AI And Robotics In The Oil And Gas Industry

Applying new technologies to industry at all levels of the organisation

Now, more than ever, we know that if companies want to survive they have to adopt new technology. But what does this look like in practice? While it is clear that the kinds of tools required – and their implementation strategies – will vary from industry to industry, it’s not always easy to understand what the most forward thinking businesses are doing.

One field which has drastically changed with the advent of new technologies is the oil and gas industry. The use of AI and robotics in particular can improve operations from the back office to the most remote offshore asset. DISRUPTIONHUB spoke to Tim Airey, Technology Principal within the Digital Innovation Organisation at BP, to find out how the business is solving problems with these specific technology solutions. 

The Digital Innovation Organisation 

One way that BP pursues technological developments is through its Digital Innovation Organisation (DIO). As Airey states, this organisation looks at key areas of digital technology, such as quantum computing, blockchain, robotics, and AI, and how they might impact the wider business. 

The DIO’s role is to look at far horizon technology and understand where that technology could be disruptive to our organisation, how it could offer opportunities for our transformation,” he says.

When we talk about nascent technologies we mean technologies that sometimes aren’t even physically available yet; they’re not really ready for commercial use. But the reality is that we need to be thinking now about how to bring those technologies in. There’s probably six to seven years of thinking as an organisation about how we change what we do, where it can be applied, and bringing a lot of people up to speed in terms of how to interact with that technology.”

A notable aspect of BP’s DIO is its holistic consideration of business and technology together. The team is made up of people from diverse backgrounds in terms of business and scientific experience, leading to fruitful collaboration of academic and corporate thinking.

So you want to talk about AI?

When discussing BP’s use of AI, Airey is first keen to clarify the company’s understanding of the term.

We consider AI as a pyramid with a number of different subsections,” he says. “At the base of that pyramid are things like machine learning. We then see that flowing up into tools which we can pull out from the AI domain – so things like computer vision, which relies heavily on AI. Then, ultimately we get to the pinnacle of the pyramid which is cognitive computing. This is the concept of mimicking human reasoning and decision making with AI.” 

The positives and pitfalls of predictive maintenance

One of the most impactful applications of AI in the oil and gas industry is machine learning, thanks to its ability to help companies monitor their assets. In any space where reasonably high quality data exists over a period of time, machine learning can be applied to identify insights. This opens up the possibilities of predictive maintenance.

By applying the techniques of machine learning,” says Airey, “we are able to look at historical signal information around various components of an asset. We map that information which can then tell us if the asset or the various devices are operating how we would normally expect – or if there is an anomaly.”

Historical information of an unplanned downtime event can then be added to the mix to find indicators that a problem may be imminent. As Airey notes, this helps the company to maintain the smooth running of assets and operations.

If we look at the results we’ve had out of this process, we can have anything up to 14 days’ notice of an unplanned downtime. That’s meaningful for us because it gives us the time to do something about it.”

Although this kind of predictive maintenance might seem like a magic wand – and has even been applied elsewhere in BP’s IT organisation and trading patterns – there are shortfalls to this technique. Most importantly, it can only be applied in areas where there is good quality, historical data. When problems occur that haven’t arisen before, the data simply doesn’t support what is happening. This makes it impossible to make predictions. As Airey puts it, “If you haven’t got the data, it’s very difficult to forward predict…”

Into the deep

As an example of BP’s use of deep learning techniques, Airey cites the quality inspections which are a crucial part of any oil and gas company’s safety procedures.

We have teams of people in inspection and maintenance that look at these critical components,” he says, “to understand if we have any degradations in there – like corrosion. We need to know what’s happening with that corrosion over time. Is it staying static, is it getting worse?”

To fulfil the obligations of a complete quality inspection, vessels need to be thoroughly inspected. This can present a significant challenge, especially if the vessel is in a remote or hostile environment.

We put people there if we possibly can,” states Airey. “But in some of our sub-sea assets we run long streams of video, taking images of what’s going on. Afterwards we have to go through a process of looking at the footage almost frame by frame, to see if there are any anomalies that we want to investigate further. This is not only human intensive, but you can imagine keeping focus on the task is difficult to do.”

By deploying deep learning techniques, quality teams can have a computer look at the imagery instead. In several effective proof-of-concepts, the DIO has fed deep learning systems thousands of images of intact and corroded assets. This enables the computer to learn what corrosion typically looks like, and independently identify areas of concern when faced with unseen footage. Combining the deep learning system with human expertise in this way increases the likelihood that damage will be spotted.

On a higher level

While the abilities of machine learning and deep learning systems might sound impressive, they cannot operate beyond the limited scope of their training. This means that – unlike humans – they cannot respond to unexpected or novel problems.

If businesses want computers which can apply their knowledge, experience, and understanding of their environment – in short, which can use the characteristics of human reasoning – to solve problems, then they need to develop a higher order of artificial intelligence.

As Airey remarks, the implications of this higher order AI in the oil and gas industry are huge, for a number of different reasons.

We’re interested in how can we mimic human reasoning in a very defined area, to a very defined problem. To be clear – we’re not creating consciousness in these systems. But when we think about humans, when we come across novel problems, generally we are not stumped, or stopped – we can think of an option of what we might want to do. We apply our experience and understanding to think about what we believe to be the best course of action, and then we can act upon it.”

“This is what we are trying to mimic within certain higher orders of artificial intelligence. So – in the context of the machine learning example – what would happen if we didn’t have all that data? What if we were able to codify knowledge instead? We could present a particular problem to a system, and without all that previous pattern matching information it could apply its knowledge, to come up with different scenarios of what it could do. It can then look at each one of those scenarios, consider the pros and cons of each and make a recommendation.”

Augment the mind with the machine

This is an incredibly important area for us,” Airey continues. “In the oil and gas industry, we don’t always have perfect information. And even if we did, one domain expert might interpret that information differently to another.”

The aim of using higher order AI in industry is to augment the knowledge of human domain experts with a computer system. The computer can look at far more data, has encoded knowledge, and can start to evaluate different scenarios of what could happen, making it a useful tool in the decision making process.

For now, computers aren’t at the point where they act upon their findings, but this may one day be the case. Crucially, it’s important to note that cognitive computing systems are fully explainable. Unlike the black box at the heart of deep learning techniques, human operators can clearly see how cognitive computers reach their decisions. As such, it is easier for people to learn from and trust the findings of the machine.

A robot reality

Together with the various forms of AI, another area of focus for BP’s DIO is robotics. For Airey and his team, robotics is also best understood as a technology stack with a series of interdependent layers. The first of these is the physical layer, which includes the fields of manipulation and locomotion – and understandably presents a significant challenge.

Our operations take place in a really different environment from a manufacturing robotics system, where robots are bolted to the floor, items move to them, and they work really well. We’re in hostile environments, we have zones which are dangerous from an explosion perspective; they’re wet, they’re oily, there’s salt… It’s endless.”

So when we have this concept of a robot coming in to do work it’s a nice vision but the reality is that there are a lot of steps underway to achieve that. Consequently, the area that we concentrate on is novel locomotion. We need to find new ways of being able to navigate our assets because from a topography perspective this is incredibly complex. None of our assets were built for robots.”

Along with locomotion, the DIO team also has to consider how to develop robots which are capable of performing certain tasks. Manipulation – such as generating torque – remains a real challenge in industrial robotics, which is an obstacle to having robots carry out things like remote maintenance. What’s more, there are also issues around how devices can be effectively controlled, and whether or not they are safe from cybersecurity threats.

What we find in our area is that one robot doesn’t do everything, you have to have a fleet of different devices to achieve a single objective,” says Airey. “So how do we control these multiple devices? Then we also need to look at the cybersecurity question – how secure are these devices that are acting on our behalf?”

Intelligent robots, unparalleled possibilities

In any discussion of AI and robotics, it’s important to consider the convergence of these two technologies. If companies can integrate higher order artificial intelligence with machines that they are able to effectively operate remotely, then this will bring considerable advantage.

The human reasoning piece, more than even the machine learning, is one of the critical components of an effective robotics system in the future,” says Airey. “Why? Well, firstly because the robot has to get on location. It needs to understand its environment, to mission plan, to work out how to get there and how to navigate a complex asset – be that an offshore platform or a refining asset. But what happens if it chooses a path which it thought would be absolutely fine, and now there’s an obstacle in the way? It needs to recognise the obstacle first of all and then figure out what to do.”

Robotic life on Mars

Robots equipped with such cognitive AI might be some way off for now, but developments are underway.

We are starting to put the artificial intelligence on a chip on the edge of the robot,” Airey notes. “We are also trying to push the processing from something that happens in the background to actually doing it live on the device.”

Deploying sophisticated robots to navigate difficult environments, manipulate assets, and process their findings on site represents an exciting horizon for the oil and gas industry, along with other heavy industrial fields. In fact, BP’s benchmark for all of this technology is the NASA spinoff Beyond Limits, an AI company which was involved with the Mars Rover project.

With cognitive AI having successfully been integrated into a robot on Mars, why not aim for similarly impressive feats on our own planet? Robots capable of some form of human reasoning are clearly on the horizon, it’s just a question of how fast we can get there…

For more interviews with industry experts, sign up to our free weekly newsletter.