts of mitigative activities. In the pipeline industry corrosion growth between two in line inspections (ILIs) has been measured by comparing one ILI run to the next. However many types of ILI comparison methodologies have been used in the past. Within the last decade or two comparison techniques have evolved from box matching of defect samples to signal matching of the total defect populations. Multiple comparison

analyses have been performed on the TransCanada system to establish corrosion growth rates. Comparison of the results from these various analyses gives insight into the accuracy and uncertainty of each type of estimate. In an effective integrity management process the best available corrosion growth data should be used. To do so it is important to understand the conservatism and the uncertainty involved in each type of estimate. When using a run-comparison to predict future growth it is assumed that the growth within the last ILI interval will continue (with associated uncertainty) during the next inspection interval. The validity of these assumptions is examined in this study. In the context of this paper these assumptions are validated for external corrosion on onshore pipelines. Characteristics of internal and offshore corrosion are very different in space and time variation. Correlations of external corrosion growth in onshore pipelines with defect size and location are also examined. Learning from multiple corrosion growth studies gives insight into the actual corrosion rate variation along a pipeline as well as general growth characteristics. Different types of corrosion growth modeling for use in probabilistic or deterministic integrity management programs are also discussed.

Copyright 2012 by ASME

INTRODUCTION

In line inspection (ILI) based integrity programs have been used for many decades to manage corrosion in oil and gas pipelines. The inline inspection detects and measures the defects at a point in time. However in following years corrosion growth needs to be accounted for and the pipeline industry has estimated this growth using many different techniques. Laboratory and field corrosion tests have improved our understanding of possible ranges of growth. However, as corrosion growth is extremely variable from defect to defect, the challenge in managing corrosion is to understand which defects will grow to a critical level and which defects will be dormant. To identify the critical defects from the many which are dormant or inactive the actual growth needs to be measured in-situ. Run-comparisons, comparing two consecutive ILI runs, measures actual growth experienced by the pipeline. These measured growth values do contain measurement errors or uncertainties. These errors depend on many factors such as time interval between consecutive runs, compatibility of technology and vendor used for each ILI, and if box matching or signal matching was performed. These causes for measurement error have been explored and are discussed in this comparison (Dawson and Kariyawasam 2009). When using a run-comparison to predict future growth it is assumed that the growth within the last ILI interval will continue (with associated uncertainty) during the next inspection interval. It is important to know if these assumptions hold in practice. The validity of this assumption is examined in this study. In the context of this paper these assumptions are validated for external corrosion on onshore pipelines. Characteristics of internal and offshore corrosion are very different in its spatial and temporal variation. Correlations of external corrosion growth in onshore pipelines with defect size and location are also examined. The characteristics of corrosion growth that are verified using these sets of data are useful in

understanding and predicting corrosion growth rates in pipeline integrity management. In reliability based corrosion management techniques, one must consider all variations and characteristics that cause variation in corrosion growth. All aspects that impact the decision to excavate and re-inspect need to be considered, so that failures are prevented without excessive excavations or unnecessary inspections. As these corrosion growth rate characterizations are to be used in a reliability based integrity management program the probabilistic characterizations are examined and discussed.
OBSERVATION AND INTERPRETATION OF MULTIPLE RUN-COMPARISONS

Some of the pipelines in the TransCanada system have experienced aggressive external corrosion and have been under ILI program for many years. Consequently, some lines have been inspected as many as 10 times. Even though the first few runs were done during the development stages of ILI and were with lower resolution tools, some have up to six high resolution ILIs performed and six runcomparisons. (Some run-comparisons are different types of comparison on the same set of ILIs.) One such pipeline is used for the main comparison of data. Six run-comparisons have been performed on this section of pipeline. They are: 1) 1990-1994 Box matching for each valve section These results were obtained as growth rates for the sample of defects that could be matched. It is not a random sample but a biased sample because the intention was to include the few defects that were found to be growing. The growth rates in each valve section are reported separately. The rates were statistically analyzed and found to form a Gumbel Max distribution which is peaked and has a long tail. This distribution is valve section specific and represents the different corrosion growth values (with measurement error) among the defects within a valve section. One such Gumbel distribution for a valve section is given in Figure 1. This Figure is plotted on Gumbel distribution paper and as the corrosion growth values plot in a relatively straight

Copyright 2012 by ASME

line it demonstrates a good fit to the gumbel distribution. A Gumbel distribution is an extreme value distribution that has a long upper tail. This shows that a few anomalies grow aggressively while the majority do not grow significantly. When comparing this run-comparison to others, the 95th percentile of each valve section specific distribution is shown in Figure 2 and 3 as a red chain line. 2) 2000-2004 signal matching of sample (population matching was not available at time of last run) This analysis segmented the pipeline into aggressive versus non-aggressive segments to identify hotspots. The results are segment specific, which means that a sample of data was chosen from each segment and the growth rate distribution for each segment represents the variability of growth in that segment among the sample of different defects. However, the sample is not random but chosen conservatively from more critical defect sizes and higher growth areas. Each segment specific distribution was described by the mean, standard deviation, and 95th percentile of growth rate in each segment. The 95th percentile value in each segment is plotted in dark blue long dashed line in Figure 2 and 3. 3) 2000-2004 signal matching of population (done in 2009 to estimate accurate growth) This is a signal matching on the whole population of defects. Therefore sampling bias and error are removed from this run-comparison. This set of data represents not only the faster growing defects but also the ones that are dormant or near-dormant which gives a more accurate representation of growth among all defects. Because the vast majority of the defects are not growing aggressively the 95th percentile for this data (solid light blue line) is significantly lower than any other estimate. The difference between 95th percentile of this data versus the 95th percentiles of the samples demonstrates the high conservatism in the samples. 4) 2004-2007 box matching across vendors The capability to perform box matching across vendors by using one set of signals and the other set

of signals in viewer form is still in development stage. This analysis was to investigate the feasibility. To check the raw box matching process a preliminary analysis was performed on static defects (defects that were recoated before both sets of ILI). The preliminary analysis found that there was a nonlinear correlation between measured sizes of static defects (probably due to different algorithms used by vendors). The non linear correction factor was applied to account for algorithm differences. A full description of this run-comparison was published by Spencer et al. (2010). The results as shown in Figure 2 and 3 are still highly variable compared to the other run-comparisons. The sample used to perform this analysis was the critical defects and faster growing defects therefore the sample is biased towards the most aggressive corrosion. The high variability in estimates is due to sampling, box matching, across vendor comparison, and the lower time interval between ILI runs. The 95th percentile is shown in grey double dashed grey line. 5) 2004-2011 signal matching of sample As the 2007 ILI was from a different vendor and as across vendor comparisons were too variable to be useful a signal matching was performed between 2004 and 2011 ILIs. First a sample was chosen for manual assessment and the 95th percentile of the results are shown in a dark purple dashed line in Figure 2 and 3. 6) 2004-2011 population automatic signal matching of

For comparison and to obtain the full population estimates automatic signal matching was performed on the 2004 and 2011 ILI data. The results are shown in the light purple as a heavy solid line in Figure 2 and 3. The light purple dots represent the defect specific growth rates. The population data is similar to the 2004-2007 run-comparison population data. Comparisons of the results for the five runcomparisons explained above are shown in the Figure 2 and 3. In Figure 2 the full trap to trap ILI section is represented.

Copyright 2012 by ASME

The 2000-2004 signal matching was done on a sample first and then on the full population. The comparison of these two lines in Figure 2 shows that the biased sample clearly overestimates the growth rate. The same trends are seen in the 2004-2011 sample and population data. The 1990-1994 and 2004-2007 were both box-matching while 20042007 was across vendors, across a shorter interval, and on a smaller more biased sample. Consequently the 2004-2007 run-comparison is highly inflated by all these uncertainties. The two sets of signal matching gives considerably lower uncertainty than the two sets of box matching, which confirms the theoretical basis for developing signal matching to improve accuracy. It is clear that sample based run-comparisons give conservatively biased growth distributions when compared to the population based results, as seen in Figure 2. The segment variation is significant and therefore if defect specific growth is not being used segmenting should be done based on aggressiveness. Box matching, across-vendor matching, and lower intervals between runs give higher variability in growth estimates. The uncertainty in the different types of runcomparisons is very different and consequently we cannot clearly distinguish a significant trend in corrosion growth with time. The best unbiased comparison is between the two sets of full population data. The growth rate distributions are very similar. Where the growth in 2004-2011 has defects that are growing faster the sample has captured these points which is reassuring. It is important to note that the general areas of highest growth remain the same across all the different growth data sets. This confirms the trend of external corrosion that has been observed by others to show that hotspots on a pipeline generally stay the same unless some remediation is performed (Desjardin 2010) When defect specific sets of growth data are added the graph involves a very large amount of data. Therefore in Figure 3 these full sets of data were added only for a small section of the pipeline shown

in Figure 2. This defect specific data is displayed in Figure 3 and it shows the significant difference between the full population growth data (light blue crosses or light purple dots) and a biased sample data (green triangles). Figure 3 shows the significant difference between the defect specific full population data and the sample defect specific data. As the vast majority of the defects are not growing significantly by taking a sample of the faster growing defects we would grossly overestimate the growth. The type of distribution that represents the population of growth rates would consequently be peaked and long tailed as shown in Figure 4. The significance of this accurate description is discussed further in the next section.
CORROSION RATE CHARACTERISTICS RELEVANT TO INTEGRITY MANAGEMENT

Corrosion integrity process is a defect management process. Defect sizing and growth rate are both defect-specific. The ILI results and specification provide the variation of sizing among defects. Defect growth rate is highly defect-specific and the results from a sample as opposed to the full population are very different as shown in above section. In corrosion growth the vast majority of the defects are hardly growing while a few defects are growing quite aggressively. This is shown by the highly peaked low value mode of the corrosion growth distribution with a thick long tail - as in Figure 4 where a typical example of corrosion growth is modeled. Consequently, the segment specific corrosion depth growth rate is modeled using a Weibull, Gumbel max, or Lognormal distribution. When a segment specific growth rate is applied to all defects the majority of the defects that are hardly growing are grown at rates higher than actual. This leads to many unnecessary future integrity excavations.

Copyright 2012 by ASME

The few defects that are actually growing aggressively are grown in the assessment at a growth rate lower than actual. For example in Figure 4 there are defects, from a defect-specific growth analysis, that are estimated to grow at 0.3 mm/yr. However, when a segment specific growth rate is used it is grown at a rate represented by a distribution that has a mean of 0.077 mm/yr and a standard deviation of 0.075 mm/yr. This is a weakness in using segment specific growth rates for a highly defect-specific phenomenon. The effects of segment specific growth versus defect specific growth on integrity decisions are explained in detail by Kariyawasam and Peterson (2010). As there are many parameters that influence the corrosion process, corrosion growth tends to be quite localized and often occurs in hot-spots as seen in Figure 2 and 3. The localized nature and consequent appropriate methods of obtaining corrosion growth rates are explored by Dawson and Kariyawasam (2009). Another very important aspect of corrosion growth is seen in Figure 5 where previously growing defects slow down in growth while other defects in the vicinity start growing more aggressively. This cannot be captured if the clusters are grown at the same rate. When individual defects within the cluster are grown at defect-specific rates this variation is captured.

Figure 6 shows that the deepest defects do not grow the fastest in future. There are only a few deep defects in a typical ILI run and they have a lower rate of growth. The smaller defects, which are the majority, have a large variation in growth; some of them grow moderately but a few of the smaller defects grow at a much higher rate. But obviously this high rate is not sustained because as they grow deeper the rate drops. For example, for a 20% wall thickness original defect to grow to a 70% wall thickness defect at a constant high rate of 0.5 mm/yr it will take 8 to 10 yrs depending on wall thickness. Because the constant high rate is not sustained it will take more than 10 years in general. Therefore we do not see smaller defects growing fast to failure within commonly used inspection intervals of less than 10 yrs. Consequently in past failures the larger defects have led to failures before the next inspection. Figure 7 shows the trend of past growth against resultant depth. Here the medium defects had the highest growth because the past smallest defects that grew aggressively grew to be midsize defects. There are very few deep defects which show that very few medium defects grew to be deeper defects. When we use run-comparison data to grow the existing defects at the rate at which it was growing at previously, the growth rate is applied as in Figure 7. This is in general a conservative assumption because the medium and deeper defects will continue to be grown at higher than actual rates. As shown in Figure 6 and 7 the rate at which it was growing is in general higher than what it will be growing at. Consequently when we use run-comparison results for future growth the medium defects that did grow aggressively are assumed to grow at the same aggressive rate which is overly conservative because, as seen in Figure 6, medium defects do not grow at that aggressive rate. When segment specific growth is applied to all defects, all sizes of defects are grown similarly. As deepest defects do not grow at the highest rates, as shown in Figure 6, growing the whole segment with same growth rate will lead to deepest defects being

The run-comparison data measures the growth between two ILI results. The external corrosion growth rate against the original depth is examined in Figure 6. Conversely the external corrosion growth rate against the subsequent (resultant) depth is examined in Figure 7. Therefore Figure 6 represents how defects will grow in the future while Figure 7 represents how defects of a certain size grew in the past.

Copyright 2012 by ASME

grown at overly conservative rates leading to over conservative defect prioritization with time.
SUMMARY AND CONCLUSIONS

aggressively growing defects are used to represent the growth of all defects. The general areas of highest growth along the pipeline remain the same across all the different growth data sets (if all operational conditions remain the same). This confirms the trend of external corrosion that has been observed by others to show that hotspots on a pipeline generally stay the same unless some remediation (or operational change) is performed. Previously growing defects slow down in growth while other defects in the vicinity start growing more aggressively. This cannot be captured if the clusters are grown at the same rate. When individual defects within the cluster are grown at defect-specific rates this variation is captured. Investigations also showed that the deepest defects do not grow the fastest in future. There are only a few deep defects in a typical ILI run and they have a lower rate of growth. The smaller defects, which are the majority, have a large variation in growth; some of them do not grow fast but a few of the smaller defects grow at a much higher rate. But this high rate is not sustained because as they grow deeper the rate drops. When we use run-comparison data to grow the existing defects at the rate at which it was growing previously, the growth rate applied is in general a conservative assumption because the medium and deeper defects will continue to be grown at higher than actual rate. The rate at which it was growing previously is in general higher than what it will be growing at in the future. Consequently when we use run-comparison results for future growth the medium defects that did grow aggressively are assumed to grow at the same aggressive rate which is conservative because medium defects do not grow at that aggressive rate. When segment specific growth is applied to all defects, all sizes of defects are grown similarly. As deepest defects do not grow at the highest rates, growing the whole segment with same growth rate will lead to deepest defects being grown at overly

Corrosion growth is highly variable between defects. Within a segment the vast majority of the defects hardly grow while a few defects are growing quite aggressively. This leads to a highly peaked corrosion growth distribution among defects of a segment with a low value mode and a thick long tail. Consequently corrosion growth in a segment of pipe with many defects is modeled by a Gumbel, Weibull, or Lognormal distribution. Due to these characteristics of corrosion growth using segment specific growth on all defects leads to inaccurate representation of growth characteristics. Applying defect specific growth will avoid unnecessary growth of the vast majority of the defects that are not growing and will apply the higher growth rates to defects that are growing aggressively. Due to techniques like run-comparison defect-specific growth can be estimated for the whole defect population. Therefore, it is now feasible to apply defect-specific growth in a reliability analysis. Many types of run-comparison have been used in the past. Multiple run-comparisons performed on the same section of pipeline enabled a critical evaluation of the types of run-comparison techniques. Signal matching gives considerably lower uncertainty than the box matching, which confirms the theoretical basis for developing signal matching to improve accuracy. It is clear that sample based runcomparisons give conservatively biased growth distributions when compared to the population based results. The segment variation is significant and therefore if defect specific growth is not being used segmenting should be done based on aggressiveness to capture variation along the pipeline. Box matching, across-vendor matching, and lower intervals between runs give higher variability in growth estimates. Defect specific data shows the significant difference between the full population growth data where many defects have negligible growth and a biased sample data where the few

Copyright 2012 by ASME

conservative rates leading to over conservative defect prioritization with time.


REFERENCES

Dawson, S. J. and Kariyawasam, S. N., Understanding and Accounting for Pipeline Corrosion Growth Rates, Joint Technical Meeting, Paper No. 22, Milan, Italy May 2009 Desjardin, G. Detection of Active Corrosion From a Comparison of ILI Runs, International Pipeline Conference, Paper No. IPC2010-31395, Calgary, September 2010 Kariyawasam, S. N. and Peterson, W. B., Effective Improvements to Reliability Based Corrosion Management, Proceedings of IPC 2010, International Pipeline Conference, Paper No. 201031425, Calgary, September 2010 Spencer, K., Kariyawasam, S. N., Tetreault, C., Wharf, J., A Practical Application to Calculating Corrosion Growth Rates by Comparing Successive ILI Runs From Different ILI Vendors, International Pipeline Conference, Paper No. 2010-31306, Calgary, September 2010

Copyright 2012 by ASME

Cumulative probability

Growth rate (mm/yr)


Figure 1: Box matching (1990-94) results from one valve section

Copyright 2012 by ASME

0.8
Valve Specific 96 Morrison (95 percentile)

0.6

0.4

Segment Specific 00-04 (95 percentile)

Growth Rate (mm/yr)

0.2

Segment specific 00-04 Auto (95 percentile)

0 0 -0.2
Segment specific 04-11 (95 percentile)

20000

40000

60000

80000

100000

120000

140000

Segment specific 04-07 (95 percentile)

-0.4

-0.6

Segment specific 04-11 Auto (95 percentile)

-0.8 Odometer (m)

Figure 2: Run-comparisons on ILI Trap to Trap Section (P95 = 95th percentile)

Copyright 2012 by ASME

Valve Specific 96 Morrison (95 percentile) Segment Specific 00-04 (95 percentile) Segment specific 00-04 Auto (95 percentile) Segment specific 04-07 (95 percentile) Segment specific 04-11 (95 percentile) Defect specific 0004 Auto Defect Specific 04-07 Defect specific 0411 Auto Defect specific 0411 Segment specific 04-11 Auto (95 percentile)

0.8

0.6

0.4

Growth Rate (mm/yr)

0.2

0 92000 -0.2

94000

96000

98000

100000

102000

104000

106000

108000

110000

112000

-0.4

-0.6

-0.8

-1

Odometer (m)

Figure 3: Run-comparisons on ILI Sub Section with Defect Specific Data (P95 = 95th percentile)

Mean = 0.077 mm/yr Std dev = 0.075 mm/yr

Figure 4 Corrosion depth growth distribution from 584 defects in a segment

10

Copyright 2012 by ASME

Figure 5 Run-comparison example of the defect-specific growth within clusters (Dawson and Kariyawasam 2009)

Figure 6: Original Depth versus Depth Growth Rate

11

Copyright 2012 by ASME

Figure 7: Resultant Depth versus Depth Growth Rate

12

Copyright 2012 by ASME

Das könnte Ihnen auch gefallen