Directional drilling is a challenging task even during the best of conditions, with many aspects described as more art than science. Successful and cost-effective directional drilling often comes down to an expert who has a good feel of how to navigate thousands of feet underground, site unseen. Therefore, the art of drilling often comes down to the artists, who are in high demand and short supply.

To address this scarcity, improve the economics and deliver consistently better outcomes, Oceanit was challenged to develop an artificial intelligence (AI)- driven system that performed as well as the experts and could augment or replicate their capabilities. The result is a system that, so far, delivers results within a 1.5% margin of expert drillers. Oceanit’s AI drilling system is playfully referred to as “Deep Thought,” a reference to the fictional computer in The Hitchhiker’s Guide to the Galaxy.

Deep Thought is capable of continuously learning and improving on its directional drilling decision-making. The scalable capabilities it provides can significantly reduce costs for operators while increasing high-performance outcomes through its reinforcement learning, replication and optimization of drilling mechanics.

Minimizing deviation, tortuosity

Working with Shell International Exploration, Oceanit developed Deep Thought to optimize value in directional drilling by minimizing deviation from planned wellbore trajectory, minimizing tortuosity, maximizing the ROP and reducing the number of personnel onboard, all factors that have serious impacts on the bottom line.

With directional drilling, expert teams maintain a stationary drillstring at the surface to achieve a curved hole. There are two main categories of systems used in directional drilling: bent-sub downhole motors and rotary steerable systems. Oceanit focused on the prior, as bent-sub downhole motors are generally more costeffective and prevalent in shale plays.

It is incredibly challenging to control the angular orientation of the drillbit toolface while ensuring adequate ROP. Eliminating trajectory deviations and the associated, costly corrective measures are imperative for improved outcomes. To reduce deviations, Oceanit used machine learning techniques in training Deep Thought to replicate the decisions of expert drillers.

Machine learning is the endowing of computers with the ability to learn without explicit preprogramming. Deep Thought uses artificial neural networks (ANNs)—a subset of machine learning—which are synthetic models of the function of biological neural networks. ANNs pass data through the network, with neurons accumulating patterns as the data flow.

A multilayer perceptron neural network is a class of ANNs that uses a supervised learning technique called backpropagation for training and can distinguish data that are not linearly separable. (Source: Oceanit)


Oceanit’s work with Shell was broken down into three primary tasks to achieve an optimized AI:

1. Information formulation, including operator engagement and data preparation;

2. ANN construction, including data immersion/analysis, evaluation and drilling simulation reinforcement learning; and

3. Supervisory mode development for augmenting the capabilities of directional drillers.

A computational model for drillstring physics was used to simulate the mechanics of directional drilling and collected data were filtered and used to structure and train Deep Thought to select appropriate drilling actions.

Machine learning

After the initial structuring, reinforcement learning methods were employed to refine the neural network behind Deep Thought. Historical directional drilling data, including that from MWD, were compiled from 14 horizontal wells in Appalachia and the Permian Basin. In addition to time/ date stamps, the data included bit depth, hole depth, hook load, weight on bit, differential pressure, gravity toolface, magnetic toolface, toolface angle, ROP, rotary rpm, rotary torque, standpipe pressure, total pump output and other more extraneous categories. The datasets were then unified and cleansed for use in training and validation.

Deep Thought was iteratively trained, and the parameters that govern learning were optimized. The result was an AI system that does not simply repeat back past results but generalizes from those instances to be able to suggest actions in new situations correctly.

As Deep Thought mastered the decision-making process in both new drilling scenarios and past scenarios, it improved upon its actions taken in response to complications, including changes in borehole friction, mud motor stall, the weighting of the bit during drilling and torque effects from the drillstring itself and low bit-rate telemetry. All of these factors are critical to preserve well trajectory and eliminate the need for corrective measures that add to well costs.

Next steps

Oceanit’s Deep Thought AI directional drilling system has demonstrated competence in historical and simulated realms and will be further trained and tested as a real-time advisory system for control of directional drilling operations.

Before moving to the field, the system will be further tested in simulation with expert directional drillers from Shell. Successful tests of Deep Thought will be those that augment a directional drilling operator’s performance and speed, enhancing their decision process and improving overall ROP.

The largest market for Deep Thought is shale wells, which are mostly horizontal and drilled with motors. The AI system will extend motor life, improve the ROP and deliver higher target accuracy.

Ultimately, Oceanit envisions Deep Thought being directly integrated into drilling operations, enabling fully automated directional drilling and addressing the scarcity of directional drilling experts and additional personnel. This will address the need for improved economics and consistently better outcomes.