When we think about artificial intelligence in oil and gas, it’s easy to think about the big heavy equipment, the rigs, the vehicles, but there’s a lot of ways to unlock the data of oil and gas from our equipment, and even from our back-office operations.
In this episode, we explore a handful of use-cases for unlocking the business value in oil and gas data. Our guest this week is Lorena Pelegrín. She’s the product manager for Iron Mountain’s insight product focused on the energy sector. She has a background in energy. Iron Mountain is a $4 billion multi-national storage company, who’s now working on digital storage and OCR in a number of industries, oil and gas, and energy is one of them.
We cover three distinct topics in this 30-minute interview:
- The uses of subsurface data
- AI for asset inspection, and
- The intersection of AI and digital twins
Listen to the full episode below, of skim our interview takeaways and the full transcript below:
Expertise: Mortgage processes and automation
Brief Recognition: Before joining Iron Mountain’s InSight product group, Lorena help positions as Principal Consultant at Restrata, focusing on safety in oil and gas, and held the role of Global Manager, Technical Safety and Risk Management at ILF Consulting Engineers, an oil and gas consulting firm based in Munich.
- Better Data Means Better Focus on Productivity. Not all oil wells have the same promise for delivering oil volumes. Whether they’re currently producing, or whether they’re still being explored – having more accurate data about different well sites allows firms to focus their resources on sites with the most upsite. This data also allows firms to more accurately make predictions about wells and production activity into the future.
- Compliance Drives Adoption. Regulatory activities change the way companies behave – and change the technologies that they adopt. As LIBOR fades away in the banking sector, companies are using AI to gain transparency on their legal documents in order to remain compliant – and in order to determine possible contractors that reference LIBOR (and now need amending). As regulation requires tighter monitoring of data from oil and gas pipelines, companies are likely to again use compliance as a prompt to action – giving them the transparency they need to comply with regulations and keep their operations running.
Full Interview Transcript
Daniel Faggella: So, Lorena, I want to be able to dive into these various use cases of how artificial intelligence can start to unlock the value of data in the oil and gas space, and make this really tangible. I know the first category we wanted to talk about was really around the value of subsurface data, that there’s a lot of subsurface data, obviously in the oil and oil and gas domain. Can you walk us through a little bit, what kind of data that is and how AI can help us get a little bit of value out of it?
Lorena Pelegrín: Yep. So as you pretty much have put it there, that we have a lot of different kinds of data, multiple document types, vast amounts of unstructured data there in many different formats. You have tapes, you have paper, we also have new digitally born data, that it also plays a role. There are multiple repositories, so you have different sources of truth there, that need to be orchestrated.
What this translates into across, pretty much across all subsurface data evaluations, all the use cases that we have in that cluster of use cases is really seeing an inability to find key data quickly. People spend most of the time really searching for data. This is a manual and time consuming and resource-intensive workflow. So what happens there is that decisions, evaluations and decisions, they take a lot longer. Also, the coverage of the analysis that the teams are doing is much reduced. You cover up to 20% of the data that you have. That also translates into reduced accuracy. So you’re seeing that teams are making decisions with high levels of uncertainty.
What we see is that if teams have access to digital versions, digital data, and to the metadata associated to those, really, processing time can be reduced dramatically. Time to decision can be reduced, and confidence on decisions associated with drilling in new places, with acquisitions considerations, asset divestiture, all those can be significantly improved.
Daniel Faggella: Yeah, so let’s talk a little bit about some of those decisions people have to make. So right now, getting the picture, and I’m sure it’s no surprise to most of the folks tuned in, that we’ve got some pretty stodgy sources of data. It’s a very old industry, relatively speaking, and we’ve got a very kludge, ugly kind of bundle of however much of it we have. It’s not very well harmonized. It’s not very well organized. Can you talk a little bit about, when it comes to this subsurface data specifically, what kinds of data, as an example, are we talking about, number one. And number two, what kinds of decisions are we using to have that data help inform? Maybe we could use those as an example to help people imagine what you’re talking about here.
Lorena Pelegrín: Right. So typical types of data are really seismic data, but also well logs. So that’s measurements on a well, and also all the reports around it. This type of data, like well logs, are huge documents. So yeah, those are papers. Those are paper documents that are not easy to digitize. Also, so the decisions around it, we think we can help around drilling decisions, and also decisions around acquiring new assets or even divesting assets. Those are the main decisions that really geoscientists have around those types of data.
Daniel Faggella: So for a well, for example, might the information we’re looking at, the how much did this, how many, whatever we measure, barrels, how many barrels did this well produce under these circumstances with this kind of equipment, at what time, et cetera, et cetera, is kind of production-related, is it maybe safety-related, I imagine? It sounds like it’s a very long document, these well reports, and it sounds like it’s also coming through in paper. I would presume a lot of the value is in how much these things are producing, but maybe you could tell me if they’re used for other purposes, as well.
Lorena Pelegrín: Yeah, so what the geoscientist is looking for is to find what type of formation is there. So the drillings that are done for wells have information around what are the properties of the subsoil there? And there are different types of soil, like sandstone or limestone, which are indications of the presence of hydrocarbon there. So geoscientists are targeting these kind of good wells. So it is a lot about looking for and finding the right soil composition information.
Daniel Faggella: When I hear well, I think a place that we pump black gold out of. But what you’re talking about is a hole in the ground that we may pump black gold out of.
Lorena Pelegrín: That’s right.
Daniel Faggella: So a well, it is something that, it’s just a damn hole, and maybe there’s something there, maybe there’s not. But the decisions are, “Do we want to keep drilling? Do we want to focus resources here?” And maybe to your point, if we can drink in that data from those different, all those different wells, and be able to visualize this stuff and at least make sense of it, we’ll know where to allocate those resources and where we have more promising potential. It sounds like that’s one of the values here of digitizing and organizing well data, if I’m hearing you correctly.
Lorena Pelegrín: Yeah, absolutely. Absolutely.
Daniel Faggella: Okay, great. Great, cool.
Lorena Pelegrín: It means teams are spending too much time looking for the right type of data there, yes.
Daniel Faggella: Got it.
Lorena Pelegrín: And we see that AI or our ML can help these teams find the data and process the data much, much faster.
Daniel Faggella: Cool, okay. Yeah, and I imagine a good deal of this has to do with, tell me if I’m wrong here, Lorena, but having an understanding of your company from working with you guys for a little while, I would imagine that the digitization of these myriad, somewhat chunky paper forms is one part of the process here, using some kind of optical character recognition stuff and working with historical records and maybe congealing and digitizing that.
Lorena Pelegrín: Yes.
Daniel Faggella: And then maybe the other part of it is connecting the dots and identifying things. So within this big ugly wacky well-drilling report, where is the part that tells me this content of this kind of soil? Because it’s such a big, custom, crazy long report, how can we zoom in on the parts of it that actually matter? I would presume there’s the digitization and then kind of like the searching and distilling of those tidbits that matter to us. Is it that where AI is able to layer the value?
Lorena Pelegrín: Yes, absolutely. So the value that we’re seeing there is. Of course, with the natural language processing part of AI.
Daniel Faggella: Yeah, yeah.
Lorena Pelegrín: So the OCR, optical character recognition, is allowing then teams to really perform an elastic search and find characteristics. As I was mentioning, sandstone or limestone or whatever other characteristics that you’re interested in, finding it very quickly. Computer Vision is also helping, so we’re able to find the targets so that the right type of soil where we want to further continue working, we’re able to find it not only by the text, but also by how those photographs and how those documents look. That is also what Computer Vision, in terms of image, similarity, for example, is allowing teams. Yes.
Daniel Faggella: Got it. And just, you mentioned mergers and acquisitions as a really important part of subsurface data. Again, as an outsider, I think about, okay, if I’m Shell Gas and I want to do business with the other players, it would be very helpful for me to instantaneously say, “Okay, let me search through all of their limestone wells and search for these kinds of criteria.” Or, “Let me search through all of these other kinds of wells based on these geo regions and search for these kinds of criteria and be able to really assess what do these people own? What am I interested in? How much promise do I think there are in these wells that somebody else owns?” That’s where my head goes when it comes to mergers and acquisitions value for AI here.
Lorena Pelegrín: Yes.
Daniel Faggella: But you let me know, Lorena, where does M&A, where does this data come in, in terms of the real value for potential M&A?
Lorena Pelegrín: Yeah. So if, for example, right now with the economic downturn with COVID, so if you have a… If you’re a company that wants to acquire, is considering acquiring a smaller operator, a smaller upstream operator who are likely to be suffering right now, if you’re looking at potentially acquiring those assets, you want to go through, let’s say a collection of the assets as fast as possible. And you want to be very focused on your search. So if you have the right data, as I was mentioning before, because you have access to text processing through OCR, if you had the metadata associated with the information of those assets, you can prioritize your search. Because at the end of the day, it’s, okay, I’m searching for which operator or which company it would be better for me to acquire. So I want to make sure that their assets are in the best shape possible, and the potential behind those assets is the best.
Lorena Pelegrín: So through this positive search, based on metadata and visual image similarity, what you get to do is, you jump from one asset to another, kind of like in the right order. So you focus your resources so that you can come to the best possible decision in the shortest time. This is how mergers and acquisitions teams can can benefit from that.
Daniel Faggella: Got it. Okay, cool. And then that’s useful context, just in terms of that. The M&A process for energy, I imagine, is somewhat new for some of the listeners. So hopefully that information was helpful.
Daniel Faggella: I know that acid inspection is another big part of the game here. To be honest, Lorena, when I think asset inspection and I think oil and gas, I think about, and we’ve had a lot of great interviews on the Computer Vision side of this equation.
Lorena Pelegrín: Yes.
Daniel Faggella: So Drone Deploy, for example, was on talking about what they do in the energy space with drones and video data to look at and inspect assets. AT&T does something similar with some of their hard infrastructure out in the world, their towers out there that they need to inspect periodically. But you folks are obviously more in the NLP and the search and discovery space. Where does AI kind of fill that gap for asset inspection there?
Lorena Pelegrín: Yeah, so where I see the, let’s say the most natural place for document-based AI is really for brownfield assets, where there is a lot of unstructured content already available. Really, what can be done is to merge it with what you were just mentioning. So if there’s new sources of data from remote inspections through drone footage, or even satellite image, what can be done is to consolidate those new sources of information with the digital data out of the legacy data, or pen and paper data from those existing assets, from those brownfield developments, and really have a more complete picture, in order to prioritize your maintenance tasks.
Daniel Faggella: Got it. Okay. Are there any particular examples that you like to walk folks through, hypotheticals that make this visual in people’s mind? Because I see where you’re headed, but I’d love to crystallize a little bit on this asset inspection side of things. And also, sorry, if you wouldn’t mind, the word brownfield, probably new for some of the folks tuned in, too. That would be helpful if you could tune us in to what that means.
Lorena Pelegrín: Right, so brownfield as opposed to greenfield. Brownfield means an existing installation or existing facility, whereas greenfield is a new facility. So it would be a new asset or new development.
Daniel Faggella: Got it – thanks.
Lorena Pelegrín: Yeah, so that’s why we think that any AI, document AI related to existing assets would be very beneficial. It’s not only about the new data that we’re capturing, but also all the context around that asset that has been captured by others previously, and that is typically captured in unstructured way on the form of reports and sheets and others. Or if there have been, of course maintenance reports, but also accident reports, perhaps. Every asset also kind of has its own personality. There is a lot of information there, and where I see that is really around the metadata. So a footage or an image, or even a CloudPoint scan, it cannot give you all the metadata. It cannot give you all the context around that asset.
Daniel Faggella: Yeah.
Lorena Pelegrín: Okay? So I do see the value there. And you were talking especially about walking through examples, right? So for in oil and gas, so remote assets, so pipeline route inspection and surveillance, that would be an example there. That would be also related with gas utilities. There’s a lot of assets out there. They’re aging. They’re aging assets. So there’s a lot of existing information there.
What we’re seeing also is the new gas mega rule for MAOP reconfirmation for gas transmission pipeline. That is a maximum allowable operating pressure for gas transmission pipelines that went in operation up to 1970s. So all those assets are going to have a lot of legacy data, a lot of reports. And the requirement from that mega rule is that operators are going to have to demonstrate that their pipes, that their systems, are operating under that maximum allowable operating pressure. They even have a requirement for so-called traceable, verifiable and complete records.
So we see that is also a use case where AI is going to make the work a lot easier for the people that have to really go back to those records, look at them and demonstrate that in disparate records, like material test reports, alignment sheets, steel drawings, that you see the same pressure, let’s say, in those documents. So we’re seeing, we think there’s a lot of potential document AI for existing assets there.
Also, we are seeing in renewable energy. In renewable energy, we’re seeing different multiple cases also on remote inspections for wind farms. So we’re also where drone footage is being used to identify potential, also prioritization of maintenance tasks there. So we’ll see a parallel there that a lot of what is done in older assets in oil and gas, and even gas utilities and electrical utilities, can also be transferred to renewable energy assets. Yeah.
Daniel Faggella: Got it, okay. Yeah, and that, I’m picking up what you’re putting down here, where basically, if we take a look at Computer Vision data, and we get a sense of, all right, how does the equipment look? Do we see any rust? Do we see any pieces or parts that don’t look the way that they should, or what have you, that’s one part. But then there’s also pressure and all the data streaming off of the thing itself, which is not visual, but has to do with gauges and other ways that information could be collected. Obviously, that’s a big part of inspection, as well.
Daniel Faggella: And it sounds like one of the factors here is that there’s going to be some new regulation around how tight and how on-point those records have to be for some of these older pipelines. They’re going to have to have really on-point records, because maybe there’s some environmental risk or something, so the government’s going to kind of crack down on them. It sounds like being able to be transparent with all that data is going to be potentially of higher value than it even was in the past.
Lorena Pelegrín: Absolutely. Absolutely.
Daniel Faggella: Got it.
Lorena Pelegrín: And it’s going to help a lot also compliance teams.
Daniel Faggella: Yeah.
Lorena Pelegrín: Yes, it’s going to make their lives much easier. And what we’re also seeing is that as oil and gas companies are really transitioning, or starting to transition into renewable and energy efficiency, they’re going to start also to be subject to that, to increased regulation. And we see that that is going to be of real help for them, yeah.
Daniel Faggella: Got it. It’s really funny to see how often compliance considerations drive interest in artificial intelligence. Because getting people to change is obviously very, very hard. But if the regulatory slap is the consequence and the technology can give you a reasonable ability to get past that regulatory slap, oftentimes that’s going to be… The avoiding of pain versus the pursuit of something new is often going to be what moves the needle. It will be really funny to see if, in oil and gas, it’s exactly the same case.
Lorena Pelegrín: That is a very good point.
Daniel Faggella: There’s a lot of things happening in finance where, all of a sudden, we need more transparency on this kind of information for this California consumer law or for this GDPR or for this other banking interest rate, London change of some kind, where now a lot of our legal docs might need to be compliant in this new way. Well, if we can have a system that can find all of those, so we’re slightly less likely to get slapped and have some exception where we get punished for it, all of a sudden now we’ve got some budget to adopt tech. And maybe in oil and gas, it’ll be the same.
Speaking of kind of where things are going, I know you wanted to touch on the topic of digital twins as our kind of third and final theme for the value of artificial intelligence in the energy space. Talk a little bit about sort of why those are valuable, why they’re being considered right now in oil and gas, and where AI can find its fit.
Lorena Pelegrín: Right. Digital twin is really a digital model of your, not only of your assets, but also of your operations. So you see there, we’re going to be seeing that the operations of these, of energy assets are more and more aided by the digital versions or the simulations of those operations. At the same time, we’re going to be seeing all the back office work associated with the operations of the asset. We’re seeing, of course, that they are more and more digitalized. And this is really what the digital twin is. So we see that, again, that document AI is going to help. It’s going to enable that digital twin or digitalization of operations, both production, I’m going to say production from the perspective of energy, but also business operations.
Lorena Pelegrín: Both elements of a business operation are going to highly benefit from that digital twin view into the processes. For that, we think that document AI is going to be needed for that. Again, it’s going to be about creating a digital version of your plant.
So for that, you probably have a lot of unstructured document information around it. One way of setting up your digital twin is by scanning, or even scanning if it’s on paper, but also ingesting the information from your digital sources, saying, “Okay, here is all my equipment, here’s all my pipes.” And make it available in a digital model, so it’s not only a digital format, but it’s a digital model, a simulation of your operations. We just think that’s going to be of help, setting up those digital, really digital processes for energy companies.
Daniel Faggella: Yeah. I can imagine a hundred reasons why. We’ve talked to folks, we’ve had a lot of great interviews in the tangential digital twin domain, heavy industry manufacturing, et cetera.
My tertiary understanding here is that, if we can relatively accurately model the interactions of these systems under different conditions, then what we’re we’re hypothetically able to do is ask ourselves, “Well, what would change in the supply chain if we were able to improve production by 4% over the next six months from these sites?” Or, “What kind of new demands would be placed over here, if these two wells shut down and we started drilling up in this area,” or something like that. And we’d be able to hypothetically see, what does this do to our numbers? What does this do to our other systems?
Daniel Faggella: Sometimes, that can be part of the value, kind of asking about potential futures and asking about potential impacts and cross-effects or from different things we would do within the business.
Lorena Pelegrín: Right.
Daniel Faggella: I’m speaking at kind of an abstract level here, Lorena. So maybe you can make it more tangible. But that’s the understanding I have. Where do you really see digital twins driving value in terms of day-to-day decisions for executives who really need to steer the company?
Lorena Pelegrín: Yeah. It’s production optimization, as you were pointing. So digital twins is not really… I mean, it’s new, but simulations have been there since, I don’t know, since 30, 40 years. So this is, it’s a way to see the, what if, like you were mentioning.
Daniel Faggella: Yeah.
Lorena Pelegrín: And to be able to do it with the latest data, the most up-to-date data, and basically any time. So to do a simulation, how it works, especially also with design, what you do is, you get the data, you give it to your simulation team, and after a week or two, they come with answers to specific scenarios. This is a way to optimize your operations in a much faster way. So that is one use case or one benefit around that. But the other benefit, really, is about automating your operations so that your back office operations can be more automated.
Daniel Faggella: Got it. Yeah. So you’re saying the simulations is not, cannot just only be for production.
Lorena Pelegrín: Right.
Daniel Faggella: But could also be for your core operations on the back office, if you can determine how the workflow of different paperwork or-
Lorena Pelegrín: Exactly.
Daniel Faggella: Whether it’s HR or whatever, if we can digital twin that ecosystem with the right data streams, then we could also understand our capacity and ask those if-then questions.
Lorena Pelegrín: Absolutely.
Daniel Faggella: Yeah, okay.
Lorena Pelegrín: Yeah, and that is also what robotic process automation is doing. Being able to have automated processes that kick off, let’s say… I think one vision would be have processes in the back office that are able to kick-off, let’s say, the procurement of parts that are going to be needed for, because a particular pump is it starting to degrade, or we have signals from the digital twin models and the predictive maintenance models that are telling us, “Hey, this is soon to be replaced.” So kick off those procurement processes, or even raise flags, telling workers that this is something that needs to be done now. I think it’s going to be around automating those back-office processes.
Daniel Faggella: Got it. Okay, cool. I think when a lot of people hear digital twin, they’re presuming we’re talking more or less exclusively about the hard assets, simulating what happens with these hard assets, this manufacturing equipment, these trucks, these wells.
Lorena Pelegrín: Right.
Daniel Faggella: But what you’re saying is, if we can have the right data streams coming off our internal processes, and that RPA might help to bootstrap this a little bit, we might be able to do the same work there.
Lorena Pelegrín: Yeah. Yep, yep.
Daniel Faggella: Cool. Okay, fascinating. Well, Lorena, man, we sure did try to pack a lot into this interview. I know ahead of time, I was talking about how fast we’d have to go. But we’re just at about a half hour and we actually made it. So three great topics. You moved quickly with us, and gave us some good examples, which I certainly appreciate.
Lorena Pelegrín: Well, thank you.