Smart Oil Platform Strikes Multiple Benefits with AI
If you want to see the future of AI, you won’t need a time machine — but you will need your sea legs. An offshore oil rig may seem an unlikely spot to find cutting-edge machine learning, but when you think about it, it’s the perfect place. Oil platforms are expensive and complex. When equipment goes offline, so does production, with the monetary damage often reaching tens of millions of dollars a year. Predictive maintenance — identifying and averting trouble before it happens — would be a boon to rig operators. But that’s just part of what AI can deliver.
Consider this: in addition to looking for problems down the road, what if we also looked into the immediate future — say 30 or 60 minutes? If we knew the health of machinery then, we’d know how much more we could push it. In effect, the AI is saying “go ahead, produce more oil, your systems can take it.”
AI, then, is optimizing on two fronts: keeping both maintenance and utilization at just the right levels. Companies aren’t servicing their equipment too soon (as can happen when they rely on vendor-set schedules based on average, not actual, operating conditions) or too late (when repairs are particularly expensive and time-consuming). And they’re getting every last drop — literally — out of their infrastructure. These are compelling propositions. So why isn’t every oil platform — indeed, every operator of complex production systems — incorporating this kind of AI?
In short: it’s hard.
Using AI to support, enhance, and optimize production processes typically requires a hefty investment — not just financial but in collaboration and effort. These are not plug-and-play algorithms. They need to be custom-built and the to-do list is onerous. You need to understand complex processes thoroughly. You need to work with equipment and sensors that — especially on oil platforms — can come from hundreds of manufacturers. You need to have access to, and work with, data that isn’t always readily available and can be in many different formats. And once you’re done, you’re not done, because there is a continual need to train and recalibrate the algorithms as equipment ages or is overhauled. That’s a lot of work. And a lot of risk. You’re jumping into deep and unfamiliar waters.
But the thing about tough leaps is that every now and then someone takes one — showing not only that it can be done, but how to do it in a smart and savvy way. That’s the case with LUKOIL, Russia’s largest private oil company. It decided that those twin benefits of AI were too great to pass up. So it set about creating a “smart platform” — an oil rig where data-driven, self-learning systems would identify and predict deviations from healthy behavior: now, a half hour into the future, and days down the road. The goal wasn’t just to spot — and prevent — potential trouble but also to generate a dynamic, readily accessible ‘health index,’ letting operators know when they could push equipment harder, and when they should throttle down.
The project to build the smart platform — known as Digital ASTRA — is particularly noteworthy because the result was not simply a proof of concept, but a product that has been deployed — and is live at this moment — aboard a working oil platform in the Caspian Sea. And that’s just the start, as LUKOIL plans to roll out the system to other oil rigs.
This is exciting stuff and it certainly warrants the newsprint. But there’s another great story here, too — that part about demonstrating and sharing how to do hard things in a smart and savvy way. The Digital ASTRA developers faced a number of challenges in implementing the right AI in the right way. How they tackled them provides valuable insight to the many companies that are getting serious about — or at least, seriously thinking about — unleashing the full power of AI. Here are the biggest takeaways from the project:
• Build the right team — and a foundation for teamwork. For an oil platform — indeed, for any complex production system — it takes a lot of people to get the algorithms right. Developers need to understand the processes and work with on-site equipment. They need to access large data sets from facilities that don’t typically open their doors to outsiders. They need to understand and “speak” both engineering and management concepts, interacting with machine operators and business leaders alike. And, they need to coordinate all of the above as seamlessly as possible. This requires a diverse and unique team — one that brings technical expertise, trust and credibility within the industry, and superior project management skills.
The Digital ASTRA team members came from an array of organizations and backgrounds: Alma Services Company (an experienced player in the oil and gas industry, they took charge of defining the overall scope and goals of the project); BCG GAMMA (experts in data science and AI); the Center for Engineering and Technology of the Moscow Institute of Physics and Technology (specialists tasked with writing the software and IT architecture and integrating the Digital ASTRA system with LUKOIL’s server); LUKOIL’s own engineers along with industry experts (advisors with deep knowledge in oilfield geology, oil production, and platform operations and maintenance); and BCG consultants (who would analyze technical processes, interview rig operators on their most pressing needs, and ensure a steady workflow — and continuous communications with LUKOIL).
A team like this brings a lot to the table, but it also creates its own unique challenge: How do you spur collaboration among individuals that aren’t used to working as a group — and often have very different mindsets? The Digital ASTRA approach for this issue? Sit everyone in the same room and get them talking. During a daily stand-up meeting, each team member would give the rest of the group a status update, sharing what they did the day before and what they planned to do now. Simple? Sure. Effective? On multiple levels. Not only did every team member now know what everyone else was working on — and pick up insights they could incorporate into their own work — but these meetings were extremely valuable ‘jump starters.’ When someone was stuck, other team members would pitch in with ideas, and help keep work flowing. By breaking down the silos, we broke down a lot of obstacles, too.
• Data gathering is time consuming, labor-intensive, and not the place for shortcuts. These kind of projects live and die on the algorithms. The LUKOIL models needed to diagnose the technical ‘health’ of equipment in real time — and predict how things will look in the near (and the very near) future. To get such models right, you need to take your initial algorithm and train it. This means analyzing a lot of past data (signaling how equipment performed over time), correlating it with past failures and repair work, and identifying patterns that signal healthy behavior (a system that knows ‘good’ patterns can flag even a small deviation — raising an alarm that trouble may be brewing).
Under the best of circumstances, it can be difficult to get the required data: remember, you need a lot of it. But with a production system like an oil platform there are additional complications. The data you need will typically come from hundreds of machines made by different vendors. And it will be stored in different formats and transferred via different protocols. In the LUKOIL project, there was another wrinkle still: some of the sensor data had been archived using proprietary OEM formats, requiring specialized knowledge to extract it.
Just collecting the data took more than a month and a half from the initial request. But taking the time to go through every hoop is vital to ensuring an accurate, dividend-paying algorithm. So if you’re going to build these models, block out your calendar.
• Your proving ground should have a history. Like many manufacturing companies, LUKOIL has multiple production facilities. While the AI system’s first ‘live’ deployment was on one of LUKOIL’s most modern platforms, on the Vladimir Filanovsky oil field, newer rigs aren’t the best choice for training and testing the algorithms. The key, after all, is to have that wealth of historical data. For Digital ASTRA, an ideal proving ground was an offshore rig on the Yuri Korchagin oil field. It had been in operation for seven years and over that time, LUKOIL had accumulated terabytes of equipment data. Moreover, like all offshore platforms, this rig had experienced teething problems, so LUKOIL also had a cache of maintenance records.
As a result, the team not only had a lot of data with which to train its algorithms, but also a lot of data with which to test them. Since actual breakdowns could be pinpointed to specific time intervals, ‘failure’ data could be isolated and run through the algorithm — to see if the model detected trouble when there was actually trouble. By iterating and fine-tuning, the Digital ASTRA team made sure it did.
• One size does not fit all. A key aspect of LUKOIL’s AI system is that different types of users — with different needs — will be using it. Operators, for example, will want to see a short-term health forecast for the equipment they’re running and be able to zero in — quickly — on specific sensors. Maintenance experts, meanwhile, need a longer-term outlook to optimize repair schedules. And managers require the ability to monitor, and get summaries on, the platform’s overall health. The key is to present the right information in the right way to each group. Ask users to click through ten screens to get to the insights they care about and it may not matter that your algorithms are awesome. This is why user experience and user interface designers should be on the team from day one. And they need to explore, test, and iterate their designs with user preferences and behaviors always top of mind.
For Digital ASTRA, BCG GAMMA’s UX and UI experts conducted more than 30 interviews with end users in order to get the presentations and the interfaces right. For many oil rig operators, it was the first time anyone had ever asked them about ergonomics. But this kind of human-centered design is a cornerstone to developing systems that are intuitive, well utilized, and effective.
• Change management is even more important than you think. Remember what we said earlier, how these are not plug-and-play solutions? Well, that’s true in a couple of ways. Not only are these highly customized systems, but they are also systems you can’t simply deploy and let be. The AI has to be continually trained: machine overhauls, for example, will alter the patterns of healthy behavior. This means that companies need to develop new roles and skill sets within the organization, especially in the area of data science. They also need to ensure that the solution is used at every level of the company. Role-specific system views and interfaces help with that, but so, too does leading from the top — and by example. Management needs to encourage, facilitate, and evangelize the tool’s use.
Most companies have yet to tap AI’s true potential. They may have a strategy in place, and perhaps even a modest initiative, but chances are, they’ve yet to really take the plunge. Here’s hoping that LUKOILl’s leap — and the key takeaways from Digital ASTRA — helps other companies take bigger steps, and reap the multilayered benefits of AI.
To learn more about Digital ASTRA and applying AI to complex production systems, we invite you to read “Smart Platform,” a detailed look at the project we co-authored with Danis Maganov, the CEO of Alma Services Company, and Andrey Skobeev, Deputy General Director, LUKOIL-Nizhnevolzhskneft. This article was published in Harvard Business Review Russia magazine (June — July 2019 issue).
