Predicting Pipeline Corrosion

Predicting Pipeline Corrosion

Dr. James D. Garber, Dr. Fred Farshad, Dr. James R. Reinhardt, Hui Li, and Kwei Meng Yap and Robert Winters

A model has been developed which can predict the corrosion rate in gas and oil flowlines and pipelines. The Windows-based program describes the physical and chemical conditions inside of a pipe. This model is capable of predicting the corrosion rate in systems containing CO2, H2S, organic acids and bacteria.

The model also predicts the occurrence of “top of the line” corrosion, wettability conditions and can describe flow dynamics in large diameter pipelines. A new addition to the model is a risk assessment component, which allows for integrity management of the system. A number of field tests were performed to show the utility of the model, and the results were very satisfactory.


Corrosion prediction model

In May 1999, the Corrosion Research Center at the University of Louisiana at Lafayette received a three-year Department of Energy grant to develop a flowline/pipeline corrosion prediction model. The objective was to produce a computer model that was capable of providing a physical description of a system and predict corrosion rates for 3-phase pipeline and gas pipeline networks. A beta version, Phase I, was released on July, 21, 2002. On January 26, 2003, the Phase II version was released with numerous technical and cosmetic changes.

On November 1, 2003, an industrial consortium was formed by several companies to develop a Phase III version. This project produced a new multislope flowline model and an oil pipeline model. It was released on January 14, 2005. The Corrosion Research Center then sponsored further work on the model and developed a Phase IV market version which, in addition to improving the physical description of slug flow, added additional elements to the model such as improved flow dynamics, wettability “ top of the line” corrosion, H2S effect and risk assessment.

All phases of the program were developed using Visual Basic 6.0 with Microsoft Access as the database. Figure 1 shows the 5 models which constitute the program. A description of the models follows.

Physical description. The first three models shown in Figure 1 give the physical description of a system. This includes temperature/pressure profile, phases present, and the flow dynamics at each point in the pipe. As is usually the case in CO2 corrosion, the flow regime is very critical to the prediction of corrosion rate. Figure 2 shows the various horizontal flow regimes that are described by the model. The system loops at this point until it converges on pressure. This usually takes two to three loops. The pressure difference for convergence can be as little as 0.1 psi. The flow dynamic model includes empirical corrosion rate prediction.

Ion, pH and scale profile. At this point the ion, pH and scale model calculates the chemical properties of water in the system. In a system containing condensed water, the pH is low and so is the ion concentration. It is possible to track the location of organic acids entering the system as well as the bicarbonates and other ions. If the system is in stratified flow and condensing, two pH values are reported, one pH for water at the top of the line and one for the bottom of the line. These values can differ by 1-2 units.

Corrosion rate profile. The final and most difficult part of the model is the determination of the corrosion rate. Using equations developed in previous models developed by the UL Lafayette Corrosion Research Center as well as from others, it has been possible to obtain accurate estimates of the corrosion rate at each point in the system. In addition to the more empirical models, a pitting corrosion model has also been included which estimates the theoretical pitting rate. This model requires that a water analysis is available. To provide the best estimate of the actual corrosion rate, an expert system has been developed which accounts for variables that can enhance or diminish the predicted corrosion rate values. The parameters that are considered are temperature, water wetting, % inhibition, scaling, and bacteria effect. In this fashion the model considers a multitude of variables before giving the user the final corrosion rate.


Risk assessment

Risk assessment was incorporated into the Phase IV model because it has become a topic of interest in the field of engineering in the past few years due to its usefulness in evaluating the life of various systems. Internal risk assessment is taken into account by this model.

Risk is defined as the probability of a failure (or frequency of failure) times the consequence of the failure. The consequence is usually ranked from 1 to 5, which correspondingly range from non-severe to catastrophic failure. Consequence is typically assessed in a qualitative fashion, which is an inconvenience for computer application. It is usually easy for companies to establish and is left for the users to determine.

General corrosion models predict corrosion rate on a deterministic basis; namely, all the input values are known and are fixed. However, in reality, each input will have some uncertainty associated with it because of the variation of production conditions and the environment. This variation can have a significant impact on the corrosion rate prediction. One way to solve this problem is to calculate the range of corrosion rates based on the whole range of input values. This process can be time consuming if many inputs are involved. Furthermore, not all of the inputs vary in a uniform manner; some of them may vary following normal distribution, and others may behave in a log-normal distribution.

Random number generators for uniform, normal and log-normal distribution were developed. By assigning a certain distribution type to each major variable in the pitting corrosion model, random numbers were generated according to the specified distribution. All of these effects were combined using Monte-Carlo simulation to give the probability of the resulting corrosion rate. Any number of iterations can be performed. However, the recommended number is 10,000 or higher. From these results the mean and standard deviation were calculated, which could then be useful in determining the probability of failure.


Pitting corrosion

There are a total of 20 parameters used as inputs to determine corrosion rate in the pitting model used in Phase IV. Variations of some of these have significant impact on the predicted corrosion rate, and others have only a minor impact. Attaching a random number generator to all these variables will be time consuming and unnecessary. In this work, major variables are distinguished from minor variables in terms of their influence on the corrosion rate, and only major variables are considered to be associated with certain distribution types. The minor variables will be evaluated on their input basis.

Specifying the typical range of each variable in the field, the corrosion rate is calculated by continuously changing one variable in the range with others fixed. The resulting maximum and minimum corrosion rates are then compared, giving the percent change of the corrosion rate for that particular variable. The effect of each variable on the corrosion rate is listed in Table 1. From Table 1, it can be seen that bicarbonate, temperature, CO2 mole fraction, pipe wall thickness, chloride, pressure and bulk iron concentration have significant effect on corrosion rate in the range assigned. They are therefore considered as major variables associated with the specific probability function.


Major variables

It is important to obtain knowledge on the distribution type for each major variable, since different types of distribution will have a different impact on the corrosion rate. Process data can provide the best basis for determining the range and type of distribution, but since it is not always available, some assumptions still have to be made. In this work, field data and assumptions from the literature are combined to determine the distribution type of major variables.

Based on of 11,838 water analyses from a major oil company, distributions have been found for the following variables:

• Alkalinity (Bicarbonate): log-normal

• pH: normal

• Chlorides: log-normal or normal.

The following criteria suggested by BP Technology regarding distribution type was used to help determine the rest of the variables. The distribution for various parameters in the Cassandra model was also taken into consideration. Based on the above criteria, the rest of major variables could be defined in terms of distribution type, as shown in Table 2.

After a number of tests were performed, corrosion rate distributions were found to consistently fit the Weibull distribution, which is a most frequently used function for failure analysis. A corrosion rate distribution of 20,000 iterations with assigned distribution type inputs is presented in Figure 3. In this case, the calculated mean corrosion rate was 20.7 mpy, and the standard deviation equals 9.54 mpy. The actual predicted corrosion rate was calculated to be 19.6 mpy by using the mean value for normal and log-normal type inputs and the average value between the lower and upper limit for uniform type inputs. Note that the predicted corrosion rate roughly equals the mean corrosion rate, and that the standard deviation is approximately 45% of the mean value. This ratio is the same as quoted by investigators of the Norsok and DeWaard Milliams models for local corrosion. This result is encouraging because it validates the random number generators and also the pitting model. In the Phase IV model, due to time constraints, the standard deviation was fixed at 45% of the predicted corrosion rate even though it will normally range from 35% to 55% depending on the input values.


Probability of failure

The DNV-RP-G101 standard mentions that the probability of failure is determined using a Weibull distribution which is described by a scale parameter (a) and shape parameter (b). The Weibull distribution is the most useful mathematical form for failure analysis. In this work, a and b are determined from mean and standard deviation with the help of published relationships, as shown in Equations 1 and 2:


(1)



(2)


Where mean & std= mean and standard deviation for corrosion rate distribution, respectively, G(x) = Gamma function, which is defined as:


(3)


The gamma function can also be calculated using a polynomial approximation of Equation 3, as shown in Equation 4:


(4)


By comparison, the error caused by the polynomial expression is less than 310-7 for 0<=x<=1 but much faster than the integration process using Equation 3. Thus, Equation 4 was adopted in this work. The gamma function for any real number can be solved by combining Equation 4 with the following equation:


(5)


Using the above equations, a and b can be solved numerically.

The probability that a failure will occur within DT years is given by the cumulative probability function of the Weibull distribution, W(CRmax, a, b) as shown in Equation 6:


(6)


where PoF = the probability that a failure will occur within DT years, CRmax= maximum allowable corrosion rate calculated by the following equation,


(7)


where tcurrent = current wall thickness, tmin = minimum allowable wall thickness (typically takes 0 for pitting corrosion, 2- 3 mm for uniform corrosion), DT = time in years, a, b = Weibull scale and shape parameters, respectively.

Note that the minimum allowable wall thickness depends on the particular operating condition. Typically, it can be calculated based on allowable pipe pressure before pipe rupture as suggested in DNV RP-F101.


Classification of risk assessment

With the knowledge of probability of failure, the risk category of the pipe can be identified based on the criteria proposed in DNV RP-G101, which is shown in Table 3. In the Phase IV model, the graph that is generated for risk assessment is placed under the “plot” button and is labeled as “risk assessment.” If “top of line” corrosion is present, risk assessment would be evaluated based on both the bottom and top of line corrosion rates, and both will show up on “select plots” in the new model.

Figure 4 shows one case of risk assessment generated by the new Phase IV model. It can be seen that without inhibitors, the pipe will reach the maximum risk category in 7 years, while the risk of the pipe is significantly reduced with the help of inhibitor of 80% effectiveness. The risk category stays at the minimum value up to 25 years, and then gradually increases to medium risk at the end of 30 years.


Conclusion

The Phase IV computer model contains a number of substantial improvements over previous versions. The physical description of large-diameter pipes has been improved with the modification of the Dukler flow regime maps. Within the flow regime area, the modeling of slug flow has been the most difficult flow regime to describe. A change in the calculation of the height of the liquid film has given reasonable slug lengths and liquid hold up values. The program no longer has problems with the negative slug lengths.

The accurate prediction of the wettability of a three-phase of gas-oil-water is important for the determination of the final corrosion rate. Using laboratory work performed at UL Lafayette in a stirred tank, and information from Ohio University on horizontal flow, it has been possible to establish the liquid velocity at which water is picked up by the oil phase. Since the Phase IV model gives the flowrates of all the fluid phase, the conditions, which allow pickup, can be found. The stirred tank information can provide knowledge as to the percent wettability of the system in this condition. This allows an appropriate adjustment to the corrosion rate to be made.

Describing “top of the line” corrosion in a condensing pipeline requires an accurate calculation of the phase equilibrium of the fluids in the system, as well as an accurate knowledge of the flow regime involved. The Phase IV model is capable of determining if the “top of the line” corrosion can occur and what the pH value is at the top and bottom of the line. By incorporating the effect of organic acids in the calculations it is possible to predict how seriously it would affect the pipe.

Depending on a variety of parameters, the presence of H2S in CO2 corrosion systems can have a variable effect on the corrosion rate. Experimental data has verified that initially the CO2 will dominate the corrosion process and small amount of H2S will contribute to an increase in corrosion. However, when the H2S species becomes dominant then there is a drop in the corrosion rate. Using the pitting model developed at UL Lafayette, it has been possible to model the effect of H2S on CO2 corrosion.

The final improvement made to the pipeline model was in the area of the internal risk assessment. The seven primary variables that affect the pitting corrosion model and their distribution type were identified in this work. It was found that the pitting model has a Weibull distribution, which is similar to other corrosion models, and its standard deviation is approximately 45% of the mean. The same standard deviation is quoted by the investigations of the Norsok and DeWaard Milliams models. The model is used to calculate the probability of the failure (PoF) and provide a risk classification using a 1-5 rating (with 1 the best) as a function of years of use. From this information, a plot of risk classification versus years to determine the expected time to failure can be developed. The effect of inhibitors on this classification can also be described.

The physical description changes in the Phase IV model on the flowline and pipeline has produced an improvement in the final predicted corrosion result. The new model predicted stratified flow cases closely except for two cases, which were at elevated temperature. Although annular flow values of corrosion rate varied widely, the model was able to closely describe the four cases in question. Slug flow condition was also highly variable but the new model showed a high level of accuracy. Overall, the improvements in the program have helped in achieving more accurate corrosion rate predictions. n


Acknowledgments

The authors wish to express their thanks for the support provided by the EETAP division of the U.S. Department of Energy, together with the various companies that have supported the model development. They are BakerPetrolite, Champion Technologies, Chevron, Coastal Chemical, Nalco, Shell and Williams. Based on a paper presented at the NACE CORROSION 2008 Conference & Expo, held in New Orleans, Louisiana, March 16-20, 2008.