AI as a concept is way past its sweet sixteen, but only in recent years has it become confident enough to learn to drive. Sooner than later it will be fully licensed and out on the streets.
It’s the same in the Oil Patch, say experts. Companies are relying on AI for more and more functions in every area: drilling and completions, production, and back office work. The goal is always to speed the workflow, allowing each human to manage more tasks with greater speed and accuracy.
SparkCognition’s Philippe Herve notes that most companies in the oil and gas sector know AI is the wave of the present as well as the future. “They all know that AI is coming, they all know that they have to get on the bandwagon—some of them have already jumped on it, some are wondering how to get started,” he said. Another question companies are facing involves whether to develop automation and AI internally or to outsource it to a tech company.
At the forefront of the movement are the international producers such as BP, he observed. Larger companies often develop their own, but even at the BP level many look to tech experts to bring best-in-class technology to the table.
Large independents are also embracing AI, mostly by bringing it in from third party developers.
Latecomers to the bandwagon are NOCs, or nationally-owned oil companies, the most prominent ones being Saudi Aramco and Petronas. “NOCs have been, generally speaking, lagging a little bit,” Herve said. But he thinks they’ve recognized the fact that in order to remain competitive in the international market from the standpoint of cost-efficiencies in a low price environment, they need to accelerate their digital transformation.
With all the challenges of 2020, he said, the NOCs “will need to wait for the rain to end, or learn how to dance in the rain. The NOCs have definitely seen the need to accelerate their transformation.”
Aerospace and Oil and Gas are probably the most advanced industries when it comes to digitalization and AI, although the oil industry lags behind what is available in the consumer world with companies such as Tesla and Amazon. Herve and others believe that recent oil price downturns, in 2015-6 and 2020, have pushed the oil patch to digitization for survival.
Any company looking to start in AI should first grab the low-hanging fruit, Herve said. “That’s where AI brings a lot of value extremely quickly. Probably the easiest way to get on the artificial intelligence bandwagon is with cyber security. Let’s protect your endpoints on your network with the most advanced protection now made available with AI based technology.”
The cost for AI endpoint protection is about the same as most traditional endpoint protection products, he said, with two advantages for AI. “First, you are going to catch zero day malware without having to update the product on a daily basis or sometimes multiple times per day. Secondly, you don’t saturate your network with constant updates for the latest malware signature.”
Few clients really understand the value of improved security unless they’ve had a ransomware attack or other cybersecurity breach. When a company has indeed been attacked, “They know exactly how much it costs them to get it remediated.” Recent well-known cyber hits have cost large companies tens of millions of dollars to recover from.
In a worst-case scenario, a ransomware in a German hospital in September caused the death of a patient. The attack had crashed the system, forcing the hospital to turn away emergency patients. This woman then died on the way to another hospital that was 20 miles away. This was the first death ever associated with a cyberattack.
AI options with more immediate paybacks involve predictive maintenance. When plugging AI into predictive maintenance, “First, it’s going to reduce your maintenance costs, and that’s always a good thing.” In that category AI will pay for itself many times over in reduced maintenance costs. “But the biggest gain is the improvement in production, and it doesn’t really matter what you’re producing,” Herve said. He noted that whether you’re producing medicine, cars, or oil and gas, you’re improving your production with predictive maintenance.
One recent SparkCognition PM project for a major producer created a $10 million value for the bottom line. In fact, Herve noted, this saving came from preventing a single shutdown event, right at the start of the project. Further savings are sure to follow.
Improvements in production also boost safety because that preventive maintenance has also prevented breakdowns that can create dangerous incidents on the site. Plus, PMs are usually done on a calm, daytime schedule instead of the emergency, 24-hour on-call situations created by unplanned breakdowns—which can lead to dangerous middle-of-the-night accidents.
Blake Burnett of IOTech notes that AI is still not at the point where it can stand alone—human intelligence is required to make sure the AI is learning the right things. He tells the story of a company a few years ago that implemented new software for their billing and accounting department—and instead of making them more efficient, the staff ballooned to more than seven times its previous head count to accomplish the same amount of work.
“So much of what has been claimed to be business automation has actually proven to be the opposite,” he said. “I think AI tools to take advantage of APIs (application programming interfaces) are what are going to really make the most difference in the next 10 years.”
The Oxford dictionary defines this as “a system of tools and resources in an operating system, enabling developers to create software applications.”
For Burnett, it’s not just about the big tasks, but about making small jobs—like billing for parts—stay small.
One idea would be imbedding a chip into a piece of equipment that is sent on consignment to a client. If it’s never used, the client never pays. But, “The minute they turn it on, it can make a call to QuickBooks to invoice them for it. I think that this will have a huge impact in the rental market.” This would take the place of someone in the warehouse writing a ticket and sending it to accounting where it would have to be manually input for billing purposes.
He continued, “Artificial intelligence software that takes advantage of the APIs of suppliers and customers and starts to tie tasks together can make things that are used to take weeks take microseconds.”
The future, Burnett believes, will allow people in the back office to “spend their time approving what AI wants to do and breaking up bad habits.”
AI has bad habits? The common belief is that it learns as it goes along, and it always learns the right thing.
Not so, says Burnett.
“My entire career I’ve had to deal with—well, we call it ‘voodoo’—but it’s actually just miscorrelations by people,” he said. “Somebody pushes a button and it didn’t work, so then they do this other thing and they re-push the button—and it worked.” That person then may falsely conclude that what they did worked, when it was actually something else that made the difference. Logicians call this “post hoc, ergo proctor hoc.” In English that’s basically, “B happened after A, so A must have been the cause.” Sometimes it is true, but it eliminates the consideration of any other causative factor—and AI is not immune to this logical fallacy, Burnett has noticed.
“From then on, they always do this other thing before they push the button because they miscorrelated.” He compared it to believing that umbrellas cause rain because that’s when you see them open.
With AI this is even more of a problem than with humans, for two reasons: First, once that correlation is made, the AI replicates the problem quickly. Second, a human must spend time trying to figure out why it miscorrelated.
“Having people understand why it happened and how to correct it, that’s going to be a huge problem.”
Overall, however, AI is making huge strides in efficiencies. Operators today are greatly interested in monitoring multi-phase flow meters on separators, Burnett observed. That gives them remote access to recorded data that previously had to be retrieved by a site visit. “We have seen an interest in anything that reduces windshield time right now, since some of our customers have reduced head count.”
He continued, “We are also seeing an uptick in offset well monitoring and jobsite radio sales. For some reason customers are really interested right now in reducing cables on job sites. This might be because they see that as an opportunity to save money, to reduce rig-up time.”
Burnett left a corporate job to start his company in 2016, with the goal of providing internet gateways to clients in order to transmit data from any device into the cloud. “As we got into it,” he said, “we found a need for smarter controllers and more robust controllers than off-the-shelf PLCs.”
The Future
It is said that science fiction writers have more and more challenges in thinking of things that haven’t been done yet, because reality is racing forward at unheard of speeds. The main issue with predicting the AI future is similar—the future becomes reality pretty much every day. With that in mind, here are some areas Philippe Herve sees as next steps.
Machinery design: “Many problems in O&G and energy in general involve complex design: refineries, individual subsystems, oil rigs, etc. Each of these is a complex mechanical and electrical system. Generative algorithms and AI approaches to ideation can help augment human creativity in the design of these systems.”
Root Cause Discovery: “An ideation engine can propose why a problem might be occurring, [and] then this hypothesis can be validated or discarded. Sometimes, humans might not be able to generate the hypotheses effectively, so this is where AI could step in.”
AI Edge Computing Boosts Production: “AI at the edge means that we can develop a greater trust in the intelligent behavior of field deployed assets; we are no longer reliant on communications links, and no longer worried about latencies. Embedding a system like SparkCognition’s SparkPredict into edge equipment means that essentially equipment will come with its own autonomous maintenance capability. We absolutely believe that part of building “smarter” equipment in the future involves the ability for the equipment to maintain and optimize itself without any external involvement.
The oil patch’s well documented resistance to change is breaking down in the current environment. Some embrace this while others are dragged into it from pure necessity—but both categories stand to benefit greatly from the flood of controls, information, and decision-making tools joining the market every day.
Geologists, petroleum engineers, and geophysicists are being joined by IT and data analysis experts in almost every E&P of any size. More than ever, everyone must arrive at work each day prepared to embrace the next round of disruptive changes.
____________________________________________________________________
Paul Wiseman is a freelance writer in the oil and gas industry.