Machine learning in oil and gas exploration and production has successfully begun to scale up from pilots to specific and bespoke industrial solutions – but must remain grounded in the real world.
If the rapid pace of technological development over the last 30 years has taught us anything, it is that Utopia is not coming. The flying car and teleportation are still a long way off, and if they do ever arrive, we all know better than to imagine that they will magically transform society. Genuine technological progress is always grounded in the real world, with all the missteps, wrong turns and imperfections that go along with it.
Machine learning in oil and gas exploration and production has successfully begun to scale up from pilots to specific and bespoke industrial solutions. © Cognite.
And so it is with the latest buzzword zipping back and forth across the oil and gas industry: machine learning. Despite the hard lessons we have learned about the application of technology to the real world, some machine learning evangelists are still enthusiastically selling a vision of sentient robots that will solve all the industry’s problems in a single stroke.
Machine learning will change the way the industry runs. Not because it is going to learn how to replace humans, but because we can teach machines to enhance the skills, performance and cognitive power of the humans involved, in ways we never thought possible.
Real-World Problems
Let’s take a very ordinary problem: paperwork. Written processes, printed forms and manual documentation are not disappearing as slowly as you might think. The oil and gas industry’s need for complex machinery, precise engineering, rigorous safety standards and responses to variable conditions mean it is paperwork-heavy; very heavy. Many companies and operators are still wedded to forms, checklists and filing in triplicate.
Paperwork – a real-world problem. © Cognite.
The issue is not that those processes do not work, but that they absorb huge amounts of time from skilled, operationally-critical staff who have better things to do. Using machine learning by teaching a piece of software to carry out that manual paperwork with minimal oversight does not need to replicate and replace the skills of scientists and engineers, but it can free them up to make the most of their specialist skillset.
Too Much Information
Another issue is the sheer volume of data now being generated and collected. We might be forgiven for thinking that any and every bit of data gathered is of value, but it is actually the opposite – many companies have spent millions on collecting, cleaning and storing data from vast numbers of sources, only to be left staring at a mountain of data that they do not really know what to do with. A recent survey by McKinsey estimates that 80–90% of the data budget of major companies is spent gathering and cleaning data, with only 10–20% being spent on building models based on that data that actually generate value.
Machine learning can help by automating the rules of collection, connection and cross-reference necessary to refine that mountain of data into recognisable patterns. Those patterns can then provide insight for the same critical staff to make more informed decisions at crucial times; enhancement of skills, not a replacement of expertise. But too many data science companies are still selling the ‘snake oil’ of algorithmic magic – the idea that all you need to do is hand over the mountain of data and they will sort it into solutions. That is simply not the case.
Numbers are Passive
Oil and gas companies need to be careful about what data they collect, collate and press into action. Numbers without meaning, purpose or application are just that – numbers. They cannot do anything by themselves. However, when data collection and machine learning are used in targeted, contextual and specific ways, they can make a significant difference.
Helping humans to make informed decisions. © Cognite.
Let’s say, for instance, that your pump has a temperature sensor on the motor, so you can prevent it from exceeding its operating temperature. However, the sensor data does not reflect the broader operational context, so it has somewhat limited potential for monitoring and decision-making. It does not know that the pump has to work a bit harder when maintenance is being performed on one of its neighbours that typically operates in parallel with it; or how long it is since the filters have been changed; or simply that it is a hot day outside. There are many factors that need to be considered. So machine learning models can watch for patterns, but the trick is to efficiently present all of this data to the engineer. That way, they can use their experience to diagnose the problem without needing to chase data from multiple systems.
Different Every Time
In oil and gas, no two locations are the same. Geography, weather, location, legislative environment and more can all vary wildly. This, in turn, requires a specific and bespoke industrial solution where existing equipment is adopted, or designed and manufactured to order, to solve unique challenges.
So when it comes to applying the power of data and machine learning to those varying environments, it follows that specific systems and software should also be adaptable, following along with a way of working that allows for unforeseen changes and can adapt to a changing system. They must be able to change as lessons are learned.
Despite the need for tailored solutions, oil and gas companies should not let perfection be the enemy of ‘good enough’. The time and trouble required to design a system that can do everything you could ever want costs far more than developing an MVP – minimum viable product – that can be up and working quickly. You need to be able to build it fast, test it fast, and change it fast, in order to keep pace with the lessons you are learning on the ground.
Chalk at Valhall: A Live Warning System Using Machine Learning
On the Valhall gas field in the North Sea off the coast of Norway, engineers working for an independent oil and gas operator faced a unique problem. Chalk silt in the seawater was gradually plugging the wells, and existing sensor data was only monitoring pressure and temperature. The huge amount of broad historical data, fed into an off-the-shelf model, was proving to be not nuanced or specific enough to detect when wells were about to clog or to predict future events, and it led to costly delays. A unique set of circumstances – the geography, the machinery and the physics of the problem – needed a unique machine learning response.
The Valhall oil field.
By combining physics-based modelling with specific and targeted sensor data, a development team consisting of subject-matter experts and data scientists created a live warning system that could recognise the unique combination of factors that indicated the beginning of chalk influx in the wells, allowing engineers to intervene before any real damage was done. This system is estimated to be saving the operator up to $15 million a year.
Teaching the Machines: People Still Matter
Many industry operators are seeing the value of this highly collaborative approach, not just in terms of involving experts from every stage of their operations in designing the models, but in encouraging separate software and data providers to work together. A rising tide lifts all boats, and the specialist solutions that collaborations can produce have developed into new products, services and business models for the industry as a whole.
In other words, it is much more about how leaders in the industry choose to behave than any mythical power machine learning may or may not have. Think less about finding machines to do the difficult jobs we no longer want to do, or to replace the skills and expense of the human beings that currently do them. Think more about finding ways in which machines can be taught to enhance and improve the power of those humans to make informed decisions.
Further Reading on Machine Learning in Oil and Gas
Artificial Intelligence and Petroleum System Risk Assessment
Mathieu Ducros and Félix Gonçalves; Kognitus
Using artificial intelligence to assess hydrocarbon charge risks.
This article appeared in Vol. 17, No. 1 – 2020
Artificial Intelligence and Seismic Interpretation
James Lowell, Peter Szafian and Nicola Tessen; GeoTeric
The key to all seismic interpretation is the interpreter’s experience and knowledge, so why should artificial intelligence change that? The reality is, it shouldn’t.
This article appeared in Vol. 16, No. 2 – 2019
One Step Closer to Fully Automated Fault Extraction in 3D Seismic Data
Sven Philit, Fabien Pauget, Sebastien Lacaze; Eliis, Caroline Guion; Eliis Inc.
To boldly go where no interpreter has gone before, getting one step closer to achieving the fully automated detection of faults in 3D seismic data.
This article appeared in Vol. 16, No. 1 – 2019
Advancing Geophysical Interpretation in Oil and Gas Exploration
Gehrig Schultz, Chris Tucker and Kirsty Simpson; EPI
Geophysics must change – but it must also remain meaningful.
This article appeared in Vol. 16, No. 3 – 2019
Artificial Intelligence in Oil & Gas Production
Bjørn-Erik Dale & Vidar Uglane; Solution Seeker AS
Solution Seeker, a Norwegian tech start-up and spin-off from the Norwegian University of Science and Technology, is developing the world’s first artificial intelligence for real-time production optimisation.
This article appeared in Vol. 15, No. 3 – 2018
Part I: An Introduction to Deep Learning
Hongbo Zhou, Statoil; Lasse Amundsen and Martin Landrø
Once, artificial intelligence (AI) was science fiction. Today, it is part of our everyday lives. In the future, will computers begin to think for themselves?
This article appeared in Vol. 14, No. 5 – 2017
Part II: An Introduction to Deep Learning
Lasse Amundsen, Hongbo Zhou; Statoil, and Martin Landrø
“We need to go deeper.” Leonardo DiCaprio, in the film Inception (2010).
This article appeared in Vol. 14, No. 6 – 2018
Seismic Interpretation with Machine Learning
Rocky Roden, Geophysical Insights, and Deborah Sacrey, Auburn Energy
A methodology to deal with ‘Big Data’.
This article appeared in Vol. 13, No. 6 – 2017
Advertisement
