Oil and Gas companies have long prided themselves on being front runners when it comes to the application, deployment, and utilization of the latest technologies to push through new frontiers. One only has to look back in recent history and see how far and how fast technology has driven oil companies to operate in environments neither feasible nor possible 30 years ago. I can remember in the early ‘90s when Shell had their first TLP (Tension Leg Platform) in the Gulf of Mexico in just over 2,000 feet of water, it was seen as groundbreaking. And now, companies can operate in depths significantly greater than that. Same with other aspects of oil & gas operations, Reservoir Modeling, Drilling and Completions, logistics; the list of achievements is indeed long.
Given everything we have stated, it is all the more surprising that Oil & Gas companies have been slower to adopt potential game-changing AI applications and platforms compared to other, traditionally more staid industry sectors.
It’s not all doom and gloom. There are examples of organizations starting to utilize AI to deliver potentially significant results. Shell is using AI and machine learning for precision drilling and recently adopted reinforcement learning, providing additional control mechanisms for drilling equipment. Reinforcement learning employs a reward system, which is “dependent on the outcome of the AI’s choices’.”
According to Daniel Jeavons, Shell’s general manager for data science, with reinforcement learning, “the key thing is you’re giving the [AI] agent the autonomy to make the decision. But you’re providing input into the model, so you’re providing reward or penalty functions on the basis of what’s happening in the model and how the model responds to the set of conditions that you give it.” Algorithms are trained on “historical data from Shell’s drilling records” and from simulated exploration and then used to “guide the drill.” The benefit gained from this AI platform is that the human interface (Drilling Foreman/Supt/Driller) a significantly improved view of the operating environment, leading to “faster results and less wear, tear and damage to machinery.” I do not doubt that for the foreseeable future, AI will not replace the more traditional composition of drilling teams completely. However, it will allow the Drilling Engineers, Supt/Foreman, to support multiple events.
Like Shell, Exxon has also rapidly made inroads into the realms of AI. They have been busy deploying AI to consolidate their data silos into one easy to access seamless repository. An effort that will potentially payoff by turning a “grueling, year-long process of churning through 2D seismic maps, tectonic and historical data into a six-month play that would detail the potential payoff of new hydrocarbon fields.” When Exxon Mobil invested in Guyana for offshore oil exploration, the company implemented its new AI-enabled data platform accelerating project development and reducing the ROI cycle. The AI-enabled platform allows experts to access data from multi-cloud applications, providing an environment to make decisions more rapidly and with increased levels of confidence. Project leader, Xiaojun Huang, stated, “any team member can collect data from any application from any source and make it available seamlessly through APIs.” Benefits realized through the shortening of planning cycles to seven months instead of nine for new well designs and 40% savings on data preparation time due to more agile processes.
TRENDS ON THE FUTURE USE OF AI IN THE OIL & GAS INDUSTRY
We have long been aware that Oil & Gas companies have been having conversations on how they can optimize all aspects of the hydrocarbon value chain using AI; unfortunately, action on the topic has been a little more scarce! Even when organizations have been successful in moving AI initiatives from the drawing board into the Field, it is baby steps rather than giant leaps. Further, it has been a frustratingly slow process. Part of the issue is the lack of a clear AI strategy; deployment has been tactical and “local” rather than strategic and company wide. Another factor is that we don’t yet see these conversations taking place at the C-Suite; instead, it is mid-level and senior managers who have been demonstrating the most passion and understanding of the value of AI.
And there is considerable value to be extracted from the utilization of AI, and especially for smaller operators. Simple things like predicting well throughput, or likely upcoming downtime events, information that is readily available within seconds through AI applications, to any person in the organization. Why would we need Field Operators to drive to well sites to only set “eyes on a well?” It no longer makes sense, from a safety, operability, nor economic perspective. Companies need to quickly jump on upcoming trends on the use of AI, like developing neural networks to help engineers understand how the yield curve on a well will change over time with significantly higher levels of accuracy and confidence. For companies operating mature assets in mature fields, this is a game-changer. An abundance of high-quality historical data results in a higher likelihood of obtaining a good model. By using numerous data points over a long enough period, we can use AI to determine how a particular well or group of wells will behave based on available inputs.
According to Rystad Energy’s senior analyst on shale, Alexandre Ramos-Peon, with enough data, the computer will “train itself somehow to guess the best value, with the goal of keeping the accuracy as high as possible.” Providing engineers and operators the ability to predict well downtime and potential down-hole failures long before they happen, thus preventing downtime, lost production, and cost to repair.
When sufficient informative historical data is not available, and speed of deployment is a priority, implementation can be accelerated by combining mechanistic models (aka Digital Twins) with AI models. This hybrid approach -which also increases the confidence in AI and accelerates adoption -is key to successful reinforcement learning and allows users to build autonomous operations for complex assets. According to Rajiv Anand, CEO of Quartic.ai. “Predictive reliability and loss of function risk of critical assets like downhole ESPs which were previously difficult or near impossible have now been enabled with AI models by companies like Quartic.ai.” Midstream producers are using AI for dynamic efficiency measurements of compression and transportation. This allows them to be more deterministic in meeting service levels and increasing asset utilization.