Hello, my name is David Skaarsjoe. I'm originally from Sweden. I've been living and working in Brazil for the last 10 years of my life. I'm responsible for an engineering team that we have in Brazil in the company HMH. I will present today a case study related to the project that we did, but I will start with introducing the company. Then we go through the problem and the problem definition, the background, the process and methods that we used, and then what conclusion that we reached.

HMH is a fairly young company. It started in 2021, actually, but it comes from 125 years of legacy through many famous companies in different industries. You can say, for example, Baker Hughes, MHWirth, Aker Solutions, GE Oil and Gas, down to where actually most of our products are organized from [inaudible 00:00:53], MH, [inaudible 00:00:55]. We are delivering equipment to offshore drilling and also construction and mining industry. We have the full suit of equipment for drilling offshore, everything in a top-side package with top drives, which is basically the drilling machine, turning the bit, mud treatment systems, mud pumps, pressure control like BOPs and well-head connectors. Between the vessel and the sea floor where you have the BOP, you have a product called the Riser, and this presentation here will actually show products needed for that type of equipment.

A marine riser is basically a piping system used for deepwater offshore drilling. It is the conduit that connects the drilling vessel with the BOP at the sea floor. It consists of multiple different joints. They are normally 75 to 90 feet long, and they are assembled in a sequence until you meet the depth of about 10,000 feet, 3,000 meter. The complete system weighs in about 2,000 to 3,000 metric tons, so it's a quite big system. Each joint is composed by an array of different lines. In the center, you have the main line or main pipe, which is where you're actually running the drilling [inaudible 00:02:14] through and running tools down the well. You have hydraulic lines responsible for providing pressure to the subsea equipment, choke and kill lines that are used in a well control situation, and the booster line that is used to boost the mud flow up through the handler of the riser back to the rig. It also have female and male connectors on each side. In this presentation, I will focus on the main line, although all the lines were assessed in this investigation.

But the main line was the one that was most significant and also to keep the presentation within the limits.

The Burst load case was dominating for the main line, and that is what we have assessed. Typical diameter of the main line, for you to get the feeling, is about 21 inch in diameter and 0.75 to 1 inch in thickness. So the background here is that we did the project for a customer where we inspected 30 riser joints onshore. During the inspection process, we found signs of galvanic corrosion where we measured the material loss about 4 millimeter per year. All these defects were assessed based on calculations, but some were on the limits, although they were approved. The complete system of the riser consists of 95 joints. We had 65 joints offshore on the vessel. They were not available for inspection, and basically, the question arose, what are the condition on these 65 riser joints, and how can we make an engineering disposition or say something about the condition of those joints? The answer to that is basically statistics. That's when I went down this rabbit hole and started to analyze what we found on the 30 joints, how can it be applied for the 65 joints that we don't have access to.

The process here that we're using is fairly straightforward. We collect the data from the 30 joints we have access to. We run those screenings, we check for correlations, interactions, identify distributions. We made the DOE design here based on the deterministic calculations to establish a mathematical model. We used that model together with distributions to run Monte Carlo simulations and come up with a conclusion to the probabilities of non-compliance. The first step, we collected data through traditional methods, measurements, visual inspection, etc. We entered all the data together with operational data for each joint into a data table in JMP. We use Graph Builder to get a good overview. On this graph here, you can see that each riser joint is represented here. On the bottom X axis, you have the rotational position in degrees, and here you have the position along the pipe, and each dot here represents a certain corrosion damage with the depth, the size of the dot, and also the colors of the different lines on the riser. We execute the screening to check for correlations.

We used the correlation matrix, one-way ANOVA, nonparametric comparisons when the variance was not equal. We made a few findings. Some of the findings here are deterministic, and they are not related to the actual problem, but we had still some findings, that we had to treat in the simulations in a separate way.

But in general, the hypothesis was confirmed that the corrosion process is highly random. It's basically depending on things such as electrical contact that is very random, where a coating damage occurred and at what time it occurred. The distributions that we had to track, they basically came or were given from the code that we used for the calculation, which is the code from DNV or corroded pipelines, where you can assess corrosion damage on pipelines. Typically, variables that are variable is the wall thickness. We have the extension of the pit, how large the damage is, how much has been lost, and also proximity to other damages. This is something that came out to be very significant in the end, and you will see that further in the presentation. We identified the distributions using JMP and the distribution platform in JMP. CFD plots was a good tool to check the fit from the distribution with the empirical distribution. There were many different distributions checked and many different found, and we had to work very carefully to ensure conservatism in all the different steps here.

For example, we had this case here, which is the position of the corrosion damage, which is not the normal distribution you can see, but it has natural explanations to that because you have some areas of the riser as it's exposed to damage and some not.

We designed the DOE as the response here is basically the result of the deterministic calculation. We use the flexible fast filling design. It's made for optimizing the coverage of design space and minimizing the rounds, basically. The continuous factors that we use for the wall thickness inside the pit, which is basically the remaining wall thickness when the corrosion has taken some material, the largest damage extension, which is the size of the corrosion bit, and also the surrounding wall thickness around the pipe, which is unaffected. We define the factor levels using the probability density functions from the inspection the result. They were extended slightly to make sure that we could capture rare events and have a good reliability on that. Graph Builder, the density function was used to check the coverage, and that was a very good tool for me as an engineer to check areas where you have not good coverage and if these areas are critical or not. The response was, as I said, this is the results from deterministic calculation, and this is basically the Utilization of Allowed.

We run the DOE. It was running through all the deterministic calculations for all the different runs. Fit the Gaussian process model, which showed to give very good agreement with what we find in the calculations. The results was that the most influential main effect for the main pipe was the surrounding wall thickness. Damage extension came out as least influential. But the interaction between wall thickness in pit and damage extension was dominating the interaction with 16.5%. Of course, this interaction here we had to keep in the model as well because there is a lot of interactions going on here. The Monte Carlo simulation was done by copying the prediction formula from the Gaussian Process Model to the DOE table, generate the distributions for the different factors based the distribution found during the inspection, and that then gave us a Utilization of Allowed prediction for each of the damages that we can see here. We also added other things to this table here. One thing that was a highlight is that we simulated, based on a random integer distribution, a number.

We assigned a number between 1 and 285, which represents a joint. It's part of the investigation, how many pits do we have per joint? We found that an average of seven pits per joint is something that is very conservative. We use that and apply that to each pipe, basically.

When we run through this again, and we could plot the different pits on these 285 joints. You see down here you have each joint, and here you have the position of the damage along the pipe. You can see here that the effects of this binomial distribution is that you have clusters where you have a higher likelihood of having corrosion which is closer to each other. We run through all the individual damages using capability analysis, but we found that the probability of exceeding the Allowed Utilization for each of these individually is extremely low. That means that we need to consider when they are close to each other, you need to combine them according to the code, and that was what we did. When we investigated it, we found basically 177 cases. It's an example here, but 177 of these cases for the main line that had to be combined that didn't meet the requirement on spacing between the defects. These 177 cases were modeled again. They get a new extension, a new depth, and a new thickness around the pipe.

We use the same prediction formula to generate new probability, new distributions.

Those distributions, we use then capability analysis to check what is the likelihood of exceeding the utilization of 1, 1.1, 1.2, up to 1.5. By doing that, we could actually plot the graph that the probability of exceeding each of these is possible to print in a graph. The interesting thing here is in the next step, because when we extrapolate this to the complete system level, we have to go from an individual defect to combine the effect. We need to go to one joint because each joint have several pipes, and then we need to go to 65 joints. We use that by basically product rule based on independent events. But what's interesting here is that the utilization is linear against the pressure applied in the pipe. That means that we could transform, instead of having the utilization level on the X axis, we could put the pressure here and correlate that, build the probability on non-compliance on the Y axis. What we found was that the main pipe is completely dominating the probability of non-compliance down to about 3,950 in this range, PSI.

Below that, even if we reduce the main line pressure, we don't get a big effect on the probability of non-compliance.

We have to go and reduce the choke and kill line pressure. You see here the red line here is the choke and kill line pressure that when it is reduced by about 30%, you can see that the main line, when it is decreasing, the probability of the non-compliance is also decreasing in the logarithmic linear way. The conclusion is basically that we could use analysis in JMP and quantify the risk for the customer. By reducing the choke and kill line pressure with 10% and main line with 30%, the risks are acceptable. There was no operational impact for the customer because the planned operations had less pressure than these. We could, by doing this, allow for more time to deal with the problem, and that saved weeks of downtime for our customer.

Thank you.