In 2017, a research paper released by the Unconventional Resources Technology Conference anticipated the new digital oilfield would become a “disruptive technology” creating new value streams for exploration and production, ranging from automated decisions and reactions in real-time to massively improved operational efficiencies, connected infrastructure platforms, and much better interaction between machines and humans.
The Rise of the Machines, Analytics, and the Digital Oilfield: AI in the Age of Machine Learning, predicted that the days of collecting and storing large volumes of data for later analysis would become a distant memory. The digital oilfield would change expectations for all aspects of the industry, ranging from how fast decisions are made to detecting patterns the human eye cannot see in order to take advantage of the insights quicker. Data silos would be reduced and information shared across all areas of the oil company.
“At a simple level, artificial intelligence (AI) will be used to increase the accuracy of predictions to near-cognitive robotic comprehension in machine learning,” it forecast.
These are still early days for AI in oil and gas. But leading technology practitioners are already seeing massive changes sweep through the sector. The question is, can the industry keep up with the pace of change?
In the view of Philippe Herve, VP Solutions at Austin, Texas-based SparkCognition, if it were a matter of cost alone, AI would be much more widely implemented. “Every project where AI is used brings a financial advantage that is very large. The challenge is more on the organisational side. It’s about change management – the three components being technology, people and process. If it was just about the technology, things would move quicker.”
SparkCognition builds AI systems, aiming to help customers analyze complex data, empower decision making, and transform productivity through machine learning technology. It has developed Darwin, a development platform that accelerates the process of building highly accurate, generalized models.
“Our software – Darwin-- goes right to the end of the data science challenge. "As a result, the model built by the automated technology, where we use AI to build AI, is then able to find models that are often orders of magnitude more performant than models built using traditional technologies,” explained Herve.
Darwin was used to create a custom model based on field data regarding monthly oil, water, and gas production, including well static features such as location and reservoir characteristics. Available data included only one data point per month, making it imperative to use advanced modeling techniques to guarantee accuracy.
According to SparkCognition, Darwin automatically cleaned this data and extracted the most relevant features for static, dynamic, and semi-dynamic variables. Forty features were generated and used to automatically build optimized, generalized classifier models (applicable to all wells in a field) to identify seven event types. Darwin completed the model generation process in days, as opposed to the weeks or months that traditional approaches would have taken.
Herve pointed out that many AI projects fail early, as the project chosen is attempting to change internal processes and the way people work in an industry where there is an inherent resistance to change."
One driver for digitalization is the significant gap between a company’s recorded data that is useable for analysis, and the recorded data that is actually being used. “Closing that gap is a low hanging fruit. Let’s make sense of the data which exist and is already available for free,” said Herve.
Some companies have moved data to the cloud, or centralised on central servers which makes it usable -- usually the big IOCs and NOCs. But when it comes to the smaller companies, data tends spread around the company. Being able to mine it requires a bit of a digitalization project on how to access the information. "Some may decide to put all the data in one place, on the cloud or a dedicated center on premise, whereas others may make it available to whoever wants it from wherever the data may exist,” said Herve.
Herve characterised the way AI brings value to any company as coming in three main “buckets”. First is cybersecurity. “AI brings a tremendous step forward in protecting companies from all the malware flooding the world. "It affords a lot of protection against cyberthreats – and oil and gas companies should move faster to protect themselves with these solutions," he said.
The second bucket is to do with predictive maintenance. “It’s where the value is easiest to realise because if things are breaking down, it’s very easy to predict what will break down using AI and data mining solutions. Predictive maintenance is the bread and butter of AI today,” said Herve.
The third bucket is about process optimization - drilling optimization, production optimization or whatever the name is given to a particular process. Achieving solid process optimization typically requires that the assets are safe from cyberthreats and operating properly or in a state that is known.
Return on investment
In oil and gas, the return on investment on predictive maintenance is enormous, said Herve. And when margins have been so thin, being able to work directly on the bottom line with predictive maintenance represents the low hanging fruit.
Oil and gas companies are applying AI to predictive or prescriptive methods and then evolving into more challenging tasks, such as optimization and finding new projects that could be solved using AI.
The first priority for companies is to clean the data. Data quality is a challenge on every project, said Herve. But whereas the old approach of requesting cleaner data may have worked when there were just 20 sensors on a rig, now there are typically thousands of sensors, leaving a vast amount of data needing to be cleaned. That takes a lot of time. Around half of a data scientist’s time is spent cleaning up data before proceeding with a project.
Data scientists may already been on the project for three or six months. “The manager then comes in and says, that’s good enough, we need you on the next project,” said Herve. “Automated model building tools such as Darwin allow for automated data clean up and model building reducing months of manual work to minutes or hours of computer time.”
Given that many oil company CEOs are in the process of absorbing the message that AI is going to help their bottom line, there is now a competitive pursuit for first mover advantage.
And if Herve’s hunch is right, there will be more and more AI-driven projects driven by ambitious CEOs who understand that making use of AI is now a matter of survival for oil companies.