Your SlideShare is downloading. ×
IPC_2010
IPC_2010
IPC_2010
IPC_2010
IPC_2010
IPC_2010
IPC_2010
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

IPC_2010

224

Published on

Estimation of Corrosion Rates by Run Comparison: a stochastic scoring methodology

Estimation of Corrosion Rates by Run Comparison: a stochastic scoring methodology

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
224
On Slideshare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
13
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Proceedings of the International Pipeline Conference IPC 2010 September 27 – 1 October, 2010, Calgary, Alberta, Canada IPC2010-31576 ESTIMATION OF CORROSION RATES BY RUN COMPARISON: A STOCHASTIC SCORING METHODOLOGY Érika S. M. Nicoletti Ricardo D. de Souza Petrobras Transporte S.A Petrobras Transporte S.A Rio de Janeiro, RJ, Brazil Rio de Janeiro, RJ, Brazil ABSTRACT Pipeline operators used to map and quantify corrosiondamage along their aging pipeline systems by carrying out NOMENCLATUREperiodical in-line metal-loss inspections. Comparison with thedata sets from subsequent runs of such inspections is one of the • σi: standard deviation of depth metal-loss populationmost reliable techniques to infer representative corrosion in a defect neighborhood [mm];growth rates throughout the pipeline length, within the period • σli: standard deviation of depth metal-loss populationbetween two inspections. Presently there are two distinct in a defect neighborhood [mm];approaches to infer corrosion rates based on multiple in-line • ∆tf: forecasting time interval [years];inspections: individual comparison of the detected defective • ∆ti: time interval between inspections [years];areas (quantified by more than one inspection), and comparison • d1: earlier inspection metal-loss depth average [mm];between populations. The former usually requires a laborious • d2: latest inspection metal-loss depth average [mm];matching process between the run-data sets, while the drawback • dfi: forecast defect depth [mm];of the latter is that it often fails to notice hot-spot areas. The • di: pig-reported defect depth [mm];object of this work is to present a new methodology which • dj: pig-reported defect depth [mm];allows quick data comparison of two runs, while still • dli: metal-loss depth local average [mm];maintaining local distinct characteristics of the corrosionprocess severity. There are three procedures that must be • Et: tool measurement error [mm];performed. Firstly, ILI metal-loss data set should be submitted • Fsi: defect scoring factor for corrosion activity at defectto a filtering/adjustment process, taking into consideration the vicinity;reporting threshold consistency; the possible existence of • Hi: defect odometer [m];systematic bias and corrosion mechanisms similarity. Secondly, • N: Population defects total number;the average metal-loss growth rate between inspections should • N1: Population 1 defects total number;be determined based on the filtered populations. Thirdly, the • N2: Population 2 defects total number;defects reported by the latest inspection should have their • n: vicinity parameter;corrosion growth rates individually determined as a function of • Rrc: corrosion growth rate determined by runthe mean depth values of the whole population and in the defect comparison [mm/year].neighborhood. The methodology allows quick and realisticdamage-progression estimates, endeavoring to achieve morecost-effective and reliable strategies for the integritymanagement of aged corroded systems. Model robustness andgeneral feasibility is demonstrated in a real case study. 1 Copyright © 2010 by ASME
  • 2. INTRODUCTION states that a metal-loss defect could have its growth rate determined as a function of the average metal-loss damage in In the past few decades, many of the most catastrophic the adjoined region.pipeline industry accidents have had as their root failure themechanism diagnosed as corrosion. Presently, operators are The concept of adjoined region or neighborhood (Lni)able to rely upon several codes, standards and consolidated is expressed by the vicinity parameter (n) and is characteristicpractices to assess the remaining strength in quantified to each location of each pipeline, being dependent on the defectcorrosion metal-loss areas, ranging from the most simple and population overall number as well as its local distribution. Thestraightforward, such as ASME B-31G, to those more refined vicinity parameter of a pipeline defect population should beand time demanding, such as Finite Element analysis. determined in order to comply with the requirements expressedAdditionally, the market is now offering in-line inspection (ILI) by Equations (1) and (2) below:technologies able to map and quantify, with reasonableaccuracy, metal-loss damage distribution along pipelines, and n >= 3 (1)the choice has grown considerably. i = N −n However, the definition of a proper pipeline integrity ∑ i = n +1 ( H i + n − H i −n ) 0.1 ≤ ≤ 10 (2)management strategy also requires damage progression N − 2(n − 1)forecasting. Indeed, growth rate estimation plays a fundamentalrole in the optimization of the re-inspection intervals andmaintenance rehabilitation scope, and therefore has a directeffect on pipeline maintenance cost-effectiveness and Simplistic assumptions1operational reliability and safety. • linear damage progression within the considered time intervals (between inspections and forecast period); The data-set comparison of subsequent ILI could • all external corrosion active sites are associated with coatprovide solid ground for such estimates. Typically, corrosion damage areas where the coat protection effectiveness isrates are estimated by individual defect matching or by assumed to be null;population comparison. The results achieved by the former • new defect generation as well as stop growth rate is notcould be quite accurate, especially when raw signal evaluation considered.takes place, but the process applicability is rather limited, due toits characteristic of being time-demanding, while also requiringspecial computational tools and skilled people. On the other CONSTRUCTING REPRESENTATIVE POPULATIONShand, ordinary procedures to perform run comparison by takinginto account entire populations could be quickly carried out, but Regarding the construction of the populations to beusually give rise to large errors in the resulting estimate. assessed, special consideration should be given to tool measurement error significance, in order to guarantee relative This paper aims at presenting a straightforward scoring significance of the ILI reported threshold. As a general rule, it ismethodology which considerably enhances corrosion rate recommendable that depth values which do not comply withoutputs from ILI population comparison. The work was Equation (3) below should be dismissed.developed based on the Principle of Local Corrosion Activity[1, 2] to take into account the neighborhood metal-loss 2 Et (3)characteristics of each defect individually. Methodology di ≥formulation could easily be put into practice on any commercial 1.28spreadsheet with basic statistic functions. A case study ispresented in order to demonstrate that, despite its simplicity, the Also, data population to be compared must have beenmodel’s results are quite robust and realistic. collected by ILI tools of similar performance (technology, accuracy, peculiarities related to each run such as, line cleaningMODEL BACKGROUND and flow regime) and they should both have been properly validated by field measurements. Possible bias should be The difficulty in accurately measuring and predicting identified and corrected. Depending on the adoptedtime dependent behavior of the parameters, which control the segmentation strategy, quality of data alignment could have acorrosion process in large structures such as pipelines, make it significant impact on model outputs.hard to accomplish effective forecasting of the metal-lossprogression rates under operational conditions, by means ofmechanistic approaches. A brief outline of segmentation strategy: the only mandatory segmentation procedure is the obvious partition of defects Stochastic modeling based on ILI data is better fittedto accommodate the inherent randomness of corrosion in 1pipelines. Accordingly, the principle of local corrosion activity Further details in references [1] and [2]. 2 Copyright © 2010 by ASME
  • 3. located on the outside of the pipeline from those on the inside. Moreover, population segmentation could be also especiallyCommon segmentation based on axial distance is not recommendable when coating material changes are involvedparticularly required, given the account of local corrosion (different degradation lags to be considered).activity. But, when it is previously known that differentmechanisms are acting concomitantly and their attack severity isexpected to be highly dissimilar, further populationsegmentation could be attempted, especially when those couldgive rise to distinct circumferential distributions. Inspection 1 Inspection 2 Reported Data Reported Data Filtering 2 Et di ≥ 1.28 Assessment of Population Significance Correction of Measurement Bias Data Alignment Segmentation Strategy Population 1 Population 2 d 1 d2 d 2 − d1 Rrc = ∆ti Figure 1: Flowchart of the primary run comparison. 3 Copyright © 2010 by ASME
  • 4. 2. MATHEMATICAL FRAMEWORK1. Two sets of ILI reported depth values; named Population 1 Each individual depth measurement for Population 2 should beand Population 2 (referred as former and latter inspections, associated to a characteristic local depth Gaussian distribution,respectively), should be “fitted” for analysis. The “fitting” as defined by Equation [5a] and [5b]. (∑ )procedure should incorporate data consistency checking, j =i + nmeasurement bias assessment and correction, besides possible dj j =i − n (5a)segmentation strategy (which could require data alignment). d li =The average corrosion growth in the time interval between (2n + 1)inspections ( ∆ti) should be determined as the ratio of thedifference between the population’s arithmetic mean and ∆ti asexpressed by Equation (4). σ li = (∑ j =i + n j =i − n (d j − d li ) 2 (5b) d 2 − d1 (n − 1) Rrc = (4) ∆ti Population 2 = (∑ j =i + n j =i − n dj ) (∑ j =i + n j =i − n (d j − d li ) 2 dli σ li = (2n + 1) (n − 1) Y d2 − di d i ≥ d Li + 1.75σ Li Fsi = 1 + d2 N d 2 − d li Fsi = 1 + d2 d fi = d i + Fsi Rrc ∆t f Forecasting of Population 2 evolving for ∆t f years FIGURE 2: Flowchart of the process to forecast defects future population. 4 Copyright © 2010 by ASME
  • 5. 3. Subsequently, a corrosion activity coefficient (Fsi) should alsobe individually calculated by Equation (6a), in order to express Table 1 gives some additional details regarding the internalthe region’s corrosion activity potential. defect populations reported in the 2007 and 2009 inspections, as well as their respective arithmetic means. Also, Figure 3 d 2 − d li presents the axial distribution of defects with reported depths Fsi = 1 + (6a) d2 over 40%. d 2 − di Procedure: as a consequence of the considerable difference Fsi = 1 + (6b) between inspection reporting thresholds, the population of d2 internal defects reported by the 2009 inspection (N2) was almost a third of the 2007 (N1) inspection. In order to deal with this3a. Hot Spot Conditional Structure. The underestimation in hot problem the following procedure was adopted:spot regions should be prevented by modifying equation (6a)into (6b), when the conditional structure stated in 1. As a starting point, it was assumed that the N2 deepestEquation (7) is fulfilled: defects reported by the 2007 inspection were the same N2 reported in 2009, those defects being labeled as Population d i ≥ d li + 1.75σ Li 1’; (7) 2. A first approximation of the average corrosion rate in the4. Finally, future defect depth should be determined by period between the considered inspections (R’rc) wasEquation (8). determined by Equation (4), considering the Populations 2 and 1’; d fi = d i + Fsi Rrc ∆t f (8) 3. All defects originally reported by the 2007 inspection (Population 1o) have had their future geometry forecastA broad outline of run comparison logic is depicted in the according to the procedure detailed in the flowchart offlowchart presented in Figure 2. Figure 2; 4. Population 1 was then defined as being the N2 defects which presented the deepest forecast depths for 2009;CASE STUDY 5. A new Rrc value was calculated by Eq. (4), considering Populations 1 and 2;Background: The methodology described was applied to 6. Finally, Population 2 growth was forecast for a timeestimate the corrosion rate distribution in a 3.5 km segment of interval of 3 years, resulting in the data set, labeled asan oil production line with 16” diameter, 9.5 mm thickness, Population 3.manufactured in API 5L X65. The pipeline had been used in theinterconnection of onshore production fields of depleted Results: Figure 4 compares the defect depth histograms ofreservoirs and the BSW content of the fluid conveyed ranged 2007 and 2009 ILI measurements (Populations 1 and 2,from 80-90%. In late 2007 and late 2009, the line underwent respectively) with 2012 predicted population (Population 3).MFL ILI surveys. Both tool accuracy and performance were The steadiness of damage progression becomes much morefairly similar but the reporting threshold was 10% in 2007, evident when those populations are normalized by the Localwhile in the 2009 inspection, only depths above 35% were Activity Principle as could be noted in Figure 5.reported. No attempt to identify and correct measurement bias Model Robustness: In order to assess the model output fitness,has been carried out. Only the anomalies located in the internal Population 1 (2007) evolution was projected towards 2009. Thepipeline surface have been considered in the analysis. histogram of these predicted depths is compared with the histogram of 2009 ILI reported results in Figure 6. The matchTABLE 1: Defect depth populations. between them demonstrates the model robustness and its suitability for common pipeline fitness-for-purpose evaluations. POPULATION THRESHOLD INSPECTION NUMBER OF AVERAGE DEFECTS DEPTH The slight tendency to overestimate the frequencies of higher % wt % wt depths can almost be overcome by dismissing the hot spot conditional structure in the algorithm. 1 12,833 32% 38.112007 1o 33,833 10% ---2009 2 12,833 35% 40.29 5 Copyright © 2010 by ASME
  • 6. 70 65defect depth, wt% 60 55 50 45 40 35 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 axial distance, km FIGURE 3: Defects axial distribution in 2009. 3500 2007 ILI 3000 2009 ILI 2500 2012 PROJECTION defects number 2000 1500 1000 500 0 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 defect depth, w t% FIGURE 4: Histogram of defect depth populations. 4500 2007 ILI 2009 ILI 2012 PROJECTION 4000 3500defects number 3000 2500 2000 1500 1000 500 0 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 local defect depth, wt% FIGURE 5: Histograms of local defect depth. 6 Copyright © 2010 by ASME
  • 7. 3500 3000 measured projected 2500 defects number 2000 1500 1000 500 0 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 defect depth, w t% FIGURE 6: 2009 Histogram of defects depth. 3. Nicoletti, E.S.M.; de Souza, R. D.A; Olivier, J.H.L.CONCLUSION External Corrosion Growth on an Ageing Pipeline: a Case Study. INTERCORR 2010. Fortaleza, 2010.A stochastic scoring methodology to estimate corrosion growthby ILI run comparison is presented. The model could easily be 4. Nicoletti, E.S.M.; de Souza, R. D.A; Olivier, J.H.L. Metalimplemented in any mathematic commercial package and loss Growth Rate Distributions: A Case Study on Ageingallows quick and reasonably accurate projections of the defect- Transmission System. Rio Pipeline Conference 2009. Riodepth population evolution throughout time. de Janeiro, 2010.Output reliability, however, is highly dependent on dataconsistency as much as the significance of the comparisonprocedure, and so the differences in the tool technologies,performances and reporting thresholds should be carefullyevaluated and assessed. Also, it is assumed that significant toolmeasurement bias had been previously identified and properlytreated.ACKNOWLEDGMENTSThe authors would like to thank PETROBRAS TransporteS.A. for permission to publish this paper, and their colleaguesDr. Sérgio Cunha and João Hipólito de Lima Oliver for manycontributions and enlightening discussions.REFERENCES1. Nicoletti, E.S.M.; de Souza, R. D. A Practical Approach in Pipeline Corrosion Modeling: Part 1 – Long Term Integrity Forecasting. Journal of Pipeline Engineering. Vol. 8, no. 1. March 2009.2. Nicoletti, E.S.M.; de Souza, R. D. A Practical Approach in Pipeline Corrosion Modeling: Part 2 – Short Term Integrity Forecasting. Journal of Pipeline Engineering. Vol. 8, no. 2. June 2009. (Correction Note Published in Vol. 8, no. 3. June 2010). 7 Copyright © 2010 by ASME

×