A colleague recently described subsea operations as “trying to hit someone in the head with a baseball bat while it dangles from an airplane.” Drillbits that are only a few inches wide must travel thousands of feet under water and then drill tens of thousands of feet farther in order to reach the desired target. Although there are advantages to subsea production, like protection from natural elements and reduced load on floating platforms, it is difficult work.

At underwater depths of several thousand feet there is no visual on corrosions or leaks as there would be with onshore wells. The monitoring of operations is not as simple as sending a worker in a truck to check it out. If a potential problem arises, investigation involves specially trained teams of divers, ROVs and more, typically resulting in lost production time for an already staggeringly expensive venture (in the realm of millions of dollars per day).

In addition to conducting this precise work with little visibility as to how it’s going, there is essentially no room for error. The Deepwater Horizon blowout in 2010 demonstrated the vital differentiator of subsea operations. While all well failures are big and expensive hassles, subsea failures can be catastrophic. With such severe consequences reliable operational insights are of the utmost importance.

Recently, companies have been turning to artificial intelligence (AI) to provide visibility into how operations are progressing and to protect against disaster. AI has already been employed across several industries, saving millions of dollars in energy production thanks to failure prediction, providing more intelligent maintenance in aerospace and increasing viable production in manufacturing. AI technology holds particular promise for subsea fields in the form of predictive maintenance uses, the ability to streamline operations, and reduced safety and environmental cost implications.

Science of AI

At its core the use case for AI is relatively simple. It can analyze large amounts of historical data from various sensor sources (temperature and pressure gauges, flow rate, etc.) during previous behaviors (including normal operations and during failures) to find patterns. Because AI is capable of considering so many variables, it can find connections and indicators of failure that may otherwise go undetected.

Previously, subject matter experts would leverage their understanding of an asset to create a physics-based or statistical-based model that predicts a desired outcome; essentially it would be the data plus the program to determine the output (data+program=output).

Now, with machine learning, algorithms can derive a predictive model by simply learning from a combination of historical data and examples of the outcomes that one would want to predict; essentially it is the data plus the outcomes to determine the program (data+ outcome=program).

There are two main areas of AI that can be utilized to help the oil and gas industry.

Supervised learning can be used to build predictive models when there are sufficient examples of the outcome to predict. These algorithms derive the pattern of information that precedes the desired outcome.

Unsupervised learning can be used to find anomalies when there are insufficient examples of the desired prediction. In these cases, the model does not look for the desired outcome but instead can be trained to determine what is “normal.” Then the software can alert users to any issues that are “not normal” or anomalous.

Finding anomalies indicative of a deviation from normal or making a direct prediction of a desired outcome can have significant implications for an oil and gas company in terms of improving efficiency, production and safety.

Aiding decision-making in the drilling process

Efficiency has always been a focus for oil and gas companies, particularly in recent years. This is especially true for subsea operations due to high cost and has driven operators to look for efficiencies whenever possible.

Subsea drilling is completely human-driven from the ocean surface: One operator ensures the drill follows the path predetermined by a computer while another controls the drill.

However, AI can aid subsea processes by finding the most efficient path to drill to a reservoir. As a drill goes through different formations, the technology could regulate weight on bit in terms of speed, torque or other variables during lengths of hard or soft rock to avoid diverting from the drilling path while maintaining the correct angle. This kind of real-time feedback helps to avoid human error while optimizing the drill path for maximum energy efficiency.

Optimizing pump configuration, predicting failures

Production also can benefit from AI’s ability to analyze large amounts of data and build models from the information. Based on historical data, production models can be developed for each well and reservoir. The advanced analytics of AI can produce a much clearer picture of production at each well, and this information can be used to decide what types of pumps should be used and how many should go in each well.

Predicting failures also leads directly to increased production. For example, Fereidoun Abbassian of BP recently mentioned that the company has completely eliminated stuck-pipe instances since implementing Well Advisor, an advanced analytics platform. With each instance costing $10 million to $50 million, BP estimates this implementation is saving the company $100 million/year.

Forewarning to avert disaster

Another major opportunity for AI is in the detection of incidents like well kicks. From an environmental and safety perspective, the unwanted entry of fluids into a reservoir can be disastrous as it can spill flammable material on the rig and into the ocean. With less sophisticated methods of detection a kick is only evident when these fluids actually reach the rig. From there the time line to resolve the incident before disaster occurs is limited, meaning more drastic measures like blowout prevention (which will wipe out all progress on the well and result in hundreds of millions of dollars in lost production) might be necessary.

AI technology can provide failure prediction further in advance than traditional methods of monitoring. It can monitor downhole sensors and inform the surface-level crew if conditions indicate a well kick. This allows the crew to make remedial decisions more intelligently since they have more information and more time to solve the problem. In addition to preventing downtime, the increased forewarning could save lives.

A more intelligent future

Although not all of these applications are flawlessly in use today, they represent the future direction of the industry. Early successes have reinforced faith in the abilities of AI, and as the field expands, more use cases are being explored. AI is defining its place in oil and gas, and the unique considerations of subsea operations are allowing for the possibility of progressive changes in the field.