The media regularly publishes news and articles about the latest clever developments in artificial intelligence (AI). Whether helping radiologists in healthcare, or indulging consumers need to find entertainment on Netflix, developments in AI always attract attention. The oil and gas industry is also at the early stages of exploring AI’s potential impact. Uses are evident, startups are forming to tackle industry problems and early adopters are gaining traction. A Google search for AI in oil and gas brings up some 26 million references, about five percent of all references to AI.
While AI is not a new field—the first exploration of AI dates back to the dawn of the computer era—it is shrouded in confusing terms and definitions. This article characterizes AI as a computerized capability to analyze data using cognitive skills that are normally associated with humans. Beyond mathematics, such cognitive skills include natural language processing, translation of languages, visual perception, auditory interpretation and tool creation.
Something that is artificially intelligent uses sensors to detect the world, historic data to provide reference to prior situations, and computers or analytics to process some rules about the world. In addition, AI is distinguished by its ability to process enormous quantities of data from a large variety of sources at super human speeds and interpret that data in the manner consistent with human intelligence.
An autonomous car is a useful illustration of artificial intelligence in action. The car and its sensors can detect and avoid obstacles, predict the likely path of other moving objects, and change the vehicle’s behavior according to driving conditions, e.g., slowing down in the rain or snow. Many robots in the future will feature this kind of other worldly intelligence, learning and adaptability.
AI is at the tipping point for widespread adoption because computer speeds, storage and mathematics can now meet the processing requirements of AI software. Companies like Tesla, Amazon, and Google are investing heavily in AI research, and as Figure A shows (see page 28), searches for terms like AI and deep learning as a percentage of search activity have been growing over time.
Why oil and gas?
Two ingredients that make for the successful use of AI are data and human talent, both of which are abundant in oil and gas. The industry has tremendous quantities of recorded data available for the purposes of training AI engines or for analysis by AI engines. That data can be dynamic and captured hourly by sensors in the field, as with motor sensors or cameras on the millions of industry field assets, or static, such as legacy seismic data recorded over the decades. The greater the volume of data, and the higher the accuracy of data, the more impact that AI can have.
The industry employs ample human talent to interpret that data. Geologists spend considerable high-cost time stitching together disparate data sets and preparing maps from seismic and other subsurface data. Expensive financial professionals laboriously read joint venture and land contracts to develop the royalty payments from production. Control room operators monitor remote gates and tank farms for health and safety compliance. Globally the oil and gas industry frequently ranks among the highest paid industry sectors.
Early indications suggest that there is no single economic driver for using AI. Motivations include supplementing existing jobs with AI to free up capacity and lower costs, improving the quality of analysis to reduce risk or variance, accelerating scale work execution and capturing latent capacity in equipment or people, as illustrated in Figure B by analysis from consulting firm EY.
McKinsey research shows that 45 percent of executives see AI as a powerful competitive differentiator. These are general drivers of performance, and applicable to many facets of the industry, suggesting plenty of scope for applying AI to the industry’s challenges. Oil and gas may be able to capture USD150 Billion in value by utilizing AI, as McKinsey illustrates in Figure C.
Use cases are beginning to appear, and, as the barriers to adopting AI solutions are very low, adoption could be very quick.
Here are just a handful of use cases that are in operation today in the industry.
- Reserves Analysis. The upstream is advancing very rapidly in the use of AI, particularly in the subsurface area. Data is abundant, the upstream field is already highly technology literate, and the value at stake dramatically outweighs the costs to conduct trials of new technology to interpret data.
Upstream companies use AI to improve their reserves understanding. When AI is applied to the interpretation and analysis of available subsurface data, recovery rates from resources improve. McKinsey estimates that AI boosts data analysis performance by 79 percent over humans. AI work is underway in conventional and unconventional resources, both on-shore and off-shore. For example, decline curves for shale resources have progressively improved over the decade but still trail conventional recoveries. The IEA forecasts that digital innovations such as AI applied to better understand porosity, permeability, fluid dynamics, and fracking performance would help to expand reserves by as much as five percent, principally in unconventional resources that are most susceptible to better analytics.
Geologists are also revisiting mature assets using AI, including resource data set aside decades earlier because the mathematics, the processing power or the compute environment had not been up to the task. Mathematics and processing power have sufficiently advanced that AI techniques are being successfully applied to the archived subsurface data and interpretation of long-discovered conventional basins to enhance the fidelity or precision of resources, enhance production and extend the life of existing wells. Data service companies are beginning to offer this as a service. For example, Enersoft’s collection of original wellbore data from 200,000 wells in Canada amounted to a quarter of a terabyte (or 250 gigbytes). Meanwhile, Enersoft’s robotic cuttings analyzer generates over two terabytes of data per well, producing resolutions that are orders of magnitude more detailed. In effect, AI allows geologists to model reservoirs at the level of a grain of sand. The amount of data now available for analysis and interpretation drives the need for new AI tools.
A recent innovation from Bluware applies streaming technology for data handling in the same way that Netflix or Spotify stream video and music data. Large subsurface data sets are compressed and “streamed” to an AI engine for interpretation without the recipient having to purchase the data or obtain a copy of the data. This development alters the concept of a restrictive data room used for mergers and acquisitions and allows many smaller players to pool their data to enhance the interpretation of their resources.
- Operations. Visual data interpretation improves safety outcomes, compliance to regulation and site operations. An AI service from Osprey Informatics interprets visual sensor data feeds and detects worker safety issues (not fitted out in safety gear, or smoking or in the wrong place) and sends alerts to supervisors. Its AI engine monitors plumes and detects the presence of harmful hydrocarbon vapors. The system raises site monitoring to 100 percent and is fully auditable, thereby raising compliance rates.
- Production. Customers of Ambyint, who supply an AI solution for artificial lift applications, report as much as a five percent gain in production and a 10 percent reduction in cost. Ambyint’s AI solution attaches to a pump jack, and optimizes the pump to eliminate over and under pumping activity.
Finally, AI makes small work of tasks that humans can do but simply take time. For example, Woodside Petroleum uses IBM Watson to catalogue all the previous engineering studies and documents about Woodside’s gas projects off the coast of Western Australia. The engineers pose engineering questions to Watson in natural language, Watson interprets the question and then presents the studies, rank ordered by best fit. Woodside estimates that prior to Watson, their engineers spent as much as 40 percent of their time searching for previous studies and analysis. This captured time is now available for more productive engineering work by reducing search times by 75 percent.
- Midstream. Midstream activities also lend themselves to new intelligent solutions. For example, tank farm operators and pipelines build digital twins of their complex network assets and apply artificial intelligence engines to optimize the network, using solutions from companies such as Stream Systems.
A fully functioning digital twin of a business includes many layers of data that work together to provide a rich, fully integrated and analytically deep software version of the business. These layers include engineering content (diagrams, specifications, configurations), physical constraints (operating capacities, throughputs and pressures), operating parameters of the assets (input energies, consumables, byproducts and emissions), financial features (fixed build cost, operating cost per unit), and uncertain elements (customer demand, weather events, supply disruption). Applying AI technology to this digital model of the network lets owners optimize the assets in ways not previously possible or simulate an asset’s lifetime. For example, Microsoft used pipeline data in a hackathon to see if an AI tool could be used to find corrosion. It was successful: the winner was able to find corrosion with 99 percent accuracy.
Asset maintenance is reconfiguring to take advantage of AI by introducing a new way for technicians to conduct work. Instead of capturing work details on paper and clipboards or typing on small keyboards to record asset and repair information, workers use cameras and microphones enabled with AI to record work in the moment. Data-driven research, enabled by AI, slices maintenance costs in half, according to Kimberlite research. The AI engine interprets the visual data and spoken words to both assist the worker and to create structured data for maintenance systems.
Carbon concerns, consumer preferences and regulatory compliance are motivating many industries to add greater intelligence to their equipment. Chief among these is the automotive industry, whose internal combustion engine technology consumes 25 percent of all petroleum. The top six automobile companies, who account for 50 percent of all industry sales, are racing to deploy AI in their products. Today’s vehicles are already rich in software. The Ford F-150, a market leading pick-up truck, has more lines of code in its various components than the Large Hadron Collider, the Space Shuttle, the F-35 Fighter Jet, or Facebook.
Next generation vehicles, which are progressively coming to the market, will incorporate greater levels of connectivity, sharing support and autonomy, elements which are dependent on artificial intelligence to operate. Early results from the use of autonomous heavy haulers in mining applications shows improved fuel usage as a key benefit, as the vehicles automatically adjust engine performance to match load and road grade features. Connected vehicles that can communicate with each other can form platoons or convoys of more closely packed vehicles. The aerodynamic effects reduce drag and increase fuel efficiency.
Moving to deployment
To successfully deploy artificial intelligence solutions, companies in oil and gas need to do three things.
1 | Robust Sponsorship and Support. Leadership must demonstrate unyielding support for AI, including such tactics as tuning performance metrics to reward behavior. Front line employees need to know that they have management support before they will invest much time in a new technology that is perceived to eliminate jobs. The success of AI in the field is not going to be based on how good the algorithms are but on how good management is at promoting change in the workplace.
2 | Education. Oil and gas companies must invest in raising the understanding of AI in their organizations. Similarly, AI providers must include training and education on how algorithms actually work, their self-learning abilities, and their limitations, to help overcome engineering suspicions about the technology.
3 | Collaboration. Proponents must work more collaboratively to overcome skills shortages. The demand for AI know-how has accelerated in many industries, triggering high demand for skilled talent in data science, machine learning, deep learning, neural networks, analytics and artificial intelligence. Working closely with labs, start-ups, incubators and accelerators is a necessity.
AI is now sufficiently advanced that some of the industry’s most difficult challenges are now impacted by this technology. There is no longer any reason to delay investments in this promising and mature technology.
Geoffrey Cann has thirty years of experience advising oil and gas, energy, and technology companies and specializes at the intersection of digital technologies and the oil and gas industry. He teaches regularly at the MBA level on energy issues.
He’s the author of Bits, Bytes, and Barrels: The Digital Transformation of Oil and Gas (MADCann Press,2019).