A comprehensive framework for evaluation of piping reliability due to erosion corrosion for risk-informed inservice inspection

Gopika Vinoda,*, S.K. Bidharb, H.S. Kushwahaa, A.K. Vermab, A. Srividyab

a

Reactor Safety Division, Bhabha Atomic Research Centre, Mumbai 400 085, India b Indian Institute of Technology, Mumbai, India

Abstract Risk-Informed In-Service Inspection (RI-ISI) aims at prioritizing the components for inspection within the permissible risk level thereby avoiding unnecessary inspections. The two main factors that go into the prioritization of components are failure frequency and the consequence of the failure of these components. The study has been focused on piping component as presented in this paper. Failure frequency of piping is highly inuenced by the degradation mechanism acting on it and these frequencies are modied as and when maintenance/ISI activities are taken up. In order to incorporate the effects of degradation mechanism and maintenance activities, a Markov model has been suggested as an efcient method for realistic analysis. Emphasis has been given to the erosion corrosion mechanism, which is dominant in Pressurized Heavy Water Reactors. The paper highlights an analytical model for estimating the corrosion rates and also for nding the failure probability of piping, which can be further used in RI-ISI.
Keywords: Risk informed in-service inspection; Erosioncorrosion; Markov model; First order reliability method

1. Introduction 1.1. Background Piping systems are part of most sensitive structural elements of power plant. Therefore, the analysis of these system and quantication of their fragility in terms of failure probability are of utmost importance. From plant operating experience, it has been found that various degradation mechanisms can result in piping failures like thermal fatigue, vibration fatigue, Erosion Corrosion (E/C), Stress corrosion cracking, corrosion fatigue, water hammer, etc. Recent inspections have indicated that carbon steel outlet feeder pipes in some CANDU reactors are experiencing wall loss near the exit from the reactor core [1]. Examination has indicated that the mechanism causing the wall loss is erosion corrosion, at rates higher than expected. Experimental observation or plant measurements strongly reveal that E/C also depends on piping layout, local

distribution of ow properties and ow chemistry characteristics. Since CANDU plants have seen various instances of E/C attack, attention has been given to estimate a realistic value for piping failure probability due to erosion corrosion. This paper presents a framework for estimating the piping failure probability due to erosion corrosion and further describes model to incorporate the effects of In-Service Inspection so that realistic estimate can be deduced. 1.2. Objective and scope of the study This study originated with an aim to nd the realistic failure frequency of piping segments based on the degradation mechanisms to be employed in Risk Informed In-Service Inspection studies. Since E/C is one of the prominent degradation mechanisms, estimation of corrosion rate due to this mechanism is the scope of the current study. However, after the corrosion rates have been established, the rest of the approach can be applied to other corrosion mechanisms in a similar way. Since First Order Reliability Method (FORM) is being widely used in Structural Reliability Analysis problems, the authors also propose the same approach [2]. After estimating the base failure

188

erosion corrosion from Indian PHWRs are very limited and not sufcient in suggesting the failure frequency of piping due to this mechanism. Not much attention has been focused by researchers for developing models for rates of Erosion Corrosion since this mechanism is predominant mostly in PHWRs. Numerous empirical and semi-empirical models have been developed, which depend on eld experience for some of the factors involved. The main challenge is to develop a complete mathematical model for erosion corrosion rate which could effectively and accurately predict the rate. Once corrosion rate is known, this can be easily interpreted as crack depth growth rate which can be used in limit state functions to predict the failure probability. 2.1. About erosion corrosion Erosion-corrosion of the material is a complex phenomenon, which is dependent on solution chemistry, properties of impacting particles and ow environment. The action of erosion may cause removal of passive corrosion lm thereby removing the ability of the material to withstand corrosion. On the other hand the effect of the chemical environment may reduce the ability of material to resist mechanical attack and cause this latter effect, the so called synergistic effect, which is still not well understood. When water reacts with iron, an oxide lm is formed on the surface of the metal. This oxide layer consists of magnetite when water is deoxygenated. The magnetite layer is porous and slightly soluble in water, which makes it less protective. The solubility of magnetite is a function of temperature, hydrogen activity, and solution composition. According to Sweeton-Baes experiments [6], four different ferrous ion complexes can be formed upon dissolution of magnetite. The chemical equation describing this process is
Fig. 1. Framework for failure frequency estimation.
22b Fe3 O4 32 2 bH 3FeOH 4 2 3bH2 O b

probability, Markov approach has been employed as suggested by Fleming et al. [3 5]. Markov model nds a realistic failure frequency, incorporating the effects of InService Inspection and degradation mechanism. Fig. 1 depicts the complete frame work for the ow of activities to be carried out towards estimation of failure frequency of piping segment.

2:1 where b 0; 1; 2; 3: The equilibrium constants Kb ; were calculated from a relationship derived by least square t to experimental data.
22b =3 Kb FeOHb =H 22b P1 H2

2:2

2. Estimation of corrosion rates The PHWR primary piping is made of carbon steel of grade A-106 GrB operating around 300 8C. Essentially major decrease in erosion-corrosion rate is found as one approaches near 300 8C. If the pH of the water can be raised to 9.5, this rate is reduced to a factor of 100 1000 compared to pH of 9.0. Obviously, carbon steel systems are operated at about 200 8C at velocities greater than 6 m/s with lower pH values. Operating experience data on piping failures due to

22b is the activity of bth ferrous ion complex, FeOHb [H ] is the activity of the hydrogen ion in the solution, PH2 is the pressure of molecular hydrogen gas.

2.2. Determining the relevant hydrodynamic parameters for E/C Mathematical models for the estimation of rate of erosion corrosion depend on large of parameters. Since these parameters are interrelated, complexity has been increased further in deriving these parameters. Recent advance in understanding of erosion corrosion mechanism

189

rate has been focused in identication of regimes of behavior of this mechanism using quantitative technique. As a result of large number of experimental works conducted, several key variables are identied that inuence the rate of attack [6]: Fluid velocity Fluid pH-level Fluid oxygen content Fluid temperature Component geometry Component chromium, copper and molybdenum content.

x 0:86 in straight pipes 0.54, when ow is fully turbulent 0.67, when ow is developing in the downstream DO2 7:4 1028 Temp 2:6 180:5 =2900:6 d diameter of the pipe U ow velocity n kinematic viscosity Cs : Surface concentration 2 :6

X X

22b FeOHb =3 Kb H 22b P1 H2

exp22FE=3RT F faradays constant 96,400 C mol21 E potential in equilibrium system Cb a given bulk concentration Total erosion corrosion rate can be dened as by
1 21 21 Rate R2 K RMT

2 :7

The decision concerning the prioritisation of various classes of piping components such as elbows, bends etc., from most to least susceptible to erosion-corrosion is very complicated. It depends on the interaction of several variables with weighing factors applied to each of the variables. Empirical models are formulated, which considers all the variables responsible for erosion corrosion to happen. These models can predict the rate of E/C with considerable accuracy. This predictive capability helps to avoid nonproductive inspection efforts. 2.3. Steady state model for erosion corrosion The erosion corrosion of carbon steel in water of low dissolved oxygen content occurs mainly due to ow assisted dissolution of normally protective magnetite lm that forms on the surface. M. Abdulsalem, proposed a steady state model for erosion corrosion of feed water piping [6]. It has been discussed that the rate of erosion corrosion is dependent on two factors (i) oxide dissolution and (ii) mass transfer based on the oxide dissolution. The kinetics of erosion corrosion is governed by two steps that operate in series. The rst step is the kinetic rate of oxide dissolution, Rk : This rate can be expected to be governed by an Arrhenius relationship given by: Rk R0 exp2Ek =RTemp 2 :3 where Ek activation energy 31,580 cal/mol R0 9:55 1032 atoms=cm2 s Temp temperature in K R universal gas constant 2 cal/mol/K. The second step involved is the estimation of mass transfer limited rate RMT ; RMT K Cs 2 Cb 2 :4 where K : mass transfer coefficient DO2 =d0:0791Ud =nx n=DO2 0:335 2 :5

2 :8

This rate can be used in models for limit state functions of pipe failure for estimating the failure probability. 3. Markov model for incorporating effects of ISI and degradation mechanisms 3.1. Discrete state Markov model for pipe failures The objective of Markov modeling approach is to explicitly model the interactions between degradation mechanisms and the inspection, detection, and repair strategies that can reduce the probability that failure occurs or the failure will progress to rupture. This Markov modeling technique starts with a representation of piping segment in a set of discrete and mutually exclusive states [3 5]. At any instant of time, the system is permitted to change state in accordance with whatever competing processes are appropriate for that plant state. In this application of Markov model the state refers to various degrees of piping system degradation or repairs, i.e. the existence of aws, leaks, or ruptures. The processes that can create a state change are failure mechanisms operating on the pipe and process of inspecting or detecting aws and leaks, and repair of damage prior to progression of failure mechanism to rupture. The basic form of Markov model is presented in Fig. 2. This model consists of four states of pipe segment reecting the progressive stage of pipe failure mechanism: the state with no aw, development of aws or detectable damage, the occurrence of leaks and occurrence of pipe ruptures. As seen from this model pipe leaks and ruptures are permitted to occur directly from the aw or leak state. The model accounts for state

190

Fig. 2. Markov model for pipe elements with in-service inspection and leak detection.

dependent failure and rupture processes and two repair processes. Once a aw occurs, there is an opportunity for inspection and repair to account for in-service inspection program that search for signs of degradation prior to the occurrence of pipe failures. Here the Leak stage L does not indicate actual leak, but represents a stage in which remaining pipe wall thickness is 0:45 t to 0:2 t (pipe wall thickness). The Markov model diagram describes the failure and inspection processes as discrete state-continuous time problem. The occurrence rates for aw, leaks and ruptures are determined from limit state function formulation. The repair rates for aws and leaks are estimated based on the characteristics of inspection and mean time to repair aws and leak upon detection. The Markov model can be solved by setting up differential equations for different states and nding the associated time dependent state probabilities. These equations are based on the assumption that the probability of transition from one state to another is proportional to transition rates indicated on the diagrams and there is no memory of how current state is arrived at. Assuming the plant life of 40 years, state probabilities are computed for the plant life. Inspect and repair aw rate, v

TI mean time between inspections for aw, its typically 10 years for nuclear power plants TR mean time to repair once detected, is in order of days, 200 h. Repair rate

m PLD =TI TR

3:2

PLD probability that leak in the element will be detected per detection period (Typically assumed as 0.9) 3.2. Piping failure probability estimation using FORM To determine the different transition rates f; lf rL and rf ; limit state functions, based on strength and resistance, are used. The rst limit state function is dened as the difference between the pipeline wall thickness t and depth of corrosion defect. This limit state function describes the state of depth of the corrosion defects with a depth close to their maximum allowable depth before repair could be carried out that is 85% of the nominal pipe wall thickness 0:45 t: The probability that pipe fall thickness reduces to 0:45 t will occur at a rate, f; which is dened as occurrence of aw. So, f represents transition rate from state S, in which aw is less than 0:125 t; to state F in which aw is 0:45 t: The limit state function can be dened as LSF1d; T 0:45 t 2 d rate T d undetected aw 0:125 t: T time of inspection usually 10 years. The second limit state function is formulated to estimate the transition rate lf : lf represents transition rate from state F, which is already crossed the detectable range i.e. 0:45 t; to the leak state L, i.e. 0:8 t: The LSF 3:3

v Pf1 PFD =TI TR

3 :1

Pf1 probability that piping element with a aw will be inspected per inspection interval. The value will be 1 if it is in the inspection program or else it will be 0. PFD probability that a aw will be detected given this element is inspected. This is the reliability of inspection program and equivalent to Probability of Detection. For most Non Destructive Examination, its values are between 0.84 and 0.95.

191

for this case would be LSF2 0:8 t 2 0:45 t rate T 3 :4 There is a probability for the piping reaching directly the rupture state, R from the aw state, F, because of encountering the failure pressure in the aw state. For this case, a different limit state function needs to be formulated. The third limit state function is dened as difference between pipe line failure pressure Pf and pipeline operating pressure Pop [2]. LSF3Pf Pf 2 Pop 3 :5

For this case, the depth of corrosion is dened as dT 0:45 t rate T 3:11

For determining failure pressure, different models are available. For the scope of this paper, two models namely modied B31G and Shell 92 are addressed. According to modied B31G model, Pf is dened as [2]: ! 1 2 0:85 dtT 2YS 68:95t Pf D 1 2 0:85 dtT M 21   LT ,4 for G 0:893 p 3 :6 Dt where s LT 2 LT 2 2 0:003375 2 2 M 1 0:6275 Dt D t for L # 50 Dt LT 2 3 :3 Dt L2 . 50 Dt
2

Both failure pressure models are used to calculate the rupture frequency from aw stage. The results are obtained from software COMREL [8], which represents the failure probability over the entire life of the plant. So the failure frequency or transition rate rf is found out by dividing this probability by designed plant life time, typically 40 years. Similarly, for calculation of rL both the failure pressure model of Modied B31G and Shell 92 are used. In this case, the failure probability is found out by considering the fact that the state transition is occurring from state L to state R which is the rupture stage, over a period of 40 years. The state transition rate rL is obtained by dividing the probability obtained from COMREL by designed plant life time. The corrosion depth for this case is computed as dT for this case 0:8 t rate T : 3:12

Normal distribution has been assumed for load and resistance variables. For longer service periods, it has been found that Shell -92 model gives higher probabilities of failure while modied B31G gives smaller estimate.

4. Case study 4.1. Corrosion rate estimation 3 :7 The PHWR outlet feeder piping system is taken as a typical case study. There are 306 number of small diameter pipes of diameter ranging from 40 to 70 mm and length 2 22 m that connects outlet header to the steam generator. The feeder pipe considered in this case study is made of carbon steel A106GrB, with a diameter d of 70 mm and thickness t of 6.5 mm. This feeder is subjected to a ow velocity U of 1500 cm/s, in a PH of 10.2 at a temperature, 280 8C. The kinematic viscosity, n is taken as 0.0179 cm2/s. The case study attempts to determine the erosion corrosion rate for one such feeder pipe. Following the methodology described in Section 2.3, the rate of erosion corrosion was found to be 0.051 mm/year, which is the mean value for the rate. The variance of the rate can be calculated by using Taylor series expansion. Corrosion rate f T ; pH; U ; d
Table 1 Parameters mean values and variances Parameters Temperature pH Velocity Diameter Rate Mean 553 K 10.2 1500 cm/s 70 mm 0.051 mm/year Variance 25 K 0.5 50 cm/s 1.48 mm 0.015 mm/year

M 0:032

for

3 :8

According to Shell-92, the failure pressure can be dened as [2]: ! 1 2 dtT 1:8UTSt Pf 3 :9 D 1 2 dtT M 21 where s LT 2 M 1 0:805 Dt D out side diameter of the pipe. L length of corrosion. t thickness of the pipe. UTS ultimate tensile strength YS yield strength of the pipe material. T time of inspection usually 10 years. LT axial length of the corrosion defect dT the depth of corrosion. The rupture stage from aw stage is identied when the nominal wall thickness is 0:55 t:

3:10

192

Fig. 3. E/C rate vs. ow velocity.

2 2 2 2 srate f =T 2 sT f =pH2 spH f =U 2 sU 2 f =d2 sd

Fig. 5. E/C rate vs. pH.

4 :1

It has been assumed that all the process parameters are normally distributed. The developed model can be simulated to get optimum design parameters, by considering the process variables of interest as the xed parameters and adjusting the others. Table 1 presents the mean and variance value calculated for the corrosion rate depending on the mean and variance of specic parameters. Figs. 3 6 present the variation of erosion-corrosion rate with parameters such as ow velocity, temperature, PH and diameter, respectively. 4.2. Piping failure probability estimation After estimating the corrosion rate, it has to be applied in the suitable limit state function to estimate the failure probability. Table 2 presents mean and variance values for various parameters appearing in the limit state functions. The software package for structural reliability analysis, STUREL, has been used to estimate the failure probabilities
Fig. 6. E/C rate vs. diameter.

from the limit state functions. The solutions obtained from COMREL module of STUREL are used to estimate the various transition rates, which are presented in Table 3. These transition rates are applied on Markov model shown in Fig. 2. Software MKV 3.0 [9] is used for determining the various state probabilities in the Markov model, as shown in Table 4. Modied B31G estimates are considered for rf and rl in Markov model. Depending on our denition of failure, state probability of either the leak state or the rupture state, can be considered as failure probability of the feeder. The failure frequency of
Table 2 Parameters for failure pressure model with mean and variance Parameters Yield strength (MPa) Thickness of the pipe (mm) Ultimate tensile strength (MPa) Outer diameter of the pipe (mm) Rate of erosion corrosion (mm/year) Load (MPa) Time (year) Length of defect (mm) Mean values 358 7 455 72 0.051 8.7 40 300 Variance 25 0.148 32 1.5 0.015 0.9

Fig. 4. E/C rate vs. temperature.

193 Table 3 Transition rates obtained from COMREL modules Parameters Values (/year) 3.812 1024 2.435 1025 0.115 1027 0.112 1026 1.486 1022 8.77 1022 LSF method LSF-1 LSF-2 LSF-3: modied B31G LSF-3: Shell-92 LSF-3: modied B31G LSF-3: Shell-92

f lf rf rl

parameters are assumed here to be normally distributed, but in actual practice this may not be the case. Nevertheless, the COMREL module has the facility to account for any kind of distribution. Instead of applying directly the probabilities obtained from limit state function in RI-ISI evaluation, it is recommended to nd the state probabilities using MARKOV model, since it incorporates the effect of repair and inspection works in the pipeline failure frequency. Markov model also allows formulating a proper inspection program and period depending on the operating condition of the plant at any given time.

Table 4 State probabilities from MKV 3.0 States Success (S) Flaw (F) Leak (L) Rupture (R) State probability 0.9956 4.362 1023 9.303 1027 3.147 1027

Acknowledgements The authors wish to thank the reviewers for their critical review and constructive suggestions to improve the quality and readability of this paper.

the feeder can be estimated by dividing this probability by the design life of the component, which value can be further employed in RI-ISI for determining its inspection category [7] for In-Service Inspection.

References
[1] Burnill KA, Chelugel EL. Corrosion of CANDU outlet feeder pipe. AECL 11965 1999. [2] Caleyo F. A study on reliability assessment methodology for pipelines with active corrosion defects. Int J Pressure Vessels Piping 2002;79:77. [3] Fleming KN, Gosselin S, Mitman J. Application of markov models and service data to evaluate the inuence of inspection on pipe rupture frequencies. Proc ASME Pressure Vessels Piping Conf, Boston, August 15 1999. [4] Fleming KN, Mitman J. Quantitative assessment of a risk informed inspection strategy for BWR weld overlays. Proceedings of ICONE 8, Baltimore, MD; April 26, 2000. [5] Gosselin, SR, Fleming KN. Evaluation of pipe failure potential via degradation mechanism assessment. Proceedings of ICONE 5, Fifth International Conference on Nuclear Engineering, Nice, France; May 2630, 1997. [6] Abdulsalam M, Stanley JT. Steady-state model for erosioncorrosion of feed water piping. Corrosion 1992;48:587. [7] TR-112657, Revision B-A, EPRI Revised Risk-Informed In-Service Inspection Evaluation Procedure; December 1999. [8] STRUREL, www.strurel.de, licensed software for Structural Reliability Analysis. [9] MKV 3.0, www.isograph.com, demo version for Markov model solutions.

5. Conclusions The paper has considered the Abdulsalam model for estimation of erosion-corrosion rate. However, these estimates should be veried against operating experience, if available, before employing in such application. If the reduction in pipeline safety is assumed for long elapsed time, then special care must be taken in characterizing accurately the coefcients of variation of the load and resistance parameters. The following priority scheme must be used for determining the actual coefcient of variation: rate of corrosion, thickness of the pipe, operating pressure, material yield strength, and pipeline diameter. The sensitivity of the failure frequencies increases with increased pipeline elapsed life. The failure pressure models considered here to dene the LSF lead to similar failure probabilities for short pipeline service periods. Various