Method yields safety factor for in-line inspection data

Iain Colquhon, Arturo Menendez
Transportadora de Gas del Norte
Buenos Aires

Ricardo Dovico
Arcan Ingenería y Construcciones
Buenos Aires

A method for dealing with uncertainties in in-line inspection (ILI) information has been applied to a segment of corroded line in the gas-transmission system of Transportadora de Gas del Norte (TGN) in Argentina.

Results of applying the method indicate that the safety factor for ILI data is equivalent to that provided by a hydrostatic test.

The methods consist of combining the mean and dispersion of corrosion-defect depth (d) and length (L) data and simulating the effect of these variations on calculations for failure pressure (PF), using Monte Carlo propagation.

A maintenance philosophy is discussed from the perspective of likelihood of failures and of information available to support maintenance decisions.

TGN system

Transportadora de Gas del Norte, Buenos Aires, owns and operates two main trunklines (Fig. 1 [86,481 bytes]): ¹

• The 24-in. (610-mm), 3,000-km North Line was built in 1960 with field-applied asphalt coating. It transports gas from northern Argentina and Bolivia to Buenos Aires and markets en route.
• The 30-in. (762-mm), 1,400-km Center West Line was built in 1981 with field-applied polyethylene tape coating. It transports gas from central Argentina (Neuquen) to markets in the western part of the country and also to Buenos Aires.

The two lines make up the original state-owned and operated system privatized in 1992 to form TGN and Transportadora de Gas del Sur (TGS). Before privatization, an acute lack of investment in the system caused the state company to discontinue important maintenance.

On one section of the North Line, for example, galvanic anodes had been installed (because of a lack of local electric power) but had never been replaced. When these anodes were consumed, the remains induced rather than impeded corrosion.

It is worthwhile emphasizing here that the North Line presents the greater challenges as far as external corrosion is concerned. At the time of privatization, only 37% of the line had adequate cathodic protection, and before privatization, more than 4,000 repair sleeves had been installed.

Maintenance

The current maintenance program is the latter part of a plan in 1993 that consisted of evaluating pipeline integrity and ranking risk against the threat of leak or rupture. This process included examining external coating, cathodic protection, soil resistivity, in-line inspection data, pipeline age, pipe temperature, and operational history.

The plan also included an inspection and rehabilitation program.

The initial in-line inspections used low-resolution technology (24-in. corrosion tool with only 22 channels) that did not convey the lengths and profiles of the defects. Inspections of the North Line in 1996 and of the Center West line in 1997 used high-resolution magnetic flux leakage (MFL) tools. Of concern here are decisions that had to be made regarding the accuracy of the 1996 inspection data of the North Line (specifically with respect to Section 7 - Lavalle to Recreo).

Choosing a safety factor

It is well established that the original B31G criteria, embedded specifically in NAG-100 (the Argentine gas code), and ASME B31.8 are overly conservative. ^2-4 In recognition of this, these major codes permit the use of alternative methods of evaluation that are to be developed by the operating company and approved by regulators.

TGN chose to use the basic NG-18 criterion with the following features:

• Flow stress equal to specified minimum yield strength (SMYS) plus 10,000 psi (6.95 MPa)
• The modified Folias factor²
• A parabolic representation of the corrosion-defect profile.

TGN refers to this criterion as "NG-18 Modified."

TGN then applied a safety factor of 1.25 so that the company would repair any defect with a calculated failure pressure (PF) less than 1.25 times the maximum operating pressure (MOP). This criterion was adopted because of its good track record with Nova Gas Transmission in Alberta, Nova being the technical operator of the TGN system.

However, some differences between Nova's and TGN's transmission systems make it desirable to refine and customize this criterion for repairs in TGN.

This desire was brought sharply into focus when the results of the 1996 ILI data for Lavalle to Recreo were first reviewed.

In contrast to other sections of the line, when the corrosion depths and lengths predicted by the ILI were compared in this section with physical measurements in the field during the repairs, predicted failure pressures were significantly higher than indicated by the physical measurements.

(Both predictions used the modified NG-18 repair criterion; Fig. 2a [139,651 bytes])

When the inaccuracy was brought to the attention of the ILI service provider, the company revised its MFL conversion algorithms. In essence, it used the fundamental MFL signal and derivatives of the signal to identify a "basic length" of defect that better represented the metal-loss profile effect.

Using these improved algorithms, the vendor produced "reprocessed" results. (TGN distinguishes between "original" and "reprocessed" results.)

When TGN applied the reprocessed results to the physically measured sample, technicians noted the results appeared to be significantly conservative (Fig. 2b).

The number of required repairs associated with the two sets of results are shown in Table 1 [148,881 bytes]. (The table also contains a column for the second provider, discussed later.)

If TGN were to have used the reprocessed data directly, the cost added to those for the original repairs would have been $3.7 million (U.S.). From a rapid and largely intuitive assessment of the data, however, TGN believed the actual situation probably lay somewhere between these two cases.

That being so, there was a potential savings of some $1.8 million if a rational basis for evaluating these data could be developed. This evaluation would relate the factor of safety to the uncertainty in the ILI results.

The first decision to be made, therefore, was from among the following options:

• To continue with the existing criterion of 1.25 times MOP and accept the new (reprocessed) results at face value
• Spend the extra $3.7 million and accelerate the review of the repair criteria
• Reject the ILI results and seek more accurate inspections.

The second option was chosen as being the most feasible. However, because winter and anticipated high gas consumption were anticipated, contingency plans were put in place carefully to control the operating-pressure profile so that a reduced factor could be used during the winter.

As an added security measure and to provide a further point of reference, TGN also elected to re-inspect the section between Lavalle and Recreo, using a second service provider.

Method for choosing

The principle behind this methodology is to achieve, with a conservative repair criterion, a level of confidence in the integrity of the pipeline equivalent to that provided by hydrotesting.

With NAG 100 and perfectly reliable ILI data, that would mean repairing all cases for which PF< 110% MOP. A glance at Fig. 2 shows, however, that there is both a bias and a distribution associated with the data.

The principle therefore is to be interpreted as making the probability p(PF

Given the degree of conservatism remaining in the Modified NG-18 criterion, TGN considers p(PF< PH)<0.05 acceptable. The conservative nature of the criterion is shown in Fig. 3 [61,233 bytes] in which the actual failure pressure is compared with PF for the Modified NG-18 criterion and B31G.

These data are taken from rupture results² and show that, on average, the real PF exceeded that calculated by the Modified NG-18 by about 30%.

Another source of conservatism can be noted in the profiles of the defects themselves. Accurate measurements of the profiles are as yet unavailable, but from a random sample of 58 defects with depths of 30% or greater of the W.T., 17% conservatively fit an elliptical profile and the remaining 83% were somewhere between a "tornado" and a triangular profile.

That is to say, they were generally punctual regions of deeper corrosion contained within a length of general corrosion and thus had considerably more steel left to resist pressure loads than the elliptical shape would imply.

The NG-18 criterion essentially supplies a function for the calculation of the failure pressure (PF) in terms of the depth of the defect (d) and the length of defect (L): PF = PF(d,L). The variables 'd' and 'L' are independent of each other (Fig. 4 [88,359 bytes]) and from sample data (reprocessed) have distributions as shown in Fig. 5 [135,129 bytes].

Monte Carlo chosen

Three methods are available to combine these variations of parameters to yield a distribution of failure pressure. ^{5 6}

• Direct combination of the mean and standard deviations without regard for the distribution type but accounting for the non-linear nature of the function PF(d,L)
• So-called "reliability" methods (from their early use in predicting structural reliability)
• Direct simulation (Monte Carlo).

As an example of direct combination, consider first the conditions typically specified in TGN's ILI contract documents: for 85% of the data reported, the depth of defect predicted is to be within 10% of the wall thickness and the length is to be within ±6.35 mm.

The governing expressions are (for mean and standard deviation, respectively): µ(PF) = PF[µ(d), µ(L)] and J(PF) = µ[(JPF/Jd)²J²(d) + (JPF/JL)²J²(L)].

For p(PF

If we now consider the data in Fig. 5 in terms of their means and standard deviations (that is, without regard for the distribution type), the failure criterion for the reprocessed data would indicate PF<1.25 MOP.

If we then consider the distributions presented by the results of the second ILI service provider, we find that the criterion becomes PF<1.31 MOP. This factor, however, must be applied to a much smaller population of repairs (Table 1) and therefore is likely to yield a lower repair cost for the same level of reliability.

Reliability methods provide an efficient alternative to the more computationally onerous Monte Carlo technique. TGN did not consider reliability methods in the present work, however, because the response function PF(d,L) is relatively simple and the probability level associated with the failure criterion is not so small as to make the computation effort prohibitive.

It is now also relatively simple to perform the Monte Carlo propagation quickly and directly from a spreadsheet. The method6 essentially involves generating pairs of random numbers, applying them to an accumulative frequency distribution of 'd' and 'L' (and thereby selecting random values of 'd' and 'L' that reflect their distribution), calculating a failure pressure from the deterministic relationship PF(d,L), repeating this process a very large number of times, then sorting the data to give an accumulative frequency plot of PF.

At present, TGN considers the Monte Carlo simulation the more representative because it makes fewer arbitrary assumptions about the distributions of the variables. The results of this process are shown in Fig. 6a for the reprocessed data and in Fig. 6b for the second ILI provider. In both cases 10,000 iteration steps were required.

These results are still somewhat preliminary; benchmarking the reliability levels achieved will occur before application of the method to the line. Reworking the methodology with RSTRENG2 as the basis of the PF criterion will also take place.

Indications are, however, that TGN should expect safety factors according to the direct simulation of the order of 1.20 for the reprocessed data and 1.24 for the second ILI provider.

References

1. Dovico, R., and Montero, E., "The Evaluation and Restoration of a Deteriorated Buried Gas Pipeline," 1995 International Pipeline Conference, Calgary. 2. Kiefner, J., and Vieth, P., "A Modified Criterion for Evaluating the Remaining Strength or Corroded Pipe," PRC report PR 3-805, 1989. 3. Kiefner, J., and Vieth, P., Data Base of Corroded Pipe Test, April 1994. 4. Kiefner, J., Vieth, P., and Roytman, Itta., Continued Validation of RSTRENG, prepared for the Line Pipe Research supervisory committee, AGA Pipeline Research Committee, December 1996. 5. Gollwitzer, S., Abdo, T., and Rackwitz, R., First Order Reliability Method. Munich, 1988, 6. Rubinstein, R.Y.,. Simulation and the Monte Carlo Method. John Wiley & Sons Inc., McCormick, N.J. 1981.

The Authors