FLORENCE, ITALY—Artificial intelligence (AI) is coming of age, and the oil and gas industry is searching for the most effective uses of this nascent technology.
Binu Mathew, senior vice president and global head of digital products for Baker Hughes, a GE company, spoke with Hart Energy about technology, digitization and how artificial intelligence can prove beneficial for the oil and gas sector.
Where is the oil and gas industry when it comes to digitization?
Mathew: First of all, you must remember that the oil and gas industry has always had digitization in some way. Just look at the original supercomputers. If you look at the oil and gas industry now, what you have is tons of data across the board, but this tends to be very siloed. Individual plants and individual machines generate a lot of data, but it leads almost to what I call a bit of a dichotomy.
An operator today will have streams of data, not necessarily actionable information, but you have many, many streams. In fact, it makes it more difficult. You’ve got all this information coming into you on different screens, and you have got to make judgement calls based on that.
If you are an operations team with a lot of experience, you make good judgement calls, and if you’re not, you don’t. So you are seeing significant shift differentials in terms of performance.
The other thing that’s also changed in the last two years with the changes in oil prices is that there is far more emphasis on operational efficiency. You’ve got to get the breakeven costs, the dollar per barrel of oil, to a significantly lower level—especially for the upstream industry. That’s changed the name of the game from leasing acreage, essentially a real estate game, to operational efficiency. But to do that you’ve got to be able to use whatever data you have to see where you can run your processes more efficiently. You need to reduce non-productive time. You need to increase overall performance and efficiency. If you’re doing field planning, you need to drill [and] you need to plan that more effectively.
There’s a lot of inefficiency in the oil and gas industry, which because of whenever you had an upcycle, you’d been able to ignore for a long time.
Do you have any examples from your experience on this fluctuating performance?
Mathew: That was one of the problems BP would see and it is why they started using our plant operations adviser. When you are dealing with the independency of multiple machines on a big offshore platform, the machines may themselves be all working fine. But if they are not working within parameters for the process, you still have a process upset. They would see several trips that were being triggered due to that.
On the one hand you have the operators, and on the other hand you have the engineers. The engineers have a lot of data, but to put it into a form that is usable you must spend a lot of time analyzing that. The tool that is probably most widely used is Excel. The engineers will do models from this. We have a lot of analytic capability in the oil and gas industry, but it takes you weeks or even months. If you take an operations team that’s working within a five- to 10-minute window and engineering teams that are operating in periods of weeks to months, there’s this big gap. That is where a lot of change is going to happen in the next few years. It’s certainly the area that we’re focused on.
When other industries such as automotive have traveled down the digitization path they have struggled with the volume of data. How is the oil and gas sector coping with this?
Mathew: Like the automotive sector we have alarms and exceptions and part of the challenge is when you get an alarm, what do you do and is the action you’re taking effective?
It all changed dramatically in 2012. This was the first time that deep neural network made a step change in image recognition. There’d been this competition out in the internet for machine learning on image recognition. The general threshold around that used to be around the 70% level, and every year it would go up by a small fraction. The reason was that up to that point it was all being done by traditional techniques.
The odd thing was that until that point neural networks were considered almost an academic curiosity. The math had been around since the ‘60s, but what people hadn’t realized was what you needed was enormous computational power, and that turned up right about then.
It has improved to such a degree that as of late 2017 we’re not just coming close to human capacity; AI is getting to the point where, once appropriately trained, it can do better than a human being. That is across the board, and you’re going to see this dramatic change over the next five years.
Can you give us an example of this form of AI in action?
Mathew: A very simple example comes from our IntelliStream product, which deals with production optimization. The traditional way of carrying out failure diagnostics on rod lift pumps is to look at the dyna card. A technician would go out, pull out a set of flat cards and compare the patterns. This is a situation that’s absolutely tailor-made for AI. We have AI-based pattern recognition that can meet or exceed human capacity in terms of recognizing the situations, and do it in a fraction of a second.
There are numerous companies dipping their toes into AI in oil and gas in all sorts of diverse applications. Have you got a clear vision of where you’re going with it do you think?
Mathew: On the demand side you have got to improve operational outcomes; you’ve got all this data and you must be able to process it. On the supply side you’ve got all these new technologies. But how to marry the two? This is part of our reason for partnership with Nvidia. Nvidia can provide a lot of the technology, but they’re having a challenge in how to take these powerful new technologies and apply it from a domain perspective.
It’s not just in the case of AI; you can see this across the board. If you take the Cloud and other technologies that have been developing over the last few years, how do you get that into the oil and gas industry? Because you’ve got to make it domain specific, you’ve got to have an end-to-end workflow; you’ve got to have something that somebody can use.
If we’re sitting here in a year’s time, where do you imagine the upstream sector will be with AI?
Mathew: I think this is going to be a continuous expansion, but I think you’re going to see the momentum continually increase, and this is true of any new technology. It’s something that has actually an exponential curve, so you’re going to see more and more and more uptake.
In the initial stages people are still in the process of looking at what it can do. You’re seeing a lot of our customers really pushing on this digital transformation story, sometimes in different ways, but I think as you see that you’ll see more and more momentum across the industry.
What about challenges? Is it finding the niche, or are there still some technology challenges to overcome?
Mathew: The technology challenge is oddly enough the rate of change. I’ve been in the technology industry for a very long time, and you’ve never seen anything like the pace of change. You say that and six months later you’ll say, ‘wow I’ve never seen anything like this fast change, because it’s just increased by another order of magnitude.’