proposed statistical model for corrosion failure prediction and l
TRANSCRIPT
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 121
University of Akron Ohiorsquos Polytechnic University
IdeaExchangeUAkron
H3 Ramp3amp$ P+amp$43ltamp D G9 B Pamp- S W--3 H3
C--ampamp
S 2015
Proposed Statistical Model for Corrosion FailurePrediction and Lifetime Extension of Pipelines
Jessica Y Ripple +923amp
F--7 43 4- 73 4 =ampamp8$ampamp3amp3amp$+amp$43
lt3 H3 Ramp3amp$ P+amp$4 3 4 4 9 ampamp amp $$amp33 9 4amp ltamp D G9 B Pamp- S W--3 H3 C--ampamp 4
IampE8$ampUA I4 3 ampamp $$amp4amp $-3 H3 Ramp3amp$ P+amp$43 9 4amp 344 IampE8$ampUA
F amp 4 -amp3amp $4$4 +amp ltamp U6amp349 A 3 O3 P-94amp$$ U6amp349 (=777amp)
Ramp$ampamp C44R-amp Jamp33$ P3amp S4434$- Mamp- C3 F-amp Pamp$4 Lamp4amp E84amp3 Pamp-amp3 (2015) Honors
Research Projects Pamp 101
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 221
983124983144983141 983125983150983145983158983141983154983155983145983156983161 983151983142 983105983147983154983151983150
983112983151983150983151983154983155 983107983151983148983148983141983143983141 983122983141983155983141983137983154983139983144 983120983154983151983146983141983139983156
PROPOSED STATISTICAL
MODEL FOR CORROSIONFAILURE PREDICTION AND
LIFETIME EXTENSION OF
PIPELINES
983114983141983155983155983145983139983137 983122983145983152983152983148983141
983105983152983154983145983148 9830901 98309001983093
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 321
Ripple 1
Table of ContentsExecutive Summary 2
Introduction 3
Background 5
General Corrosion 5
Oil Pipeline Corrosion 5
Corrosion Models 6
Model Discussion 8
Data Input 8
Monte Carlo Simulation 9
Economic Analysis 13
Recommendations and Conclusions 17
Recommendations 17
Conclusions 17
References 18
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 421
Ripple 2
Executive Summary
Corrosion is an unavoidable global issue that has serious safety and environmentalconsequences if left unmitigated Currently the majority of methods used to address corrosionare reactive in nature meaning that industries are responding to leaks spills and catastrophes
after they have already occurred Ideally the asset owners of the corroding infrastructure andequipment should use predictive modeling to take proactive measures before a corrosion inducedrisk becomes a danger to society
This project focuses primarily on internal pipeline corrosion The US has hundreds ofthousands of miles worth of pipelines currently in operation These pipelines are oftentransporting hazardous materials such as natural gas and crude oil To protect society and theenvironment from loss of containment the US Department of Transportation requires mostpipelines to be analyzed for internal corrosion every five years using an in-line inspection (ILI)tool
The intended audience for this model are pipeline asset owners and operators The goal ofthis project was to develop a ldquouser-friendlyrdquo model that allows a user to analyze the datacollected from the ILI tool The user can choose to accept the default analysis options or easilymake adjustments to tailor the results to their data
This report presents the results from developing a predictive model that estimates theprobability of failure of a pipeline The model utilizes data collected with an in-line inspectiontool to generate a Monte Carlo simulation based on a desired statistical distribution The resultsof the Monte Carlo simulation provide the user with the probability of failure that each anomalyon the pipeline presents over a specific time period Using these probabilities the modelconducts an economic analysis to recommend an optimal repair schedule
Going forward this model can be improved by adding new distributions which willallow the data to be categorized more accurately and new predictive equations which permits theuser to adjust how conservative the results should be Also the model would greatly benefit froma second set of ILI data of the same pipeline to validate the calculated results
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 521
Ripple 3
Introduction
The target audience for this research project is oil and gas pipeline operators owners orother responsible parties The final product a computer model utilizes raw industry dataaccepted industry standards statistical analysis and economical evaluations to recommend
optimal mitigation methods and scheduling As with any model the results are only as accurateas the data and engineering judgement that is input
Figure 1 shows the main process steps the model utilizes The raw data comes from in-line inspection (ILI) tools also known as ldquosmart pigsrdquo and consists of wall thickness andanomaly width and depth An anomaly is considered to be any defect in the pipe internal surfaceSmart pigs have the ability to catalog hundreds of kilometers worth of pipeline data Within themodel this data is then fit to an appropriate distribution curve which can be chosen based onminimizing standard error The next step entails performing a Monte Carlo simulation togenerate probabilistic pipeline conditions These conditions are individually assessed whichprovides a wide range of operating pressure estimates Once the pipeline operation has been
simulated over a specified range of time an economic analysis can be performed which willmake recommendations based on specified costs of failure and repair The final output of themodel includes a recommended year of repair at each specific anomaly location and theeconomic value of risk avoided by performing said repair
Every step is completed independently and requires various amounts of user inputs Theoverarching idea for this model is to allow the user to be a detailed or simplistic as he or sheprefers The model has default settings for the steps but has the flexibility for tailoring at anypoint
Figure 1 Pipeline reliability model execution steps
983093 983109983139983151983150983151983149983145983139
983105983150983137983148983161983155983145983155
983092 983119983152983141983154983137983156983145983150983143
983120983154983141983155983155983157983154983141
983109983155983156983145983149983137983156983145983151983150
983091 983117983151983150983156983141
983107983137983154983148983151
983123983145983149983157983148983137983156983145983151983150
983090 983108983137983156983137
983108983145983155983156983154983145983138983157983156983145983151983150
983110983145983156983156983145983150983143
1 983122983137983159 983108983137983156983137
983113983150983152983157983156
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 221
983124983144983141 983125983150983145983158983141983154983155983145983156983161 983151983142 983105983147983154983151983150
983112983151983150983151983154983155 983107983151983148983148983141983143983141 983122983141983155983141983137983154983139983144 983120983154983151983146983141983139983156
PROPOSED STATISTICAL
MODEL FOR CORROSIONFAILURE PREDICTION AND
LIFETIME EXTENSION OF
PIPELINES
983114983141983155983155983145983139983137 983122983145983152983152983148983141
983105983152983154983145983148 9830901 98309001983093
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 321
Ripple 1
Table of ContentsExecutive Summary 2
Introduction 3
Background 5
General Corrosion 5
Oil Pipeline Corrosion 5
Corrosion Models 6
Model Discussion 8
Data Input 8
Monte Carlo Simulation 9
Economic Analysis 13
Recommendations and Conclusions 17
Recommendations 17
Conclusions 17
References 18
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 421
Ripple 2
Executive Summary
Corrosion is an unavoidable global issue that has serious safety and environmentalconsequences if left unmitigated Currently the majority of methods used to address corrosionare reactive in nature meaning that industries are responding to leaks spills and catastrophes
after they have already occurred Ideally the asset owners of the corroding infrastructure andequipment should use predictive modeling to take proactive measures before a corrosion inducedrisk becomes a danger to society
This project focuses primarily on internal pipeline corrosion The US has hundreds ofthousands of miles worth of pipelines currently in operation These pipelines are oftentransporting hazardous materials such as natural gas and crude oil To protect society and theenvironment from loss of containment the US Department of Transportation requires mostpipelines to be analyzed for internal corrosion every five years using an in-line inspection (ILI)tool
The intended audience for this model are pipeline asset owners and operators The goal ofthis project was to develop a ldquouser-friendlyrdquo model that allows a user to analyze the datacollected from the ILI tool The user can choose to accept the default analysis options or easilymake adjustments to tailor the results to their data
This report presents the results from developing a predictive model that estimates theprobability of failure of a pipeline The model utilizes data collected with an in-line inspectiontool to generate a Monte Carlo simulation based on a desired statistical distribution The resultsof the Monte Carlo simulation provide the user with the probability of failure that each anomalyon the pipeline presents over a specific time period Using these probabilities the modelconducts an economic analysis to recommend an optimal repair schedule
Going forward this model can be improved by adding new distributions which willallow the data to be categorized more accurately and new predictive equations which permits theuser to adjust how conservative the results should be Also the model would greatly benefit froma second set of ILI data of the same pipeline to validate the calculated results
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 521
Ripple 3
Introduction
The target audience for this research project is oil and gas pipeline operators owners orother responsible parties The final product a computer model utilizes raw industry dataaccepted industry standards statistical analysis and economical evaluations to recommend
optimal mitigation methods and scheduling As with any model the results are only as accurateas the data and engineering judgement that is input
Figure 1 shows the main process steps the model utilizes The raw data comes from in-line inspection (ILI) tools also known as ldquosmart pigsrdquo and consists of wall thickness andanomaly width and depth An anomaly is considered to be any defect in the pipe internal surfaceSmart pigs have the ability to catalog hundreds of kilometers worth of pipeline data Within themodel this data is then fit to an appropriate distribution curve which can be chosen based onminimizing standard error The next step entails performing a Monte Carlo simulation togenerate probabilistic pipeline conditions These conditions are individually assessed whichprovides a wide range of operating pressure estimates Once the pipeline operation has been
simulated over a specified range of time an economic analysis can be performed which willmake recommendations based on specified costs of failure and repair The final output of themodel includes a recommended year of repair at each specific anomaly location and theeconomic value of risk avoided by performing said repair
Every step is completed independently and requires various amounts of user inputs Theoverarching idea for this model is to allow the user to be a detailed or simplistic as he or sheprefers The model has default settings for the steps but has the flexibility for tailoring at anypoint
Figure 1 Pipeline reliability model execution steps
983093 983109983139983151983150983151983149983145983139
983105983150983137983148983161983155983145983155
983092 983119983152983141983154983137983156983145983150983143
983120983154983141983155983155983157983154983141
983109983155983156983145983149983137983156983145983151983150
983091 983117983151983150983156983141
983107983137983154983148983151
983123983145983149983157983148983137983156983145983151983150
983090 983108983137983156983137
983108983145983155983156983154983145983138983157983156983145983151983150
983110983145983156983156983145983150983143
1 983122983137983159 983108983137983156983137
983113983150983152983157983156
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 321
Ripple 1
Table of ContentsExecutive Summary 2
Introduction 3
Background 5
General Corrosion 5
Oil Pipeline Corrosion 5
Corrosion Models 6
Model Discussion 8
Data Input 8
Monte Carlo Simulation 9
Economic Analysis 13
Recommendations and Conclusions 17
Recommendations 17
Conclusions 17
References 18
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 421
Ripple 2
Executive Summary
Corrosion is an unavoidable global issue that has serious safety and environmentalconsequences if left unmitigated Currently the majority of methods used to address corrosionare reactive in nature meaning that industries are responding to leaks spills and catastrophes
after they have already occurred Ideally the asset owners of the corroding infrastructure andequipment should use predictive modeling to take proactive measures before a corrosion inducedrisk becomes a danger to society
This project focuses primarily on internal pipeline corrosion The US has hundreds ofthousands of miles worth of pipelines currently in operation These pipelines are oftentransporting hazardous materials such as natural gas and crude oil To protect society and theenvironment from loss of containment the US Department of Transportation requires mostpipelines to be analyzed for internal corrosion every five years using an in-line inspection (ILI)tool
The intended audience for this model are pipeline asset owners and operators The goal ofthis project was to develop a ldquouser-friendlyrdquo model that allows a user to analyze the datacollected from the ILI tool The user can choose to accept the default analysis options or easilymake adjustments to tailor the results to their data
This report presents the results from developing a predictive model that estimates theprobability of failure of a pipeline The model utilizes data collected with an in-line inspectiontool to generate a Monte Carlo simulation based on a desired statistical distribution The resultsof the Monte Carlo simulation provide the user with the probability of failure that each anomalyon the pipeline presents over a specific time period Using these probabilities the modelconducts an economic analysis to recommend an optimal repair schedule
Going forward this model can be improved by adding new distributions which willallow the data to be categorized more accurately and new predictive equations which permits theuser to adjust how conservative the results should be Also the model would greatly benefit froma second set of ILI data of the same pipeline to validate the calculated results
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 521
Ripple 3
Introduction
The target audience for this research project is oil and gas pipeline operators owners orother responsible parties The final product a computer model utilizes raw industry dataaccepted industry standards statistical analysis and economical evaluations to recommend
optimal mitigation methods and scheduling As with any model the results are only as accurateas the data and engineering judgement that is input
Figure 1 shows the main process steps the model utilizes The raw data comes from in-line inspection (ILI) tools also known as ldquosmart pigsrdquo and consists of wall thickness andanomaly width and depth An anomaly is considered to be any defect in the pipe internal surfaceSmart pigs have the ability to catalog hundreds of kilometers worth of pipeline data Within themodel this data is then fit to an appropriate distribution curve which can be chosen based onminimizing standard error The next step entails performing a Monte Carlo simulation togenerate probabilistic pipeline conditions These conditions are individually assessed whichprovides a wide range of operating pressure estimates Once the pipeline operation has been
simulated over a specified range of time an economic analysis can be performed which willmake recommendations based on specified costs of failure and repair The final output of themodel includes a recommended year of repair at each specific anomaly location and theeconomic value of risk avoided by performing said repair
Every step is completed independently and requires various amounts of user inputs Theoverarching idea for this model is to allow the user to be a detailed or simplistic as he or sheprefers The model has default settings for the steps but has the flexibility for tailoring at anypoint
Figure 1 Pipeline reliability model execution steps
983093 983109983139983151983150983151983149983145983139
983105983150983137983148983161983155983145983155
983092 983119983152983141983154983137983156983145983150983143
983120983154983141983155983155983157983154983141
983109983155983156983145983149983137983156983145983151983150
983091 983117983151983150983156983141
983107983137983154983148983151
983123983145983149983157983148983137983156983145983151983150
983090 983108983137983156983137
983108983145983155983156983154983145983138983157983156983145983151983150
983110983145983156983156983145983150983143
1 983122983137983159 983108983137983156983137
983113983150983152983157983156
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 421
Ripple 2
Executive Summary
Corrosion is an unavoidable global issue that has serious safety and environmentalconsequences if left unmitigated Currently the majority of methods used to address corrosionare reactive in nature meaning that industries are responding to leaks spills and catastrophes
after they have already occurred Ideally the asset owners of the corroding infrastructure andequipment should use predictive modeling to take proactive measures before a corrosion inducedrisk becomes a danger to society
This project focuses primarily on internal pipeline corrosion The US has hundreds ofthousands of miles worth of pipelines currently in operation These pipelines are oftentransporting hazardous materials such as natural gas and crude oil To protect society and theenvironment from loss of containment the US Department of Transportation requires mostpipelines to be analyzed for internal corrosion every five years using an in-line inspection (ILI)tool
The intended audience for this model are pipeline asset owners and operators The goal ofthis project was to develop a ldquouser-friendlyrdquo model that allows a user to analyze the datacollected from the ILI tool The user can choose to accept the default analysis options or easilymake adjustments to tailor the results to their data
This report presents the results from developing a predictive model that estimates theprobability of failure of a pipeline The model utilizes data collected with an in-line inspectiontool to generate a Monte Carlo simulation based on a desired statistical distribution The resultsof the Monte Carlo simulation provide the user with the probability of failure that each anomalyon the pipeline presents over a specific time period Using these probabilities the modelconducts an economic analysis to recommend an optimal repair schedule
Going forward this model can be improved by adding new distributions which willallow the data to be categorized more accurately and new predictive equations which permits theuser to adjust how conservative the results should be Also the model would greatly benefit froma second set of ILI data of the same pipeline to validate the calculated results
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 521
Ripple 3
Introduction
The target audience for this research project is oil and gas pipeline operators owners orother responsible parties The final product a computer model utilizes raw industry dataaccepted industry standards statistical analysis and economical evaluations to recommend
optimal mitigation methods and scheduling As with any model the results are only as accurateas the data and engineering judgement that is input
Figure 1 shows the main process steps the model utilizes The raw data comes from in-line inspection (ILI) tools also known as ldquosmart pigsrdquo and consists of wall thickness andanomaly width and depth An anomaly is considered to be any defect in the pipe internal surfaceSmart pigs have the ability to catalog hundreds of kilometers worth of pipeline data Within themodel this data is then fit to an appropriate distribution curve which can be chosen based onminimizing standard error The next step entails performing a Monte Carlo simulation togenerate probabilistic pipeline conditions These conditions are individually assessed whichprovides a wide range of operating pressure estimates Once the pipeline operation has been
simulated over a specified range of time an economic analysis can be performed which willmake recommendations based on specified costs of failure and repair The final output of themodel includes a recommended year of repair at each specific anomaly location and theeconomic value of risk avoided by performing said repair
Every step is completed independently and requires various amounts of user inputs Theoverarching idea for this model is to allow the user to be a detailed or simplistic as he or sheprefers The model has default settings for the steps but has the flexibility for tailoring at anypoint
Figure 1 Pipeline reliability model execution steps
983093 983109983139983151983150983151983149983145983139
983105983150983137983148983161983155983145983155
983092 983119983152983141983154983137983156983145983150983143
983120983154983141983155983155983157983154983141
983109983155983156983145983149983137983156983145983151983150
983091 983117983151983150983156983141
983107983137983154983148983151
983123983145983149983157983148983137983156983145983151983150
983090 983108983137983156983137
983108983145983155983156983154983145983138983157983156983145983151983150
983110983145983156983156983145983150983143
1 983122983137983159 983108983137983156983137
983113983150983152983157983156
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 521
Ripple 3
Introduction
The target audience for this research project is oil and gas pipeline operators owners orother responsible parties The final product a computer model utilizes raw industry dataaccepted industry standards statistical analysis and economical evaluations to recommend
optimal mitigation methods and scheduling As with any model the results are only as accurateas the data and engineering judgement that is input
Figure 1 shows the main process steps the model utilizes The raw data comes from in-line inspection (ILI) tools also known as ldquosmart pigsrdquo and consists of wall thickness andanomaly width and depth An anomaly is considered to be any defect in the pipe internal surfaceSmart pigs have the ability to catalog hundreds of kilometers worth of pipeline data Within themodel this data is then fit to an appropriate distribution curve which can be chosen based onminimizing standard error The next step entails performing a Monte Carlo simulation togenerate probabilistic pipeline conditions These conditions are individually assessed whichprovides a wide range of operating pressure estimates Once the pipeline operation has been
simulated over a specified range of time an economic analysis can be performed which willmake recommendations based on specified costs of failure and repair The final output of themodel includes a recommended year of repair at each specific anomaly location and theeconomic value of risk avoided by performing said repair
Every step is completed independently and requires various amounts of user inputs Theoverarching idea for this model is to allow the user to be a detailed or simplistic as he or sheprefers The model has default settings for the steps but has the flexibility for tailoring at anypoint
Figure 1 Pipeline reliability model execution steps
983093 983109983139983151983150983151983149983145983139
983105983150983137983148983161983155983145983155
983092 983119983152983141983154983137983156983145983150983143
983120983154983141983155983155983157983154983141
983109983155983156983145983149983137983156983145983151983150
983091 983117983151983150983156983141
983107983137983154983148983151
983123983145983149983157983148983137983156983145983151983150
983090 983108983137983156983137
983108983145983155983156983154983145983138983157983156983145983151983150
983110983145983156983156983145983150983143
1 983122983137983159 983108983137983156983137
983113983150983152983157983156
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 621
Ripple 4
Currently the main limitations of the model include the inability to distinguish betweeninternal corrosion and erosion as well as the inability to evaluate external corrosion Also themodel does not differentiate a failure as either a leak or a rupture Lastly the model assumes aconstant corrosion rate with time It cannot predict a sudden acceleration in corrosion Howeverfrequent and thorough data collection should be able to give an indication of a step change in the
corrosion rateThe data presented in this report is not meant to be extrapolated to represent other
pipelines The purpose of displaying and analyzing the data in this report is to showcase thetypes of analyses that the model can perform
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 721
Ripple 5
Background
General Corrosion
Corrosion is simply defined as the ldquodegradation of a material due to a reaction with its
environmentrdquo (1) A wide variety of materials can fall victim to corrosion including metalspolymers and ceramics Additionally these materials can undergo various forms of corrosionmost often simultaneously (2) Figure 2 displays the corrosion mechanism using a basicelectrochemical cell schematic (2) Corrosion occurs due to the oxidation reaction taking place atthe anode The loss of electrons at the anode is balanced by the reduction reaction occurring atthe cathode Depending on the chemical concentration and properties of the environment (shownas the electrolyte in Figure 2) the corrosion rate may be almost negligible (less than 1 mpy) orvery severe (greater than 200 mpy) (3)
Figure 2 Corrosion representation using electrochemical cell
Unmitigated corrosion has significant safety environmental and economicconsequences on a global scale In 2002 a comprehensive study determined the total direct costsof corrosion in a wide variety of industry sectors in the United States totaled $276 billion (4) Tocombat corrosion several mitigation strategies are available The most common include materialselection coatings inhibitors and cathodic protection (5)
Oil Pipeline Corrosion
For this specific project the primary focus was of internal corrosion of oil pipelines in theUnited States Crude oil itself is not corrosive Instead the corrosion is caused by carbon dioxide(CO2) hydrogen sulfide (H2S) and especially water (6 ) At high velocities the water will remainentrained in the crude oil and will not be able to settle out and cause corrosion (7 ) However atlower velocities there is a greater risk of two phase flow and thus corrosion can occur in thewater phase Light crude oils containing water are at a higher risk due to the greater difference in
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 821
Ripple 6
density to water (7 ) Therefore as refineries are forced to run heavier crudes over time the crudeoil pipelines will actually decrease their risk of corrosion
Today there are limits in place on water concentration in transporting crude oil inpipelines In 2012 the length of US oil pipelines was approximately 152000 miles (8 ) Forabout 90 of hazardous liquid pipelines pipeline operators have the option of using a ldquosmart
pigrdquo to collect integrity data (9) Figure 3 illustrates a smart pig created by Nord Stream whichuses magnetic flux technology which is one of the most common methods (10) These devicesare capable of detecting the location of pipeline defects such as wall thinning mechanicaldamage material defects and cracks (9) The data often referred to as ldquoin-line inspectionrdquo or ILIdata was the primary input to the developed model The United States Department ofTransportation requires pipeline owners to complete a smart pig run every five years although itis common for them to be performed even more frequently (11)
Figure 3 Nord Stream magnetic flux smart pig
Corrosion Models
Currently in industry it is common to use one of the previously developed deterministicmodels to estimate the service lifetime before corrosion induced pipeline failure Figure 4 compares the pipeline failure pressure as a function of defect length using different models (12)The models all require similar information (anomaly length and depth pipe thickness etc) tocalculate the failure pressure For the purpose of this model the Modified B31G was used sinceit is a standard produced by the American Society of Mechanical Engineers (ASME) and is lessconservative than the original B31G model (13) The ASME Modified B31G method wasprimarily developed on older lower strength steels while some of the ldquonewerrdquo models including
the DNV-99 method were developed based on modern pipeline materials (14)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 921
Ripple 7
Figure 4 Pipeline failure pressure vs defect length for various models
A Level 1 Modified B31G evaluation was conducted based on the steps outlined in theASME Technical Standard document (13) Equation 1 shows the formula used to calculate thefailure pressure for the model where YS is yield strength t is the pipe thickness d(T) is theanomaly depth at time T L(T) is the anomaly length at time T and D is pipe diameter (12) Thisequation assumes a ldquoflow stressrdquo expression appropriate for a material with specified minimumyield strengths below 483 MPa and operating temperatures below 120degC (13)
Equation 1
1 06275
0003375
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1021
Ripple 8
Model Discussion
Data Input
The ILI data used to develop this model spanned 100 kilometers of industrial pipeline
The source of this pipeline data cannot be identified due to confidentiality although it wasprovided to the University of Akron with the intention of creating a predictive model It is worthnoting that only the raw data points were provided Any subsequent analyses statisticalmodeling or calculations were completed as part of this research project
Within the 100 kilometers of data over 3400 anomalies were documented Figure 5 displays the frequency of the data utilized in the model This analysis was carried out Potentialoutliers were not removed because these sites likely represent spots of local acceleratedcorrosion which was considered important to keep in the model Corrosion rates were determinedbased on the assumption that the pipe had been corroding for 10 years Ideally data from twoseparate ILI audits would be used to estimate corrosion rates more accurately
Figure 5 Pipeline ILI data histograms from a private industrial source
Figure 6 shows the probability density function (PDF) curves that were necessary forrunning the model The corrosion rate is shown on the left while the longitudinal growth rate isshown on the right These distribution curves are required for performing the Monte Carlosimulation Figure 7 shows an example of how the distribution curves were chosen for thelongitudinal growth rate The fit that yielded the lowest standard error was selected to representthe data shape Therefore the corrosion rate was fit to a normal distribution curve while thelongitudinal growth rate was fit to a lognormal distribution curve
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1121
Figure 6 Probability density fu
Figure 7 Distribution curve fitti
Monte Carlo Simulation
Once the raw data is enteevaluated the probability of failMonte Carlo simulation (MCS)
ction curves used in the model
ng for longitudinal growth rate
red and the appropriate distribution curve parare over time for each anomaly location is calcurocess The Monte Carlo simulation is defined
Ripple 9
eters arelated via theas ldquoa numerical
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1221
Ripple 10
experimentation technique to obtain the statistics of the output variables of a system computationmodel given the statistics of the input variablesrdquo (15) Figure 8 outlines the individual stepsrequired to complete a MCS A random number generator was used in Step 1 This randomnumber was associated with a corrosion and anomaly longitudinal growth rate using the inversenormal and lognormal functions respectively A MCS simulation including 1000 trials was
performed for each specific anomaly location Therefore the rate of corrosion depth andlongitudinal growth were statistical inputs to the simulation while the anomaly length and depthwere not For each anomaly the cumulative distribution function (CDF) of the calculated failurepressure was plotted The CDF is defined by NIST as the ldquoprobability that the variable takes avalue less than or equal to xrdquo (16 ) For this model the ldquovariablerdquo is the pressure calculated usingEquation 1 and ldquoxrdquo is the pipeline operating pressure Therefore the failure probability is foundby plotting the CDF and finding the probability that corresponds to the pipeline operatingpressure as shown in Figure 10 The model was set up on the basis for 15 years prediction
Figure 8 Calculation steps of a MCS process
1 Choose a random number between
0 and 1
2 Calculate the desired variable usingoptimal distribution arameters thatcorrespond to the random number
3 Repeat the first two steps for manyrepetitions (n = 1000)
4 Plot the cumulative distributionfunction (CDF)
5 The intersection of the CDF curvewith a value of interest represents the
probability of occurrence
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1321
Figure 9 shows CDF equdata (17 ) (18 ) (19) (20) Whenrepresents a probability labeledlength etc) is then calculated byInverse distribution equations ar
and MATLAB The inverse distrand standard deviation) that are
Figure 9 Commonly used distri
Figure 10 shows the MCcorrosion The operating pressurapproximately 71 From this dparticular anomaly corroding to
983118983151983154983149983137983148
991266 19831342〖2991266 σ983155983156983137983150983140
983116983151983143983150983151983154983149983137983148 991266 1〗9831342〖991266 σ983155983156983137983150983140
983127983141983145983138983157983148983148
991266 1991266 α983155983139983137983148983141
ations for distributions that are commonly seena random number is selected in Step 1 of the Mp The desired variable (whether it be corrosionsolving for the x value that corresponds to a spcommonly available in software including Mi
ibution equations require the parameters (suchhown below
utions and their respective cumulative distribu
S output for a singular anomaly after 15 years oof the line is 56 MPa which intersects the C
ta the conclusion is that after 15 years the prohe point of failure is 71
radic2 int983135infin983134 983134〖〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
radic2 int9831350983134 983134〖983148983150 2 〗9831342 〗983154983140 983140983141983158983145983137983156983145983151983150 μ983137983158983141983154983137983143983141
983134〖 〗983134 983137983154983137983149983141983156983141983154 β983155983144983137983152983141 983152983137983154983137983149983141983156983141983154
Ripple 11
in corrosionCS process itrate anomalyecific value of pcrosoft Excel
s the average
ion functions
f predictedF curve at
bability of this
〗
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1421
Figure 10 Monte Carlo simulati
To draw conclusions aboused Figure 11 shows the rangeyear for the same anomaly consiand minimum pressures that werpressure begins to extend belowfailure probability as shown in t
Figure 11 Estimated pressure aat 82190 m from pipeline start
on for a single anomaly located at 82190 m fro
t the anomaly over time the average calculateof pressures calculated and the overall probabilered in Figure 10 The data bars correspond tocalculated in the MCS At year 4 the minimu
the pipeline operating pressure This results in she graph on the right
d probability of failure prediction for singular
Ripple 12
m pipeline start
pressure wasity of failure perthe maximum
calculatedome amount of
nomaly located
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1521
Ripple 13
Lastly Figure 12 illustrates how the individual anomaly analyses were combined toproduce a failure probability ldquosnapshotrdquo of the bulk pipeline after 15 years of predictedcorrosion The percentages represent the probability of failure at each specific location Whilethis image was created manually using MATLAB the intention is to have the simulationgenerate all images automatically
Figure 12 Failure probability of pipeline segment after 15 years
Economic Analysis
There are many ways of accounting for costs associated with corrosion and how toaccount for these savings when employing mitigation strategies Previous studies on costs wereevaluated from the following perspectives ldquothe cost to the economy of a nation and the cost ofselected corrosion control measuresrdquo (17 ) Figure 13 lists some of the costs typically associatedwith corrosion (17 ) For this project capital and design costs were not considered since the focusof this work is protecting existing assets Thus the economic analysis focused on the trade-offbetween control costs (ie repairing the anomaly using inhibitors) and associated costs (ie lossof containment due to pipeline failure pipeline capacity diminished)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1621
Figure 13 Various costs associa
Initially the base cost offailure was estimated as $1000analyzed separately In each yea
the product of the failure penaltythan the economic risk of failureOnce the model recommends a yreduced to zero For example Fi
anomaly and how this comparesis calculated where the ldquonrdquo subsin year 1 should be calculated usthe probability of failure increasanomaly as compared to repairin8
Equation 2
Figure 14 Cost analysis for sin
CapitalDesignCosts
bull Replacementof equipment
and buildingsbull Excess
capacity
bull Redundantequipment
Co
bull Maand
bull Cocon
ted with corrosion
repair was estimated as $10000 and the penalt00 To determine the optimal time for repair e the cost of repair was compared to the cost of
and the probability of failure If the cost of repin a given year it was recommended to do fix tear of repair the probability of failure of the sugure 14 shows the total risk for each given yearto the base cost of repair Equation 2 shows hocript refers to the beginning of year n For exaing the failure probability that corresponds to ys such that it becomes a greater economic riskg it Thus the recommendation is to repair the
le anomaly located at 805 m from pipeline start
trol Costs
ntenancerepair
rosiontrol
Design Costs
bull Materials ofconstruction
bull Corrosionallowance
bull Specialprocessing
bull Lpr
bull TS
bull In
bull In
Ripple 14
of a pipelinech year wasfailure which is
ir was lowerhe anomalysequent years isfor a specific
w the total riskple the total riskar 1 In year 8o not repair thenomaly at year
ssociatedCosts
ss ofoduct
chnicalpport
surance
ventory
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1721
Figure 15 shows the copipelines discounted back to therisk that is accrued as the probabaccumulated with repairs is the abefore the model recommends re
basis the total risk of operating tas adding inhibitors have the poamount
Equation 3 shows how tthe beginning of year n and the sconsidered from the beginning oEquation 3 was used as explicitlbut once the anomaly was repair
Equation 3
Figure 15 Discounted risk comtime
Lastly Figure 16 compaaccepted by not employing mitig
lower than the evaluated cost ofand thus the cost of mitigation sl12 showed it is very common torepair this segment of pipeline ssuch as excavating and labor coscorrosion is estimated at over $4the pipeline is predicted to avoid
parison of risk between repairing and not repaibeginning of year 0The risk of not repairing isility of failure increases annually The amountmount of proportion of the failure penalty that ipairing the anomaly By optimizing repairs on
he pipeline can be dramatically reduced Otherential to reduce the cost of risk by an even mor
e discounted total risk was calculated in Figur
ummation is over the entire length of the pipelithe year the first value of n is 2 For the case
y written For the case with repairs Equation
d the risk for subsequent years was set to 0
arison of repairing vs not repairing all pipeline
res the predicted annual cost of repairs vs the cation methods Excluding year 14 the cost of r
risk Year 14 had the largest number of recommightly exceeds the amount of risk accepted Hosee anomalies that are close together Thereforould be lower than what is predicted due to shats Over 15 years the economic risk that will be1 million By implementing the recommended r$135 million in risk over 15 years
Ripple 15
ing thethe incrementalf risks acceptedn economic
echniques suchdramatic
15 where n ise Since n isith no repairswas still used
anomalies over
st of risk that ispairs is always
ended repairsever as Figure
the cost tored expensesaccrued due topair schedule
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1821
Figure 16 Risk associated withrepairs
not repairing all pipeline anomalies vs the ann
Ripple 16
al cost for
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 1921
Ripple 17
Recommendations and Conclusions
Recommendations
While the model is performing as intended the following recommendations would bebeneficial to its future users
bull Test the predictions of this model by acquiring a second set of ILI data of the samepipeline
bull Diversify the distributions available for fitting the ILI data
bull Include analysis of field data such as the effects of product chemistry waterconcentration and solids content on growth rates
bull Apply the same methodology used in this work for external corrosion
bull Determine methods to estimate the effects of external corrosion mitigation
bull Add additional failure estimation models (such as the Shell or DNV-99) as a comparison
to the Modified B31G
bull
Have model display not only the anomaly locations along the pipeline but also theirorientation within it
Conclusions
This model analyzes federally mandated data for pipeline owners by combiningdistribution fitting techniques statistical analysis and an economic evaluation The model hasbeen constructed to permit easy modifications which allow the user to make the data analysis assimple or as complicated as desired Based on the specific pipeline data analyzed and estimatedrepair costs it is predicted that approximately $135 million dollars of risk can be avoided byimplementing an optimized repair schedule for the entire pipeline over 15 years Based on the
model output the majority of repairs should take place between years 13-15 While the modelstill requires sound engineering judgement the results can still be used to drive management andbudgetary decisions Also as heavier crudes become more popular and general corrosionawareness increases it is predicted that the number of anomalies per pipeline will decrease overtime
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2021
Ripple 18
References
1 NASA Fundamentals of Corrosion and Corrosion Controlhttpcorrosionkscnasagovcorr_fundamentalshtm (accessed March 23 2015)
2 NACE International Corrosion Basics httpswwwnaceorgCorrosion-CentralCorrosion-101Corrosion-Basics (accessed March 23 2015)
3 Callister W D Rethwisch D G Material science and engineering An introduction 8thed Wiley amp Sons New York 2010
4 Koch G Brongers M Thompson N Virmani P Payer J Corrosion costs and
preventive strategies in the United States US Federal Highway Administration McLean2002
5 ASM International Corrosion Understanding the Basics ASM International MaterialsPark 2000
6 Nyborg R Controlling internal corrosion in oil and gas pipelines Business Briefing Exploration and Production The Oil and Gas Review 2005 No 2 70-74
7 Larson K R Managing Corrosion of Pipelines that Transport Crude Oils Materials
Performance 2013 52 (5) 28-35
8 United States Department of Transportation US Oil and Gas Pipeline Mileagehttpwwwritadotgovbtssitesritadotgovbtsfilespublicationsnational_transportation_statisticshtmltable_01_10html (accessed March 23 2015)
9 US Department of Transportation The Changing Face of Transportation Bureau ofTransportation Statistics Washington DC 2000
10 Nord Stream Intelligent PIG httpwwwnord-streamcompress-infoimagesintelligent-
pig-3467page=3 (accessed March 23 2015)
11 Alyeska Pipe Pigging the Trans-Alaska Pipelinehttpdecalaskagovsparperpresponsesum_fy11110108301factsheetsfact_Piggingpdf (accessed March 23 2015)
12 Caleyo F Gonzalez J L Hallen J M A study on the reliability assessmentmethodology for pipelines with active corrosion defects International Journal of Pressure
Vessels and Piping 2002 No 79 77-86
13 ASME Manual for Determining the Remaining Strength of Corroded Pipelines IndustryStandard ASME New York 2009
14 Mustaffa Z Developments in Reliability-Based Assessment of Corrosion In Developments in Corrosion Protection Aliofkhazraei M Ed Intech 2014 pp 681-696
15 Mahadevan S Monte Carlo Simulation In Reliability-Based Mechanical Design CruseT A Ed Marcel Dekker INC New York 1997 p 123
16 NIST Related Distributionshttpwwwitlnistgovdiv898handbookedasection3eda362htm (accessed April 152015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)
8182019 Proposed Statistical Model for Corrosion Failure Prediction and L
httpslidepdfcomreaderfullproposed-statistical-model-for-corrosion-failure-prediction-and-l 2121
Ripple 19
17 Kruger J Cost of Metallic Corrosion In Uhligs Corrosion Handbook 3rd ed Revie RW Ed John Wiley amp Sons Ottawa 2011 pp 15-17
18 MathWorks Normal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatsnorminvhtml (accessed April 20 2015)
19 MathWorks Lognormal inverse cumulative distribution functionhttpwwwmathworkscomhelpstatslogninvhtml (accessed April 20 2015)
20 MathWorks Weibull inverse cumulative distribution functionhttpwwwmathworkscomhelpstatswblinvhtml (accessed April 20 2015)