Received: 8 January 2022 Doe 10.1002/prs32368 Revised: 7 March 2022 | Accepted: 23 March 2022 era ORIGINAL ARTICLE Probabilistic failure assessment of oil pipelines due to internal corrosion Sina Khakzad* | Ming Yang? | AliLohi® | Nima Khakzad*® *acuty of Enginering and Applied Science, Memorial University of Newfoundand, St Johns Canada| Abstract il and gas pipelines play a key role in the safe and efficient delivery of energy resources around the world. Crude oll by Itself is not corrosive, but oll extracted from geological reservoirs is accompanied by varying amounts of water and acidic gases such as carbon dioxide (CO), which can form a corrosive combination, Estimating the corrosion rate and depth in pipelines is essential for predicting their failure proba- bility. In the present study, a Bayesian network has been developed for predicting the distribution of corrosion rate in oil pipelines given the point estimates generated Using an empirical corrosion simulation model. For this purpose, the simulation model considers corrosion parameters such as pipe diameter, flow temperature, flow veloc~ ity, and COz partial pressure, among others. With the corrosion rate distribution predicted by the Bayesian network, corrosion depth-rate relationships have been 2acty of Technology. Plc, and ‘Management Delt Univesity of Technology Dal, The Nethertnds Departrnent of Chemical Engineering Ryerson University, Toronto, Canada “Schoo of Ocaspationl and Pui Heath, Ryerson University, Toronto, Canada Correspondence Nims Knee, School of Occupational and Pubic Heath, Ryerson University, 288 Church Steet, Tron, MSB 125, Canad, Ema ninahalzadaryeronca Funding information Publication Gran, Fay of Community employed to convert the corrosion rate distribution into failure probability Series, Ryerson Uriverity distribution. keyworDs Bayesian network depth-rate relationship, allure probability assessment, ol pipeline 1 | INTRODUCTION Corrosion of pipelines by aqueous CO» and H.S has long been 2 ‘concern in the ol and gas industry and stil represents a serious threat Pipelines are stil considered the safest and most cost-effective means ‘of mass transportation in the oll and gas industry. Meanwhile, even ‘minor failures in pipelines may lead to catastrophic consequences. ‘Among the failure modes of pipelines, corrosion appears to be one of the most feared and frequent culprits. Corrosion is a naturally occur- ring electrochemical reaction resuiting inthe degradation of material ‘Aeevitons EN, js netwof Cerason dept Pipe ameter i Pbsiey ‘erat ron cmon ate Cause Seney non af eon a Pp Corrosion ate va: Age of De expourtime to cores 5, Shear tes xp ron & Cuero creson 0 fax, Panes nto isis an open access at ‘the ters of the Creative Commons provide the ognal works propery ced. te pipeline integrity and plays key ren ifecyle assessment of oll pipelines Ol and gas pipeines ae commonly made of carbon stet because ofits good mechanical properties, low cost, and wider avai ily despite having a celatively low corrosion resistance® Appving vanced techniques in ol and gas extraction, such 35 CO, injection for enhanced ol recovery and extraction of deep natural gas reser: ‘ots, has brought this problem to attention again® Dry gas and ol by themselves are ot corrosive: however, the presence ofan aqueous phase on the meta surface along with CO> and HS provides a suitable condition fo corrosion to accur® When cil wel age the production of ol stats to delne whereas the flow rates of water and gas, which usualy contain corosve agents such as Oz, start to increase, accelerating the intel coresion of the pipeline. 1© 2022 The Authors. Paces Softy Pees publshed by Wiley Periodicals LLC on behalf of AmoianIstiuteof Chemical Ergicers. Pres Sof Prog 202283:790-800, wieyontnelbray-consjoumaliers 793KHAKZAD era. | say ‘Although there are steel alloys that resist corosion effectively, mild steel (a type of carbon steel with low carbon content) is widely sed inthe construction of pipelines and process equipment inthe oi and gas industry, because of ts cost effectiveness, The corrosive envi- ronment in the presence of acidic gases of CO, and H,S is classified as sweet and sour corrosion, respectively.” The severly of both types (of corrosion is controlled by the combination of several operational parameters such as temperature, pressure, flow velocity and the envi- ronmental parameters such asthe pH of the fluid insde te pipeline, Data about the rate and extent of crresion ae usually collected by inteigentpigging. which isa highly sophisticated instrument for mea- suring the pipeline wall thickness through electromagnetic waves or lectrochemical potential nose" I the corrosion rate cannot be deter- ‘mined from thichness inspection data, rough estimates may be established using expert elation” or predictions can be made using eterinistic and probabilistic models based on measurement of the key corrosion parameters” Generic fale frequencies of process equipment and pipelines, ue to intemal and external corrosion, are also available” but they need to be tailored before use by taking factors such as management systems, inspection history, an soon, into account. Based on the underlying physicochemical processes of corro- sion, many models have been developed to calculate the corrosion rate of pipelines, including mechanistic models,“ semiempirical models" empirical models * neural networks #9 and numerical simulation of corrosion ferential equations. The Bayesian net- Work (BN) has been effectively used for modeling and failure assess- ment of engineering systems and structures where uncertainty may impede the application of conventional techniques.” BNs devel- ‘oped oF enhanced on the basis of physical models (also known as physical model-based BNs] have recently been employed asa prom- Ising alternative for corrosion anaiysis"®*? and for estimating the remaining usefl life and reiabity of pipelines” and for other engi- neering systems 5 The superior performance of BN aver most + entity corrosion parameters + Set up the corrosion simulation model + Generate point estimates fr corrosion rate + Develop 2 BN based on the output of Step 2 ‘Run the BN to identify the distribution of corrosion rate conventional techniques fs mainly due to its ability in reasoning under uncertainty and updating the prior probabities should new information become avaiable, Inthe present study, a probabilistic method based on BN and ‘Weibull distribution is proposed for predicting the fallure probability ‘of an oll pipeline due to CO; corrosion, Considering the radial corro- sion rate a5 a key factor in corrsion-relatd failure of pipeines.* 2 corrosion rate simulation model” is used to generate point estimates (of corrosion rate (mmv/year) given the pipe diameter, flow tempera- ture, flow velocity, COs concentration (as its partial pressure, and so ‘on. The output ofthe simulation model is used to develop a BN and lean its parameters via the maximum fikethood estimation algorithm, With the slstribution of the corrosion rate predicted by the BN, 3 ‘Weibull distribution is used to predict the corrosion depth and the respective flue probabilities, ‘The steps taken to develop the methodology are ilustrated in Figure 1. In Section 2, the basics of CO corrosion and BN are briefly reviewed (Step 1 is covered in Section 2.1). Section 3 demonstrates the development and validation of the developed BN model for predicting corrosion rates through an ilusrative case study (Step 2 is covered in Section 3.1 while Step 3 is covered in Section 32) In Section 4, a Weibull cstibution is used to estimate the corrosion depth and predict the respective failure probabilities (Step 4 is cov- ‘ered in Section 4). Section 5 concludes the paper. 2 | THEORETICAL BACKGROUND 21 | COz corrosion mechanism “The initiation of corrosion damage i a random process, depending on the microstructure of the material, the surface condition (eg, the “= Use the distribution generated in Step 3 to learn the parameters of Weibull distribution + Use Weibull distribution to convert corrosion rate distebution Into falure probablity distribution FIGURE 1 Steps taken to generate the failure probability distribution inthe present study jorayoie/:sdn wis papeowunoa ‘y 2202 ‘EL6SLYSL sua] sowos anneesyaigeaydde ap hq pousenoS ate sop iue yo ‘98n Jo st Jo) Axe au Kay Uo (sop puaD -pue-suuai/worAapmArezqpuuo//sday)suopipuoy pue sua] ayh 295 [ez02/60/I] vo AzexQr BUNLO hay eURUeBrysueAIDO> Ke ¥9f21'sA/Z00L OWVNOP/AzeD Aan AseLKHAKZAD era presence of surface defects), and other environmental factors.” ‘Aqueous COz corrosion of mild stel is an electrochemical process Involving the anodie dissolution of iron (Fe) and the cathodic evolution ‘of hydrogen (H,) 25° Fe CO2+HyO—FeCOs + Ho ry ‘The presence of CO» can increase the rate of hydrogen evolution and thus accelerate the rate of corrosion. As an oll well ges, the pro- duction of cil starts to decine, whereas the flow of water and gas, hich usually contain highly corrosive agents such as CO>, tends to increase, This, in turn, accelerates the corrosion process inside the pipeline.” Compared to that of COs; the role of iron carbonate (FeCO:), which i usually formed at higher temperatures in the form of soli lms (also known as scales) is twofold: it can be protective and decelerate the corresion, ort could be non-protective, depending ‘nthe environmental conditions under which it is ereated.* tron can aso be anodically dissolved individually in acid solutions as. Fe Fe! +26 @ [As a result, carbonic acd (H;CO,) can react with free electrons and enable hyérogen evolution even at pH > 2HsCO; + 26° Hy 4 2HCO; @ However, carbonic acid is believe to serve as an extra source of hydrogen fons (H"), wii evolution of more hydrogen: in retraction with electrons an lead to the 2H! 420 Hy @ ji has a strong influence on corrasion rate; at pH <4, the reduc: tion of H' ions is important particulary at lower partial pressures of C02 (pCO, directly affecting the corrosion rate. On the other hand, at high pH the solubility of iron carbonate decreases, which in turn Increases the precipitation rate and formation of scales, thereby reducing the rate of coroson? CO», temperature flow velocity, and corrosion inhititer are the other factors that play a key role in the rate of corosion* The increase in pCO, increases the concentration of carbonic acd and accelerates the cathodic reaction and thus the corrosion rate. The role of temperature, however, s controversial: at low pH the increase in temperature accelerates the corosion ate, whereas athigh aH, a high temperature speeds up the formation of protective scales and decreases the corrosion rate. By almost all accounts lw velocity at both low and high pH increases the rate of corosion. Given the corro- sion depth (2) and the pipe wal thickness (the corosion depth can be catepoied as shalow if 01 < <0.2, moderately deep if 0.2 < f -<04, and deep iF 0.4 << 0.6. A detaled analyse of CO, corosion mechanism canbe found elsewhere? aa 22 | Bayesian network | ‘ABN can be defined as an acyclic directed graph BN = (G0, where {G denotes the structure ofthe graph (nodes and edges) and @ denotes the parameters.” Based on the chain rule and the separation ritria, te joint probability distribution ofthe nodes (each node rep- resents a random variable) ina BN canbe factrized as the product of the conditional probabilities of the nodes given their immediate parents PKs XarnKs) =[[ Ppa), 6 ‘where Xs random variable, and plX) isthe parent set of X; that is, the set of nodes from which there are direct edges to X, The condi tional probabilities 6=PIX;pa(X)), which are also known a5 the parameters, can either be estimated by subject matter experts oF be learned from data (observations) using algorithms such 35 the max: ‘mum likelihood estimation (MILE) Given a set of observations = (ds, ‘observation assigns a value to each node as 4 lkelnood function of the parameters can be developed as sd) in ich exch v4). 0 u0:6.0 -Poe~ Frau) TE {T°(ee(4)). By maximizing either the lkelinod function in Equation (6) or its natural logarithm (og-tkeinood function), the parameters of the BN, ‘an be estimated, An ilustraive example on the aplication of MLE is presented in the Appendix A. 3 | MODEL DEVELOPMENT 3.1 | Data generation Considering the infuentl parameter in CO, corrosion, as described in Section 21, the M-508 corrosion rate model which isan empiial model developed by NORSOK.” is usd to generate the data ecuited for estimating the conitonal probaities of the BN model for predicting the corrosion rats. experimental data oF field measure ments are avaiable.” the probabilities needed inthe BN canbe est: trated using data mining techriques. nthe present study, for illustrative purposes, the data generated by the M-506 model is used {or mode! development and parameter lamin “The M-506 has been developed using Microsoft Excel and has a rin datogue box Figure 2) that can perform al corrosion ate caleu- lations wth no ned for (or the possibilty of) changing the program settings ll the input parameters can be entered inthe dlalogue box for pont calculation of the corrosion rate mim/yeal. The model is based on the NORSOK Standard M-508 and is developed by the jorayoie/:sdn wis papeowunoa ‘y 2202 ‘EL6SLYSL sua] sowlos anneal aigeaydde ap hq pousonoS ate sopiue yo ‘98n Jo st Jo) Axe aU Kay Uo (op pUaD -pue-suuai/worAapmdrezqpuuo//sday)suopipuoy pue sua] ayh 295 [ez02/60/I] vo Azexqr BUILO hay euRUeBrysueAIDO> Ke ¥9f21'sA/Z00/ LVNOP/AzeD Aan AseIKHAKZAD era. ne | say —=aiMIOMSO S] FIGURE 2 ierace of 506 ‘corrosion rate model”? = exc ma Scoxtway Sa — = om = — a —— ae Fae coe Ie aa) a = = a — ir eed eee ss een — = = cottesy le eo Snel Soe ear cos Fe son armel uedin the worse so-mox —O3-05bn 7-145 3555 Moin mau gman Nonwesian Oi ndstry Assocation ad the Federation of Norwetan calculates the corosion rte bth the presence and absence of “Manufacturing Industries.“ ‘The M-506 uses, among other parameters, the flow temperature and pressure, flow velocity, and the internal dlameter of the pipe to caleulate the shear stress on the internal wall ofthe pipe. Similary, the temperature, pressure, and pCO, are used to calculate the DH. Having the shear stress, pH, and efficiency of corrosion inhibitor, the M-506 caleulates a single value fr the corrosion rate (mm/yex? ‘The steps taken to enter the input parameters inthe M-506 simulator ate as folow: ‘+ Step 1 Inset the id temperature in °C) in the Temperature box. ‘+ Step 2 Insert the intemal pressure (in bar) in the Pressure box. + Stop 3 Insert the CO, partial pressure (in bar in the box with the ‘+ Step 4, Click “Calculate shear stress” to calculate the shear stress ‘The calculated value wil automatically appear in the Shear stress box fin Pa. Stop 5. Click “Calculate pH” to calculate the fluid pH. The caleu- lated value wil automatically appear in the pl box. ‘+ Step 6. If glycol has been added to the fluid to prevent it from freezing, inert its concentration in the Glycol concentration box in %. ‘+ Step 7 Insert the efficiency ofthe corrosion inhibitor in the Inhibi- tor efficiency box (in ‘+ Step 8. Press the “Calculate corrosion rate” button to run the model and calculate the corrosion rate (in mm/year). The model the corrosion inhibitor. “The mode! is valid for temperatures 5-150, pH 35-65, shear stress 1-150 Pa, and pCO; 0.1-10 bar, although it may mis-predict the corrosion rate when pCO; i ess than 0 bar. The model is appli ‘able to corrosion rates only when CO, is the corrosive agent. In ‘other words, it does not include the addtional effects of other con: stituents that may influence the corrosivity,inckiding contamination ‘of Op and HS, which are more common in water and gas pipelines. ‘To validate the developed BN in Section 3.2, the experimental values reported in Peng and Zeng forthe temperature (1 pCOs, and ‘low velocity (Ware adopted inthe present study, a iste in Table 1, ‘As the same parameters will be used forthe oot nodes ofthe BN, to facilitate the discretization of the nodes later in Section 3.2, the values reported in Table 1 were discretized as T = (60, 70, 75), P=(60, 65, 70, pCO; = (0.3, 04,05), and V = (07, 1.1, 1.6) Further, since the inhibitor efficiency isnot reported in Peng and Zeng.” three values wore considered for the inhibitor efficiency (%) 35 1 = (40, 50, £60), Owing tothe uncertainty regarding the distributions ofthe basic parameters (the root nodes}, uniform distribution was considered for all the rat nodes 50 as to not impose the analyst's prior and possibly biased) knowledge on the analysis: this way, given a root node, al the states of the node would have equal chances to contribute to the probablltycitribution ofthe corrosion rate and consequently the fail Lute probability. Considering 2 wall thickness of 12 mm fo the pipeline of interest, the M-S06 mode! was run 243 times to account forall jorayoie/:sdi wos papeowunoa ‘y 2202 ‘EL6SLPSL sua] sowo anneesyoigeaydde ap hq pousonoS ate soe yo ‘98 Jost Jo) Axe aU Kay Uo (op }pUaD -pue-suuai/worAaymArezqpuuo//sday) suopipuoy pue sua] ayh 295 [ez02/60/I] vo Azexqr UIKLO hay euRUeBrysueAIpO> Ke ¥9f21'sd/Z00/ OWVNOP/AzeD Aan KseLKHAKZAD era TABLE 2 Sample of data generated pe £ TCC) Plbse) COs tba) Vim/s) 11%) pH Shear(Pa)_Coroson rate (mi yest Using the M-506 mode © 6 05 1 42 2 22 76503 or 5044 16 @ 7m 05 16 50 42 4 a7 7% 7% 08 16 4 49 4 28 lO corrosion rate la] Jaze joaox| 20% [© _—mhvbter pe | ES ame fea jes |S239%} pre [633% ona. 33% sss [o peO2 fo 7 io B jo v fers] frv33% brava] To cosas yr2s0% 30% | Wve39%| lc233%| {73.33%6| \ps.33%6| {v333%| FIGURE 3. BN developed for predicting CO, comosion rate possible combinations ofthe five tertiary roat nodes (3° — 243). Given the input parameters, the discrete values of BH and shear stress were calculated by the model as pH = (4.2, 43, 44) and Shear = (1,2, 4) while the corrosion rates were observed to vaty from 1 to 4 mm/year. ‘A sample of the generated datas presented in Table 2 3.2 | Development of the Bayesian network ‘The parameters and the way they are considered by the M-506 to cal- culate the shear stress pH, and, nally, the corrosion rate can be used to construct the structure (nodes and the causal aes, among others) of the BN, 2s shown in Figure 3, Since the M-506 uses the tempera- ture, pressure, and flow velocity to calculate the shear stress, the corresponding nodes are considered as the parents ofthe node Shear (shear stress) inthe BN. Similarly, since the M-506 uses the tempera- ture, pressure, and pCO, to caleulate the pH, the corresponding nodes are considered as the parents of the node pH in the BN. The nodes T, , pH, and Shear, together with the node Inhibitor (inhibitor eff- ciency), are connected to the node Corraion rate as its parents. The odes ofthe BN and their states are presented in Table 3 Using data generated in Section 2.1, the parameter leaming mad Ue of the GeNle software was employed to learn the conditional probabilities of the nodes pH, Shear, and Corrosion rate in the BN using the MLE algorithm presented in Equation (6. Since the values of Corrosion rate were continuous and varying from 1 to 4 mm/year, for the corresponding node in the BN, four discrete states were consid fered as A = 0-1 mm/year, 2 = 1-2 mm/year, 3 = 2-3 mm/yr, and et = 3-4 m/e. With the parameter learning implemented, the BN was quantified to obtain the probability distribution ofthe corrosion rates as Pr 2, 13, r) = (0.05, 0.28, 0.48, 0.20), with the corrosion rate of 2-3 mm ‘year 35 the likeliest corrosion rate with a probability of ~48%. This is In agreement with the experimental result reported in Peng and Zeng,” whore an average corrosion rate of ~3-4 mm/year was reported. The reason why the experimentally measured corrosion rate is slightly higher than the one predicted by the BN can be explained as due to a more acidic condition (pH = 3.5-4) in the experiments conducted by Peng and Zeng. which could not be exactly simulated using the M-506 model. This f because the M-506 calculates pH as a function of T, P, and pCO2, without leting the analyst to arbitrarily set or manipulate its value 3 B 2 i i i z g i Ide ap hq pousonos ate sap ie yo ‘98m Jo st 10) Axe aU Kay Uo (uo}}pUaD Im "eunuabsysuespo> Ae "ySe21s14/2001 ‘aay suo anees 9KHAKZAD era. TABLE 3. ‘Thenodes of the BN in Parameter Node states Probation Figure and ther states. The probably Temperate T n= (033,093,039 distributions ofthe rot nodes are 2-8 identified by the user (uniform 13-75 Aistrbtion we te dstrbutions of Pressure bar ° A 60 (038,093,033) the child nodes are identified bythe BN 7265 sing the maximum ketiood funtion 3-70 02 pressure ae) coz ci-03 (38,033,039 @-os G05 Fw velocity is) v vi=07 (33,023,029 von1a va=16 iio efisecy 8) Iniitor n=40 (033,093,039 2-50 no oH oH oH1=42 (017,050,053) pHa 43 pHa 44 Shear stress Pa) shear st (026,099,035) 2-2 3-4 Conesionatelmmiyex)—Comosionrate 1-0-1 (005,028,048, 020) 2-12 =23 a4 © conesion rate in ox] \22o% lars 425% [Santor or = a lees oa Ex oon o%| “f [soon d Ie COD \o T Io id Io cd ler 6%] 67% aT [pr 47%] \v132%] cz ox [2274 loz lvez0% [C387%| {73 6%) [P327%| [v339%| FIGURE 4 Updated BN given an instantiation of pH node. Corrosion rate distribution is skewed to the right, given a more acidic pH To examine the impact of pH on the corrosion rate, node pH was from the updated BN in Figure 4 this more acidic pH makes the prob- instantiated to pH 1 = 4.2, which is the most acidic state among the ability distribution of Corrosion rate more skewed to the right, others, to update the states of node Corrosion rate. As can be noted increasing the probablty of r4 ~ 34 mm/year from 0.2 (hn Figure 3) 3 B 2 i i i z g i Ide ap hq pousonos axe sap ie yo ‘98nJo st 10) Kreg aU Kay Uo (op pUAD In "euauabsysuesoo Ae "ySe21s14/2001 ‘aay sow anees 9 o/s) suopipu0n pur sui aut 995 (ez02/60/ -pue-sau/io>KapdveKHAKZAD era and 0.26 (in Figure 4). As such, 2 more acidic pH (3.54) would be ‘expected to further increase the likelihood ofr = 3-4 mm/year inthe BN, making the prediction even more consistent with the experiment. 4 | ESTIMATION OF FAILURE PROBABILITY The aim of pipeline corrosion analysis is to estimate the residual strength and failure probability of pipelines with the purpose of risk assessment or identifying maintenance schedules. One approach to relate the corrosion rat tothe probability of failure (POF) of the pipe- line is to first find the corrosion depth given the age ofthe pip, and then to calculate the fallure pressure of the corroded pipeline. Based ‘on a comparison between the failure pressure of the corroded pipe- line and the operational pressure af the pipeline." the POF of the pipeline can be estimated using, among others, Monte Carl simu- lation. This approach, however, is computationally demanding since both the corrosion width and length need to be calculated in addition to the corrosion depth, ‘One simpler approach i based on a comparison between the cor- roslon depth and the nominal wall thickness ofthe pipe.” As pointed ‘ut in Larn et al.” atthe operational pressure, a corroded pipe may incur plastic strains at spots where the corrosion depth (dis larger than half ofthe pipe thickness (h). They also argue that at startup and shutdown pressures, which may be twice the operational pressure of the pipe, there is a 70% probability of falure where the corrosion 22), 9) Using the probability cistibution of the corrosion rate, the POF presented in Equation (9) can be calculated, Corrosion rate, as a rar ‘dom variable, is commonly believed to satisty the Gamma, Weibull, or generalized extreme value cstributions."" Inthe present study, we ‘adopt a Weibull distribution fr the corosion rate, with a probability density function ff) and cumulative density function FR) as tot F(R) ~PirsR)—1-e (10) an) ‘here k> 0 and y »0 are, respectively the shape parameter and the scale parameter of Weibull cstribution, Considering temporal varia: tion of corrosion rate, a value of k< 1 indicates a decreasing corrosion rate, k= 1 indicates a constant corrosion rate, and k> 1 incates an increasing corrosion rate overtime. By combining Equations (9) and (11), the POF can be calculated using Equation (12) ifthe values of andy are known: a2) ‘The probablity cstribution of the corrosion rates predicted by the BN in Figure 3 can be used to estimate the values of k and y for Equation (12). Since there are two unknown variables (k andy), two ‘equations would be required. Considering the corrosion rate probabil: ties, one equation can be P(2 Ke 921/200! VNOP/AzeD Aan AseLKHAKZAD era oo | Bsa roroyee (A) ae 8)" gem, ay Using Equation (14), the failure probability distribution of the pipeline can be presented, asin Figure 6, For example, att = 3 years, the cumulative failure probability is calculated as POS) O23") 073, 4.2 | Second approach The corrosion depth in uniform coerasion mechanisms can be ‘modeled asa power aw function of the exposure time as"? ape, (a5) where d is corrosion depth, ¢ is exposure time, and a and pare con- stants to be determined using experimental or field messurements of 4. Assuming a constant [but random) corrosion rate overtime, would be equal tr, and « would be equal to 1.0. Thus, Equation (15) can be simplified as (18) Plugging Equation (14) in Equation (7), the failure probability can be calulated as POF — Pid 20:3h) =P(rt20.3h) =P a7 With h = 12 mm, te probability mass distribution derived from the BBN forthe comosion rate can be used to quantity Equation (17) for eit- ferent values oft This mass distribution is presented in Figure 5 for clar- ity, For instance, fort = 4 year, Equation (17} would be quantified as ori (222222) pds) 1-096 =1- (PlOsr<4)4P(dsr<2) +P(2sr<3) +P(3sF<3.6)} (2a) corouonate mest Probably FIGURE 5 Corrosion rate mass distribution derived from BN in Figure 3 FIGURE 6 Failure probability distributions. The Second approach has resulted in higher failure probabilities due to the assumption of constant corrosion rate overtime According to Figure 5, P(0s1< 1} = 005, PLL s <2) = 0.28, and 2 <1<3) = O48, Further, assuming. that the probability mass is equally spread over the corrosion interval r = (3, 4), AIS <1<3.6) = 06 x 020 = 012. Having the constituent proba: bites, POF (¢ = 1) = 1 - (005 +028 +048 + 0.12) = 007. “The probabilty of failure for other consecutive yeas can be cal- ‘ulated similarly. The resultant probability distribution (second approach) is depicted in Figure 6 along with the results obtained from the fist approach. The area with meaningful differences between the results is denoted with a red solid line, where the falure probabilities ‘aleulated by the second approach are higher than those of the fist ‘approach fort = 2, dand 4 years. “The higher failure probablities in the second approach can be attributed to higher corrosion depths calculated using Equation (16) than Equation (6), This becomes clearer when we note that Equation (16) is a special case of Equation (8) in which w = 1.0. In Equation (8), the role of a (03-10) is to consider the decelerating effect of environmental and corrosion parameters (eg, corrosion scale) on the corrosion rate as time passes. Corrosion scales consist of hard mineral coatings, and corrosion deposits bull up overtime at the location of corosion, providing 9 protection layer and reducing the corrosion rate.® Since most empi ‘cal models assume constant corrosion rates, deceleration of the corro- sion rate can be taken into account through the value of «. In other ‘words, «'< 1 denotes a corrosion depth advancing with 2 decreasing pace (slowing corrosion rate), whereas a = 1 denotes a corrosion depth increasing with a constant rate (constant corrosion rate). There- fore, the difference between the failure probabilities can partly be attributed to the assumption of a more aggressive corrosion rate in the second approach (a = 1.0) han the first approach (x = 0.6) 5 | CONCLUSIONS In this study, we presented a methodology based on BN for predicting the probability distibution of corrosion rate in oll pipe lines. An empirical model, namely the M-S06 corrosion rate i : F i : i : % i i i 3 q 2 i z : = Ide ap hq pousonoG axe sap ie yo ‘98n Jo sant 10) Axe aU Kay Uo (Uap pUaD in "eupuabaysueaine Ae ySe21=4/200" jorayoie/isdn wos papeowunoa ‘y 2202 ‘EL6SL¥SL ropnuookendce i iKHAKZAD era simulator? was used to generate the data required forthe develop- ‘ment and parameter leaning of the BN. The corrosion rate probabil- ity distribution predicted by the BN was shown to be consistent with the experimental resuts.* Compared with the empirical mode, the BN model was shown to account for the uncertainty of input variables, enabling the analyst to easily update the probability distr- bution of corosion rate when new information (e.., new data for ol temperature and pressure) becomes available. This updating feature of the BN Is particularly important because recalibration of parame~ ters in semiempirical and empirical models based on new data can Usually be a tedious task®* We demonstrated that the result of the BN (le. probability istribution forthe corrosion rate) can be combined with corrosion opth-rate relationships to convert the corrosion rate's probabilty istibution into the pipelines failure probability distribution. To do 50, the rate distribution was processed under two separate assump- tions, resulting in two approaches: In the frst approach, the rate distribution was considered to have a Weibull distribution and a power-law function for the corrosion depth-rate relationship was sed to imply that the corrasion rate might decelerate overtime. In the second approach, however, the rate distribution derived from the BN was used with no further assumption regarding its distribution type, and a linear corrosion depth-rate relationship was employed to denote that the corrosion rate remains constant aver time. Both approaches were illustrated to resuit in similar trends forthe faliure probability distribution, although, overall, the near depth-rate relationship yields higher fallureprobabiites. ‘With new information regarding the input parameters, the devel- ‘oped methodology can be used to update the distributions of both the corrosion rate andthe failure probability. To do so, the BN can be sed directly to update the rate distribution, which can subsequently be used to update the failure distribution. For instance, regarding the first approach (Section 4.1), glven an updated rate distribution, Equation (13) can be re-solved to calculate the updated values of and 1 {the parameters of Weibull distribution, and then Equation (14) can be recalculated using these updated k and y to obtain updated failure probabilities. ‘The present study has not been aimed at advertising or validating the M-506 model but it was to demonstrate how empirical models like M-506 can be coupled with BN and corrosion depth-rate rela- tionships to predict the corrsion-induced failure of pipelines. As such, the accuracy ofthe method fs expected to improve if more accu- rate models than the M-506 were employed. ACKNOWLEDGMENTS ‘The authors are grateful to the two reviewers for their instructive comments that helped enhance the readability and quality of this study. The financial support provided by the Faculty of Commurity Services, Ryerson University, via the Publication Grant is much appreciated. CONFLICT OF INTEREST ‘The authors declare no conflict of interest. Bsa | oo AUTHOR CONTRIBUTIONS ‘Sina Khakzad: Formal analysis lead: methodology (lead) writing, ~ ‘original draft (lead). Ming. Yang: Validation (equal: writing ~ review ‘and editing equal. Alf Lohi: Validation (equal; writing - review and ‘editing (equal, Nima Khakzad: Conceptualization (lead); supervision (ead); writing ~ review and exiting lead), DATA AVAILABILITY STATEMENT “The data that support the finding ofthis study are availabe from the ‘corresponding author, [NK], upon reasonable request. orciD ‘Nimo Khalead @ https /orciore/0000-0002-2899-6830 REFERENCES 4 Khakzad S, Khan F, AbbassiR. Khakrad N. Accident risk-based fife ‘oye assessment methodology for green and safe fuel selection. Proc Safty Environ Protect. 2017:109-268-287. 2. Bonis MR, Crolt JL. Basics of the Prediction ofthe Risks of 2 CO Cone sion in Oi and Gas Wels. Corrsin/ 89, Paper no. 464, NACE Intern tora: 1989. 4, man MN, Kusmono, Analysis of internal corrosion in subsea ll pine line. Case StudEng Fail rel 2014:20}1-8.htips//doiorg/10.1016/, ‘sefa.2013.12003 4. Zhao G, Lu X, King J, Han Y. Formation characteristic of CO, corr Sion product layer of PL10 stel investigated by SEM and electro: chemical technique. J Ion Ste es Int. 2009-1614 89-94, 5. Lee K. A Mechanistic Moding of CO Corrosion of Mid Sta in Pres lence of H2S. [PRD thesis). Ohio University: 2008. 6. Rodtiguer M, Delgado D, Gonzalez Ret al. Corrosive wear fare analysis in a natural gs pipeline. Weer. 2007:263:567-571, 7. NesicS. Key issues related to modeling of internal corrosion of oil and gs pipelines ~ a review. Coos Sl 2007:49:4808-4338. 8. Bhandari J, Khan F, Abbassi R, Garaniya V, Ojeda R. Modeling of pitting corrosion in marine and offshore steel structures ~ a technical review 1 Los Preven Proces Init 2016:97:9-62, 9, Kamsu-Faguem B. Information structuring and rk-baeed inspection forthe marine cl pipelines. Appl Ocean Res 2016:56:192-142 10. Kiefner JF, Veth PH, New method corrects criterion for evaluating corroded pipe. Oi Gas .1990:656-59. 41 Abammed M, Melchers RE. Reliability estimation of pressurized pipe lines subject to localized corrosion defects. nt J Press Vessels Pip. 1996:69:267-272 132 Anammed M, Probabilistic estimation of remaining fe of 2 piping in the presence of setve corrosion defects. Int J ress Vessels Pig. 1998,75:321-200, 18, Melehers RE, Jeffrey R. Probabilistic modes for stel corrosion loss and pitting of marine infrastructure, Reab Eng Syst Sofety. 2007; 99t3}429-432. 14. LS. Yu, Zeng H, etal. Predicting coroson remaining life of under ‘sound pipelines with 2 mechanicaly-based probabilistic mode Peto Se ng. 2009:65:162-166. 45, Netto. A simple procedure for the prediction af the collpse pres sure of pipelines with narrow and long corrosion defects ~ correla tion with new experimental data. App! Ocean Res. 2010:32(): 132-134 16, Bazan F, Beck A. Stochastic process corrsion growth models for pipeline reliability. Cons Si 2019:74:50-58 417. Latin 0, Barkanov E, Vota ©. Prediction of reliability ofthe corroded Pipeline considering the randomness of corrsion damage and its sto chasic growth. En ai Anal, 2016:66:60-71 juorayoie/:sdi wos papeowunoa ‘y 2202 “EL6SL¥SL sua] sowio anneesyo}geaydde ap hq pousonoS axe soi yo ‘98h Jo sat Jo) Axe au Kay Uo (op pUaD pu -suuai/worAapmArezqpuuo//sday)suosipuoy pue sua] ayh 295 [ez02/60/I] vo Avexqr BUNLO hay euRUeBrysueAIDO> Ke ¥9f21'sA/Z00/ WVNOP/Aze> Aan KseLKHAKZAD era. | sy 12. 1. 21, 2 2 24, 2, 2% 2. a1 22 Eo 36 37 g8 Leia BU, Nass A, Ness OB. Realty analysis of corroding pipelines by enhanced Monte Carlo simulation, Int J Press Vessls Peng. 2016; 144197, [ASME-B31G. Manual for determining the remaining strength of cor roded pipelines. Supplement of ASME 831G Code for Pressure Pin. ‘American Society for Mechanical Engineers; 1995, {e Waard C, Miliams DE. Prediction of Carbonic Acid Corrosion in "Natural Gas Ppeines, in: Fst Intemational Confrence onthe nter- ral and External Corrosion of Pipes. paper F2, Univesity of Durham: 1975 {de Waard C, Lote U. Predton of CO; Corrosion of Carbon Stee Corrosion’ 93, Paper. Vol no. 69. NACE International 1993. Dugstad A, Lunde L, Videm K. Parametric Study of CO> Corrosion of Carbon Steel, Coresion/94, Pope. Vol no. 14. NACE Intemational 1994 [NORSOK M-506 COs corrosion Rate Model. (2005) hits, standar.no/en/sectrs/energ-og-klma/petroleum/norsokstandard- categories /n-material/m-5061/. Accessed July 28,2020 "Nesic 8, Nordsveen M, Maxwell N, Vitovac M. Probabilistic mode ling of C0 corrosion laboratory data using neural networks. Coos Sei, 2001:49:1979-1392, Yao ¥, Yang ¥, Wang Y, Zhao X Ail intligence-based hull structural plate corrosion damage detection and recognition using «onvolutional neural network. Appl Ocean Res. 2019;90-101823, Sanchez JF, ama F, Moreno JA. An efficient and reliable model based on network method to simulate CO, corrosion with protective Iron carbonate fis. Comput Chem Eng. 2012:9:57-64, Khokaad N, Khan F, Amyatte F. Safety analysis in process faites comparison of tau ee and Bayesian network approaches. Read ng Syst Safety. 2011:96:925, 982. ‘Shabarcin O, Tesfomariam S. Ineral corrsion hazard assessment of oll & gs pipelines using Bayesian belief network model. J Loss Preven Process Indus. 2016:40:479-495, ‘Ayo F, Liu G, Zhang J. Decision making through the application of ‘Bayesian network for intemal corrosion assessment of pipelines, Pro ceedings ofthe ASME 2018 37th International Conference on Ocean, COfshore and Arctic Engineering. Volume 4: Materials Technology. ‘Madi, Spain une 17-22; 2018 ‘ArzaghiE, AbbassiR, Garaniva Vet al. Developing a dynamic model for pitting and corrsion-fatigue damage of subsea pipelines. Ocean Eng. 2018;150:391-996, ‘Abubakiro R, Yang M, Khakzad N. risk-based approach to determi ratlon of optimal inspection intervals fr buried ol pipelines. Process Safty Environ Protect. 2020;13495-107. ‘Yacl M, Khan F, Abbas! R. Operational subsea pipeline assessment affected by multiple defects of microbiological iftuenced corrosion, Process Safety Eniron Protec. 2022:158:159-171 (Ca 8, Shao X, Lv Y, eta. Remaining useful fe estimation of struc ture systems under the influence of multiple causes: subsea pipelines 2353 case stud. IEEE Trans Indus Electron. 2020,67(7}5737-5747. (Cai B, Sheng C. Gao C, etal. Artifical intetigence enhanced reSaiity assessment methodology with small samples. IEEE Trans Newral Netw Learn Sst. 2028113, doi:10.1109/TNNLS.2021.3128514 ‘Khakaad N, Van Gelder P. Vulerabilty of industrial plants to food- Induced natechs: A Bayesian network approach. Relab Erg Syst Softy. 2018:169:403-411, ene 5, Zeng Z. An experimental study on the intemal corrosion of a subsea multiphase pipeline. Petroleum. 2015;1(1}75-81 ear J Probabilistic Reasoning in inelgent Systems. Morgan Kaufmann 1928, "NeapoltanR Learning Bayesian Networks, Prentice Hal 2008, Ricker RE. Analysis of piptine steel corosion data from NBS (NIST) studes conducted between 1922-1940 and relevance to pipeline ‘management J Res Natl Inst Stand Technol, 2010;11515:73-392. 40, NORSOX Standard M-S06, Draft 1 for Rev. 2, March. 2005. hitps:// ‘ikiolsystemscom/wiki/Images/e/eb/Norsok standard, M-506 pa. ‘Accessed December 12, 2021. 41. Goll, Academic Installer, Decision Systems Laboratory, Universty ‘of Pitsburg hts! download bayestusioncom/leshtmiZeategory= ‘Academia Accessed July 10, 2021 42. DNV. Corroded Pipelines Recommended Practice DNV-RP-F101, October 2010. 48. Opevemi D, Patel E, Beer M, Timashev S, Comparative Studies on Assessment of Cavosion Rates in Pipelines a5 Semi-Probablstic and Fully Stocoste Values. International Conference on Applications of Statistics and Probability in Gil Engineering: 2015, 444, Hasan 5, Khan F, Kenny 5. Probability assessment of burst limit state due to intemal corrasion. Int J Press Vessels Piping. 2012:09 48-58, 45. Caleyo F, Velizquez J, Valor A. Hallen J. Probablty distribution of pitting corrasion depth and rate in underground piptines: a Monte Caro study, Corts Sl, 2009:51:1925-1994, 46, Velazquez JC. Caleyo F. Valor A, Hellen JM. Predictive model for pitting corrosion in buried oll and gas pipelines. Corrosion. 200965: aa2-a42 47. BubenikT, Olson , Stephens D & Francini R Analyzing the pressure strength of coroded pipeline. In Proc. L1th int. Conference on Off shore Mechanics and Arctic Engineering. Vol V, Part A. ASME 1992 pp.225-231, 48. Baker M. Stress Corrosion Cracking Stes, Integrity Management Pro ram DTRS56-02-0-70036, Department of Transportation; 2004 49. Nahal M, Khoi Falure probability assessment for pipeline under the corrosion effect. Am J Mech Eng. 20142(115-20, 50. Cabrini M, Lorene S, Pastore T. Corson behavior of carbon steels in CTS environment. Int J Comes. 2016:9121247, doi:10.1255/ 2ore/a121247 How to cite ths article: Khakzad , Yang M, Loh A, khakzad N. Probabilistic fllure assessment of cl pipelines due to{nteral corrosion. Process Saf Pog. 2022:44(4)793-B08. sdoi:10.1002/prs.12364 APPENDIX A: Application of maximum likelihood function Consider a BN consisting of three binary nodes, two root nodes X ‘and , and a child node ZX, Y, and Z as discrete random variables can assume values of 0 or 1. Given a dataset asin Table 4, the probability of @ = 1| x= 1, y = 0) among others, can be estimated using the MLE algorithm, To do so, the estimated probabilities canbe presented 250=P\2=1|x=1,y=0) and 1-0=P(2=0|x=1,y=0) ‘As such, the lkethood function (t) for observing the data pairs lsted on lines 2,4, 9, and 10 in Table 4 can be developed as L0x (1-0) x0x0=08 oF “To find the value of 0 that maximizes the tketnood function, we ‘an calculate the fist derivative of the likehood function with respect to 0, and then solve forthe value of @ that makes this deriva tive equal to zero: jorayoie/sdi wos papeowunoa ‘y 2202 ‘EL6SLPSL sua] sowlos anneesyo|geaydde ap hq pousenoS ate sop iue yo ‘98n Jo st 10) Axe aU Kay Uo (op pUaD -pue-suuai/worAapmArezqpuuo//sday)suopipuoy pue sua] ayh 295 [ez02/60/I] vo Avexqr UINLO hay euUeBrysueAsDO> Ke 921/200! OWVNOP/Aze> Aan KseLKHAKZAD era TABLE 4 Exemplary datast to estimate the conditional probability of Z siven X and ¥ Line No. x y z 1 1 1 1 2 1 ° 1 a o o o 4 1 ° ° 5 o 1 1 6 ° ° 1 7 o o o 8 1 1 o ° 1 ° 1 10 1 ° 1 u 1 1 1 2 ° 1 ° a Ah at 4 0-8 0-0-0 4 ThertoreBe=the=1.y=0)=2 ‘As expsnedin Section 22, one may decide to maximize the log- Iiketnod function to fnd the vale Logl =Un(o* x (1-0)) =3tn(0) +Ln(1 0) allogl)_3__4 a a0 1-0 Oe Using the MLE algorithm, the CPTS of nodes pH, Shear, and Corrosion rate in the BN have been determined by GeNle."” The CPT ‘of node pH is show in Table 5 as an example al TABLE 5 CPT of node pH in the BN in Figure 3, which has been developed using the MLE algorithm in GeNle™= pH No pCO, T= Ppt pH pH. aaemeacl 1 PL 00xs mss saa 2 4 Tm P2083 oss sss. soa T P3088 onss sss 4 4 PL amas oss sss. 5 cl m2 2 033 aos asaa CRC SSS) 7 G@ 7 PL oo oo asaa ER so) o G@ 13 8 oo oo ayaa 100] 933 0089 mo o@ T2033 9330089 2 @ TPB 003309830088 Ee co nen rE EE O19 C9 CL 4 00330983 0.088 6 @ m2 mi OOS OF ams 16 @ TS pL 0033983 0.088 v @ 7 m2 O03 O7m ams wT] aos 088 Po Eo cy a @ 1 8 oss ons ose 2 @ fm PL oss cox 099 mB & 2 Pm 00% 993 099 mms osas os iE cone ro PE O10 1959 OOS 2% TS R033. 0983 0.088 2730330983 0.088 jorayoie/:sdiy wos papeowunoa ‘y 2202 ‘ELSSL¥SL s2pnze Yo 125 jo san 40} Ig BuO AOI Uo (BUoIpUOD z a 3 s q g rH < 3 = a 8 : ‘say sowios anneesy aigeaydde ap hq pouseno sua 24 995 1202/60 -pue-suiay/wo> Aer krexqeuuo/fsday)suoppue> Bu