Name: Corrosion control: predictive power play with Monte Carlo simulation
Start: 2024-12-04T11:00:00.000-08:00
End: 2024-12-04T11:34:00.000-08:00

Hello, everyone, and welcome to today's webinar on corrosion control. We're thrilled to have you with us and whether you're joining from home, the office, or anywhere else in the world, we appreciate you taking the time to join us for this session. My name is Emma Becha, and I'm the Technical Sales Manager here at Wood and I'll be facilitating this session. Before we get started, just few things to note. Firstly, this is a pre recorded presentation but members of the team are on the call so we'll be able to answer your questions as we go through this. Secondly, if you'd like to receive a certificate. If you'd like to receive a certificate of attendance, there is a minimum time requirement as well as a quiz at the end, so please do fill that out if you'd like the certificate. Also, I'd like to thank everyone who's already submitted questions during the registration process. We'll be addressing some of these during the webinar, but also if we haven't, we don't address them. We will be following up afterwards. And the last point, if you'd like to hear more about what we talk about today, you can visit our website and you can download our free trial of the of ECE software and you can also contact one of us directly. So we'll begin today's webinar with a brief safety moment followed by an introduction to intelligent assets, which is a part of Woods business where ECE has been developed. After that, I'll introduce our speakers who will be taking us through a more detailed overview of ECE software and show the new Monte Carlo simulation modelling with a live demo. We will then wrap up with some questions. Atwood One of our values is to ensure everyone gets home safely. The environments we work in can be harsh and dangerous, so I wanted to bring one very relevant incident to your attention. During the filling of the LPG sphere, the legs collapsed whilst at 80% capacity. This resulted in one fatality and another serious injury. An investigation found the last hydrostatic test was conducted 10 years prior to the incident and the last inspection of the legs was conducted five years prior to the incident. It was identified that the primary cause of failure was fear corrosion of the legs beneath the concrete, Fire Protection and this had resulted from water ingress. All the protective cap that was covering the concrete was inadequate to prevent water from entering. The key point underscoring the entire incident was that the failure to recognize the legs of the sphere had begun to be corroded. This oversight ultimately led to the collapse and really highlights the importance of testing and inspection. Now today, we're here to talk about corrosion control and the Monte Carlo simulations in the context of ECE, which is one of Ward's digital products. For those of you who are not familiar with WOOD, we are a global leader in consulting and engineering our core values, our care commitment and courage, and we are united by a shared purpose to unlock solutions to critical challenges. As you may already know, WOOD has three core areas of business, projects, operations and consulting. This webinar is brought to you by our Intelligent Assets team, which is a part of the consulting business. Within Intelligent Assets, we have a portfolio of 10 different products which are used throughout the life cycle of the energy industry, right from the design and commissioning with our Go Technology software all the way to operations and maintenance with our Virtuoso Digital Twin and Maint AI digital solution. ECE is one of the products in our portfolio and we have clients using ECE on projects worth billions of dollars. However, it's also used by academics and subject matter experts. It's a very versatile tool and you'll hear more and see more about this during the webinar. So without further ado, I'd like to introduce you to my colleagues who will guide us through the remainder of the webinar. And I'll be back at the end to wrap things up. First, we'll hear from Johann Henrickson, our Product Manager, who will give a deeper introduction to ECE. After that, Johann will pass the floor over to Andrew SIM, our specialist Services technical manager, who will talk you through the Monte Carlo simulation method and demonstrate it in action. So, Johann, over to you. Thank you very much for that introduction, Emma. So this brings us into to the topic of today, corrosion modeling or rather corrosion modeling software with a focus on our ECE software. So if we look at the life of an asset and we look at when corrosion modeling implies, of course, the first thing that comes to mind is the design stage where we want to understand how we can create the asset to make sure that it lasts for its its expected life. So this is where we need to look at expected conditions during life of the assets, precious temperatures, flow rates, but also of course fluid compositions and the existence of corrosive species. So with that, we can then start to look at what's the right material for the job and also how much wall layer do we need to have to get to the right corrosion allowance to ensure that the asset will last through its lifespan without leaks. All of that of course leading up to that, once we go into the execute state, putting steel to the ground, we know that we have the right material for the job, but we also have the right amount. We haven't over ordered or we haven't under ordered. So with that you can then say is this the end of corrosion modeling for an asset now? Because if we go into operations, then a lot of the parameters that determine the corrosion rate, they are actually measured. We have measurements of precious temperatures, composition, flow rate, so on and so forth. And all of that data is gathered in data platforms or historians. And what we see on the screen right here is dashboards where we have connected ECE to this data, pulling it into ECE and performing live corrosion assessments. So determining the corrosion rates at various points in our system right now. With that, of course, we can then also accumulate this to determine the accumulated war loss over time to see how much of the corrosion allowance is left. Of course, with this, we can also see if we encounter intolerable conditions, having excessive corrosion rate at some point and understand its impacts, do we need to take mitigative actions here and now? Do we need to do proactive maintenance on this one to avoid lease further down the line? Now if we look at this and we look at corrosion modeling or most physical modeling properties depend on the large number of parameters and it's very difficult to assess them exhaustively. And This is why we have introduced Monte Carlo simulations into EC. And this is not to gamble with our corrosion assessments. It's not to go to the casino and put it all on red and cross our fingers. Now it's a means to ensure that we have the right design without over design. Because if we go into physics and we look at a lot of properties, it's impossible to exhaustively evaluate them throughout the domain of validity. If we use this surface plot here on the left hand side as an example as a representation, we can create a high fidelity map of it here with looking at this heat map, high resolution yellow indicating the peaks, blue indicating the cross. But to do this is infeasible for most of the cases we are looking at. So when you come to that, what you do, you look at the parameters, you select parameter values and you can create all possible combinations as we see here on the right hand side. So each dot here represents a simulation and we have created a granular representation on the heat. We see all the details, but not at a high resolution. This works well at lower dimensions, but as the number of parameters grow, which it does rapidly coming up to 5 or 10 parameters, it just grows exponential and it's impossible to carry out the number of simulations required. So this is where we come into the work which was pioneered by Ulam von Neumann and others in the 1940s, which is Monte Carlo simulations. So their idea here was let's look at the randomized set of parameter values. So let's consider that an experiment, a numerical experiment, and then we carry out these experiments and see what we get. So if we look here on the upper panel, this is 1000 experiments randomly selected over the domain and we see that we get an irregular representation of the heat map we previously saw. If we think this resolution is too low, we can add more experiments and gain better certainty and better resolution for this. But this one also allows us to exploit properties of the parameters. If we recall the heat map on the previous slide, we saw that not much was happening at the bottom and the top of that plot. And we can then make sure that we skew the selection of parameter values to areas of interest. And we see here, we have applied at the lower panel a normal distribution, making sure that the selection in the Y direction is centered around the middle and we get a much better resolution. All of this while still looking at outliers, ensuring that we're not missing out on anything. Now, with thousands and thousands of experience at hand here, how do we analyze it? Do we take the worst value and add a market? Well, that's one way of doing it and you design me to the worst case. But most of the time that means that you will over design your system. So to avoid that, you can do a probabilistic approach to analyzing this. You look at all your experiments and you create a frequency analysis, creating a histogram as you see here on the right hand side. So basically looking at the number of experiments between minus .75 and -.7, the number between -.7, -, .65, so on and so forth. And once you've done that, you can go in and you can create the cumulative probability distribution as illustrated in the red. So if you take a number here, 75% for example, we go in and find that value on the curve here, also known as the P-75 value. We can go down to the X axis and check what value does this correspond to and it's around 0.45. The interpretation of this is that this is 75% probability that the property we have investigated will be at 0.45 or lower. So meaning a 25% probability that we will exceed this value. So this is how we can look at this and we can determine the level of confidence we want to have that we are not going to exceed the value 75 percent, 99% all depending on the properties we are looking at. And by that, I will hand it over to Andrew to bring us through how this works for corrosion models. Thanks very much, Johan. So Johan has given an overview of Monte Carlo simulation in general, and now I'm going to show how we can apply it in particular to corrosion modeling. So what do we need to perform a Monte Carlo corrosion simulation? Essentially, we need two things. We need a corrosion model with built in Monte Carlo support and we need an important data set that has variability. So for this, we're going to use the electronic corrosion engineer ECE model, which since version 5.8 release has supported Monte Carlo functionality and the flow line corrosion predictor. And the variability in our data set is either going to come from the design stage, multiple design cases to consider, or the operational stage is going to come from data logs of the various parameters that go into corrosion modelling. So considering the data sets that we typically receive for corrosion studies, which of these, how can we categorize these inputs and which of those would have variability? And which of them are fixed single values? And if we're considering the modelling of a pipeline, then the fixed values are typically the physical parameters. And so the diameter, the wall thickness, the pipeline length, the grade of carbon steel, the heat transfer factor, while there might be some manufacturing tolerances for the purposes of corrosion modelling, we consider these to be single fixed values. Now the variability in our data is going to come from the operational parameters. So the pressure, the temperature, the flow rates of oil, water and gas, the fluid composition, CO2H2S content and water chemistry, in particular corrosion inhibitor availability, these parameters are all going to have significant variation. So once we've identified the parameters that have variation, we now need to look at the types of probability distributions that we could use to describe them. And ECE supports the following four probability distributions, which we can see on the screen here. On the left hand side, we have the simplest 1, so uniform here has a defined minimum value, a defined maximum value, and we say there's an equal probability that any single Monte Carlo iteration would select a value from anywhere within that range, and that's the simplest distribution we can have. If we move down to triangular, it's slightly more complicated. Again, we have a defined minimum value, a defined maximum value, but the probability that we would select a value from within that range increases linearly from the minimum and the maximum values to reach a maximum area of maximum probability at the mode position in the center of our range here. And if we look on the right hand side, we have more complicated mobility distributions, the normal and the log normal distributions. Now, these distributions are not so easily bounded within a minimum and maximum boundary. Rather, the probability reduces as the distribution approaches the access, but it never actually reaches it. So in EC we have to artificially bound these distributions to fit our range. This means that there's a small probability that simulation may be run outside of the entered range. So we need to be aware of this obviously, because if he uses them to the specific range and then they find that the simulations are being run outside of that range, it would be a bit confusing. But also we need to prevent the model from running simulations outside of the validity range for the model itself. So ECE handles those eventualities. So then how do we go about selecting the correct probability distribution to describe our data? So one way about of doing it we've got on the left hand side here is if we've got a large lots of data, which likely this would come from an operational study where we've got a large data set to consider. We can use commercially available probability distribution fitting software that will analyse our data and apply the best fit probability distribution to describe our data. And as an output of value, you'll simply be told which is the correct probability distribution of which of the parameters to describe that distribution, and you can enter those directly into the C software. Alternatively, if you don't have that software available, or perhaps it's a design stage study that only has four or five discrete data points that are going to fit very well, then we can use the following rules of thumb. We can use a uniform distribution for pure sensitivity analysis, so this would work well if we've got been provided with a range of parameters but we have no further guidance. We have no idea of where the typical value is, for example, so we just select our minimum and our maximum value where we have a good idea where the typical value is, but we need to account for variabilities both above and below that value which have a lower probability of occurrence than the normal distribution. Works really well. Or alternatively the triangular distribution is a good fit to describe the behaviour where this manual intervention involved. So for example, corrosion inhibitor availability, which can go from 100% to 0% over very short periods of time is described quite well by the triangular distribution. So now we're going to run through a worked example with our ECE software performing Monte Carlo simulation. So this is our case study. We've got a four kilometre pipeline constructed of carbon steel, 6 inch outer diameter. We're looking at an operating line of life of five years. Corrosion allowance is supposed to be 6 millimetres and we've got our variables on the right hand side here. So our inlet and outlet temperature, inlet and outlet pressure provided with minimum and maximum values for these, but no specific guidance on what the typical value would be. For our CO2. We do have a typical value, but we need to consider the what these upper and lower values around that typical value. Chloride, we only have a single value of chloride to consider. So we're not going to look at any variability on chloride bicarbonate, we've got a typical value bicarbonate. But again, we want to also consider what would happen if we have some higher and lower values then our oil and our water. We have a range of these values with no typical value and again a corrosion inhibitor, we've got a range there, but no typical value providing. So now I'm just going to share screen into our software. OK, So we're in our software and we've pre populated it with our typical values of our physical promises. On the left hand side, we're going to the conditions tab. We've can see that we selected the temperature and the pressure minimum and maximum values as described in that table. We don't know the typical values in this case. So we selected a uniform distribution for each of those CO2 where we did know that we had a typical value which was 1.7. But we also have these higher and lower cases to consider. We're selecting a normal distribution which is centered around that 1.7 value. We have no H2S in our example, so we just leave those both as zero. And in that case, it doesn't matter what distribution we select. Our water chemistry, we've got this 150,000 PPM chloride value and we're not going to look at any variability on that value. So our minimum, our maximum, both size equal, doesn't matter what the probability distribution is in that case. Bicarbonate, we had our typical value of 300 and we need to look at these lower probability 200 and 400 PPM values and we're going to look at normal distribution for that. If we go to our Freeport tab, we've got our variability, our oil flow rate, our water flow rate, both with a uniform distribution because we've got no idea on the typical value and there's no gas present. So the gas is it right advanced going to set initially going to set the inhibition to non. So I'm going to look at the absence of inhibitor. So if I now run an initial 100 simulations and we can see what happens. So that quickly populates the graph and we can see we've got our P25 green line here, P50 light blue, P-75 blue all the way up to our P99 value. So 100 simulations is good for a quick answer, but if we want something with a bit more accuracy and a bit smoother graphs, we need to move up to 1000 simulations. So I'll quickly run that. So that's running. Now obviously you'll notice it's taking somewhat longer, simply 10 times longer, because we're running 10 times as many simulations as before. And we do have to remember that ECE being a pipeline model, it's not only running 1000 calculations, we actually simulate 1000 segments within the pipeline. So we're we're running 1,000,000 corrosion calculations now as the bar passes across the screen. So that is just about finished now. So you can see that our lines did move around a little bit. It's looking a lot smoother now. There was some adjustment in those positions of the P numbers, but approximately it's roughly the same answer from our P number that P 25 is about nought .82mm a year going up to our P99 of about 1.7mm a year. So if we just drop back into the presentation now and we look at a summary of those results, so this is our only inhibited table and we've calculated total service loss now from those corrosion rates based on our five year design life. So what is that saying is that our P25 value here is our value where we say with 25% confidence our total service loss will not exceed 4mm. So that's quite a low level of confidence if we increase to our P90 value here. So we're saying with 90% confidence now our corrosion rate will not exceed, our corrosion loss will not exceed 7.4mm all the way up to our P99 value here, where we can say 99% confidence that our total service loss will not exceed 8.35mm. So as we increase our level of confidence in the answer, our total service loss is increasing. So what does that mean in terms of a 6mm corrosion allowance? Well, 6mm fits just between our P50 and our P-75 value. So we can say we only have 50% confidence in this case that a six millimetre corrosion allowance would reach the required design. Five year design, which most people would consider 5050 is not acceptable confidence in this design. But that's OK because this was an uninhibited case and now we want to look at what benefit corrosion inhibitor could have. So if we go back to our software. So we're back in the software now and we go back to the Advanced tab and we're going to select an efficiency corrosion inhibitor model. We've got the inhibitor availability as described in the table with minimum of 50 to a maximum of 90. We're going to use the triangular model distribution as suggested. Go back to the project now and we'll rerun that with 1000 simulations, OK. So obviously that's going to take a little bit of time. We do expect that with the presence of a corrosion inhibitor that our corrosion rates are going to drop. So we're currently we're seeing range from about naughty .8 to about 1.7. So applying a corrosion inhibitor with 90% efficiency and looking at a range of availability from 50 to 90%, which is actually realistic. I think in my experience, we should see those rates drop now. So now we can see that didn't make quite a big impact. So our P25 values there about nought .31 millimeters a year going up to P99 value is about nought .8mm a year. So let's go back now look at those results. So now this is our inhibited tables. So if we recalculate our service losses now total service loss based on those numbers from we've got P25 number of 1.55mm all the way up to our P99 number which is 3.95mm. So actually we can say now with that relatively low, but I would say realistic level of inhibition, we can have 99% confidence that a 6mm corrosion allowance would reach it's five year design life. In fact, we could go further than this. We could say we could reduce the corrosion allowance with 99% confidence. We could drop the corrosion allowance to 4mm, or if we're willing to accept a lower level of confidence in our answer, if we're willing to accept a 90% level of confidence in our answer, we could drop the corrosion allowance to 3.2mm. So the question now is a little bit different. So what else can we do with the model? So the model is calculating all the other parameters in the background as percentiles as well that are required to calculate the corrosion result. And these can be useful in themselves. So for example, here we've got shear stress. So shear stress is very useful if you're looking to specify a corrosion inhibitor that has been tested to a certain shear stress level or if you need to define the parameters of the test program perhaps. And you need to understand what are the range of shear stresses that will be applicable to your corrosion inhibitor or what, what is the probability that certain shear stresses could be reached pH if a full spread of the potential pH range. This can be very important if you've got alloys selected somewhere within the system and you need to know what ranges of pH windows they're tolerated too. In addition, we also that all the other parameters such as like temperature, pressure, all the flow velocities will also be calculated and be available as percentiles. So to summarize. The. ECE Monte Carlo implementation allows the user to analyse data sets, large data sets with more variation faster than traditional sensitivity analysis techniques will take you. The model has been designed to be simple and easy to use by corrosion engineers. You don't necessarily have to have an in depth understanding of Monte Carlo mathematics. It does now mean that rather than the traditional perhaps designing to the worst case scenario, you would have to consider a level of risk or confidence in your answer, which is a new question for corrosion engineers to consider. So I'd just like to highlight now if you're interested in the Monte Carlo simulation tool I've been presented today, you can go to our website, fill in a form and you can download a trial version which will install on your computer for seven days. And you can play, play at home with that at your leisure. So I'm just going to hand back to Johan now to go through the question and answer session. Right. Thank you very much for that presentation, Andrew. And I see that we're coming up here on the 30 minute mark now, but I hope that you'll have time to spend a couple of more minutes with us and we'll address just a few questions here before handing it back to Emma to to wrap up the session. So the first question here for you, Andrew, is to generate a normal distribution and mean value at standard deviation are required to describe the shape of the bell curve. How are these determined as they did not appear to be entered during the demo? Yeah. So the the mean is just simply calculated from the minimum and the maximum values entered. And the standard deviation uses the what's called the 689599.7 statistical rule, where the standard deviation of distribution is equal to three times the mean value. We've done that just to keep it easy to use and simple software in terms of just the parameters that are required to be entered, but there's absolutely no reason why in the future we couldn't allow the user to enter those variables directly in the future update. OK, good. Next one here is How can large quantities of raw operational data be quickly reviewed and entered using the Monte Carlo model to reduce the overall number of individual simulations normally required. Yeah. I mean, one way to do it depends on the time period, but you could break down the data sets quite easily into perhaps like quarterly chunks and then just review them and identify what is the minimum and maximum value in that range. And then just add to them for each parameter. Or into EC Monte Carlo simulation that way. And then use that to build up percentile corrosion rate predictions for each quarter, which you could then convert into from the cumulative percentiles of corrosion allowance loss. Good. And I see here I think we take one last question before handing it back to him. And this is 1. I think many are asking how many iterations should be performed. Yeah. So obviously the more iterations you do, the longer it takes. You saw on the demo there the 100 iterations and the 1000 iterations took somewhat longer. You see, supports up to 100,000 iterations, but we're not going to show that in a live demo. And 100 iterations is good for updating the graphing screen. It does it pretty much instantaneously. So if you change the parameter and you want to see instantly what effect that would have roughly, you can use 100 iterations, but it's only considered accurate at around P-75 level. And the answers you get from that 1000 iterations takes some that little bit longer as you saw about 30 seconds. And that gives you accuracy up to around P-95. But if you really want the P99 accuracy, you're looking at 10,000 iterations, which is going to take some few more minutes with shown today. OK. Thank you very much for that, Andrew. And by that, we'll hand it back to you, Emma, to to wrap it up. Great. Thanks, Johan, and thanks Andrew and thank you everyone for joining and for all of your questions. As myself and Andrew have already mentioned, if you have any questions or if you would like to get more info on ECE, you can contact us. And you can also get access to a free trial version of the software. And, and if you follow QR code that's on the screen that will actually take you straight there. So it's nice and simple for you guys. Before you go, if you would like a certificate of attendance, please complete the quiz at the end. And we'd also really appreciate your feedback. So if you have a moment at the end to complete that, that would be really valuable to us. We do take that feedback on so we know how to improve them going forward. Lastly, thank you and goodbye. Bye. Thank you very much.

Inaccurate corrosion predictions can lead to pipeline failures, environmental damage and expensive downtime.

The problem

The solution

Who should attend

Certificate

Questions?

Other sessions