Creating an Open Source Drilling Community

Good morning, good afternoon, good evening. I'm Paul Pastusek. I'm a drilling mechanics advisor in the Wells Technical Organization at ExxonMobil. With me today is Dr. Greg Payette, one of our researchers in the Upstream Research Company. We'd like to talk to you about the efforts that have gone into creating an open source drilling community for the industry.

We'll go through the objectives of this effort, how we got started, the Steering Committee. And then for those that are not involved in drilling, I'll go through a brief explanation of the drilling process. Greg will then talk about the importance of modeling the current issues, and he'll talk about the benefits of open source and the validation process. And I'll end up talking about how to join the effort in contributing your time and talent.

The objective of this effort is to form a coalition of industry and academic leaders to encourage the use and re-use of continuous improving world class models. Our key efforts have been to set up a repository for source code to identify benchmarks and documentation requirements. We have also wanted to document good coding practices, encourage a number of subject matter experts to review the models and the data submitted, to test, use, and improve the code that's submitted, and to collect drilling data for validation. And if that data doesn't exist for a particular validation, need to request that data from the industry. Last, we'd like to attract talent to the effort and mentor those getting started.

This really started in 2018 with a question at the Deep Drilling Colloquium that Greg and I posed. And the question that we asked was, is it time for us to start using open source models in the industry? The answer was a resounding, yes. We subsequently wrote a paper with about 14 authors supporting the effort and outlining what it takes to create an open source community.

We chartered this effort under the SPE drilling systems automation technical section as an open source subcommittee, and that was in 2020. However, 2020 did not work out the way we expected to, and it was really 2021 before we actually got started in a serious way, holding bi-weekly meetings with the Steering Committee. We had a soft launch of this year in September.

ExxonMobil challenged the industry and gathered a coalition of experts to get started in this effort. We've identified over 160 people, subject matter experts that have contributed models and explained a number of the drilling physics that are going on that are-- many of them are joining this effort now. The University of Calgary-- Roman Shor at the University of Calgary is hosting the website and GitHub sites, and models to date have been submitted by University of Calgary, Texas A&M, Scientific Drilling, NORCE, and ExxonMobil. And we have a number of other models that are coming this way.

The ExxonMobil model is an axial torsional friction drill string model, lumped mass parameter model. We've worked with MathWorks to improve the use, and interface, and modularity, converting it to Simulink We've been optimizing the execution speed and stability of that model with their help. And we're currently documenting how to set this up and to use this model and the other models that have been submitted.

For those of you that are not involved in the industry, I mean, when we talk about drilling, you may think about that mast in the center of the screen as the drill rig. That's just what's going on at the surface. The actual drilling process can go on several miles down hole under the ground and under the ocean. There may be 150 to 160 people on an offshore rig, supply boats, helicopters, ferrying people back and forth and equipment. That operation can easily exceed a million dollars a day. Therefore, it's imperative for us to reduce the amount of time used in drilling and reduce any dysfunctions that may occur.

A typical land rig is shown on the left and the drill bit in the center. The video is of a PDC drill bit cutting rock and air with the side of the rock removed so you can see the process of cutting. In order to understand the system, we need to be able to model the cutting process, that's the loads on the individual cutters, and the drilling process as the lower boundary condition. In addition to the lower boundary condition for drilling at the drill bit, we need to be able to model the control systems at the surface as they're the upper boundary condition.

Along the way, we need to be able to model both the function and the dysfunctions that can occur of the drawstring. So several dysfunctions I wanted to show you are indicated here with these videos and animations. This indicator that you see in the center of the screen is a weight on bit indicator, and it is fluctuating between 10,000 and 100,000 pounds weight on the bit. That is not an acceptable range of vibration. It's damaging to the tools and the surface equipment. And it's indicated by the blue arrows as an axial motion of a bit bounce.

We also have a torsional oscillation, a low frequency torsion and high frequency torsional oscillation. What's shown here is called stick slip. The blue is the surface equipment and it's showing the top drive rotating at a constant speed, and the candy stripes on the drill string show the torque that's wound up and unwinds as the bit stops and starts, or sticks and slips. This can be very damaging to downhole BHAs as well as overloading the top drive and causing dysfunction all along the drill string.

Another dysfunction that we don't like is forward whirl. This is a drill coder in the master in the drill rig and you see that it has what looks like a bend to it. It is going through a jump rope motion, and that can occur anywhere along the drill string all the way down to the BHA. That puts extra bending stresses on all the parts, takes off extra torque, and scuffs off energy that we don't like.

It can get severe enough if that OD surface interacts with the borehole, if you have a high enough friction, you can go into what's called reverse whirl. And that's where you walk around the inside of the borehole with the pipe. Again, that can occur anywhere along the drill string. It is extremely damaging to the tools and the equipment. And as I said, both forward whirl and reverse whirl can occur anywhere.

This next video is of a drill bit. It looks like it's drilling smoothly, but if you notice, there's water splashed on one side of the hole. That bit is pushed off center and it means that the cutters on one side of the bit are loaded more heavily than the other. That's forward whirl, or a jump rope motion occurring at the drill bit itself. And one of our most damaging vibration modes is reverse whirl at the drill bit.

If you look, this is the gearing motion of the drill bit grab grabbing a hold of the borehole wall and it's walking around the inside, or gearing around the inside. A good model for this is the toys you had as a kid is a spirograph where you're walking around the inside of the gear with the prior gear. So when we're modeling the drilling process, we need to be able to model the normal function of everything from the bit all the way up to the control systems at the surface, and we need to be able to model and account for not only normal functions, but all of these dysfunctions as well.

Thanks, Paul. Hi, everyone. My name is Greg Payette. I'm also with ExxonMobil. I'm in our upstream research organization, primarily work in drilling. And so I get involved a fair amount in some of the more advanced modeling work that we do. So why modeling? Paul mentioned normal drilling activity, normal drilling behavior and also the dysfunction.

Certainly when we select rigs and also designer our well construction programs, we're going to use modeling tools to ensure that equipment is up to capacity, that equipment can basically operate in the loading environments that we expect to see. And so modeling can be used on a number of different levels. It can be used in very detailed ways to study drill bits, drill bit design.

It can be used for understanding directional drilling equipment, and it can also be used for modeling more holistically the drilling assembly. And so really modeling can come into play on many different length scales and time scales. And our perspective is we're interested in all of them. We're interested in having good models for evaluating the individual tools, but we also want to understand how those components integrate and how they behave when they're basically combined with the other components of the drilling assembly.

On the left here, the image is kind of a cartoonish image of a drilling assembly from the top drive all the way down to the bit. And so there's a number of different things here that we can model that really go all the way from that surface equipment down to the bit. In the middle here, we're showing you a simplified model of a drilling assembly. And in this case, this is a lumped parameter model. You can see that we've discretized the string using springs and damping-type elements, and then we've got some other assumptions down at the bit in the surface.

And our perspective is it's very important to be able to model and understand these different components at the component level, and then at the system level. On the right over here, this shows you an image from a paper of modeling work that was done by Detournay's group from the University of Minnesota looking at the impact of, basically, blades and cutter arrangement on the stability of a drilling assembly to torsional vibration.

The ultimate driver behind the need for modeling, really, is cost. And cost comes into play really in a major way through the failures that we see. When vibrations culminate in broken bits, destroyed MWD equipment, or in really bad cases, twists offs, all of those things end up being large cost to the operator and potentially other players at the rig site.

Unfortunately, one of the realities that we just have to live with is that we collect limited data measurements on our drilling rigs. We collect quite a bit of data at the surface, we collect some data in the BHA, but we really don't collect any data along the drill string, at least not typically. And so modeling can help us close the gap of understanding between what we actually measure and then what we can predict.

And so in terms of prediction, we often think of modeling as a way to predict behavior. Really what we want for modeling, though, is to understand. We want to understand the physics of the systems that we're going to run at an individual level and a system level so that we can then basically make decisions that can minimize the chance of those bad things from happening that we talked about earlier. And so our position, really, is we think it should be inconceivable that new tools are put in the ground, or new control systems are placed on rigs without full component testing at the numerical level, but also system testing as well.

What are some issues with current modeling efforts that might motivate the need or desire to set up an open source community? Well, first of all, I certainly want to mention that there are a number of really phenomenal commercial models that exist on the market today that are offered by a number of different vendors.

And at ExxonMobil, we even have some internal models that we use by our subject matter experts and also our drilling engineers. The problem that we run into sometimes with commercial models is that they're often opaque, in the sense that we may not know fully what is implemented within the models. And so that can be an issue. Academic models, there's a wide variety of those as well.

The issue with those often is that they're usually single use. Often the purpose of those models is to demonstrate capability and publish nice work in papers. But often those get created in different groups around the world and they're fragmented in that sense. Managing incremental improvements can be a challenge when you're working in the space where different groups maintain different models.

How can open source address some of the issues that we just talked about? Well, transparency is something that, by definition, comes with open source initiatives. Anyone can see the source code that gets published. Open source brings a huge potential for collaboration. Whereas a lot of modeling previously has been done in silos, we think that open source can allow different players from many different organizations to come together and work together on a bigger initiative.

Often the SMEs that are doing modeling work, they're often concerned about very individual components of the drilling assembly, the system. And we think that's OK. We think that's actually great. And we think that those folks are best suited to focus on developing those types of models. But what we would really like to see is kind of a common framework set up where once those models are developed they can be plugged into other models that can then integrate into a larger system.

Another benefit of open source is model benchmarking. We can bring transparency in performance of how different models actually behave, either from a prediction perspective, or even from a computational performance perspective, how quickly can solutions be made. We're moving into a world that is much more integrated, so that's becoming very, very important in the world. We think that the more open source our tools are, the easier it will be to integrate into cloud environments and those sorts of things.

Now all of this said, we still think that there's a place for proprietary code, and we don't want to rule that out. And in fact, one of the things that we want to enable with this initiative is to let people continue to develop proprietary models that they can use internally in conjunction with the work that comes out of this community.

One of the really important things that we're going to do as part of the open source effort is set up a process for verification and validation of code that is submitted to the effort. And so at a high level, this process is summarized here in the slide. Basically, the key three things to look at is ultimately we're concerned about natural phenomena. And in order to model that, we're going to create what we'd call physical models. And we arrive at those by basically applying physical principles and different assumptions.

Once we have the physical model, which is basically partial differential equations, we need to solve it. Unfortunately, those models can't be solved directly, so they have to be converted into numerical models, which are then implemented as algorithms in source code.

And then finally, really what we're after is using those models to make predictions. And then we need to assess how well those models actually predict reality. Verification and validation are basically the two critical steps in this process. Verification, the question that we ask, really, is how well does the numerical model solve the equations of the physical model? The goal here is to basically ensure that the numerical model represents the mathematical model. It's not-- the goal isn't to see how well the numerical model represents reality.

Once we do that check, then we're ready to do tests against natural phenomena. And so that's what validation is. For validation, that's where we ask, are our predictions useful? And for this step, really having good data is key. One of the things that we're going to be asking for is part of the open source effort is for donations of data sets that we can use in order to validate the models that are submitted.

I'd like to-- if you're interested, I would like to ask you to join this effort, just add your contact name to the mailing list under the contribute tab in the Open Source Drilling Community website. Again, we're sponsored by the SPE Drilling Systems Automation Technical Section as a subcommittee. In order to contribute and/or use these models, you do not need to be an SPE member. You do need to be a member of SPE DSATS to join the subcommittee.

University of Calgary, Roman Shor is coordinating these websites and GitHub sites, and we're publishing everything and ask you to publish everything under the MIT Open Source license. In summary, it basically states that all the modeling data are freely available for anyone to use and reuse, both in the academic world and in commercial use. There's no limitations of using source code other than to attribute the original source for the sections of code that you use from this site.

If you have models and data that you've already published through peer-reviewed journals, then submit the model to the GitHub site and we will expedite the public release of that to the community. If you have models that have not been peer reviewed, you can submit the model, the data, and documentation to the Open Source Drilling Community, get your organization's release of information form as well, and submit the model to the GitHub site.

The Steering Committee will set up a review plan. We do not endorse models, but we do help coordinate the amount of documentation and test cases that they go through so that potential users will know what the errors and limitations are of every model.

In addition to modeling and data, you can contribute your time and talent to this effort. You can contribute funding as well. There's no requirement to join this effort, in terms of a fee, but if you want to help the industry, you can set up a bilateral agreement between your organization and a research company and/or university to help solve some of the hard problems that we've identified in this industry. With that, we'd like to open this up for questions that you may have, and we certainly appreciate your time and attention today.