Artificial intelligence applications promise improved drilling efficiency

Dec. 11, 2023
Research suggests that AI tools can help predict, assess downhole conditions

Editor's note: This cover story first appeared in the November-December 2023 issue of Offshore magazine. Click here to view the full issue.

By Bruce Beaubouef, Managing Editor 

Artificial intelligence is increasingly being examined by the upstream oil and gas industry as a means of improving drilling efficiency and lowering completion costs. Researchers, vendors and operators are studying the possibilities of a range of AI applications that include drilling parameter optimization, downhole environment detection, intelligent completion design, and more. 

Numerous studies have focused on intelligent algorithms and their application. Advanced technologies, such as digital twins and physics-guided neural networks, are expected to play roles in drilling and completion engineering.

And there has been a rapid increase in the number of AI tools deployed within the petroleum industry in recent years. AI has been utilized to tackle numerous challenges within the oil and gas sector, including seismic pattern recognition, reservoir characterization, permeability and porosity prediction, PVT (pressure-volume-temperature) properties prediction, drill bit diagnosis, pipeline and well pressure drop estimation, oil well production optimization, oil well performance, portfolio management, and general decision-making operations, among others.

Recently, oilfield equipment supplier WWT International laid out its views on how AI technologies can advance the science and operations of oil and gas drilling. With the advent of advanced algorithms and machine learning technologies, drilling operations have become more efficient, safer, and cost-effective. The company sees the following four areas where AI applications could be used to advance drilling safety and efficiency.

          Predictive maintenance. One of the most significant ways AI is transforming the drilling industry is through predictive maintenance, WWT says. Drilling equipment is subject to a lot of wear and tear, which can result in costly breakdowns and delays. AI algorithms can analyze sensor data from drilling equipment in real-time and predict when maintenance is required, helping to reduce downtime and increase efficiency. By identifying potential problems before they become significant issues, AI can help companies save time and money while improving safety.

            Automated drilling. Another way AI is changing the drilling industry is through automated drilling, says WWT. Automated drilling systems use AI algorithms and machine learning technologies to control the drilling process, from drilling direction to mud pressure and torque. These systems can automatically adjust drilling parameters based on changing conditions, such as rock hardness, and optimize the drilling process in real-time. By eliminating human error and improving drilling accuracy, AI-powered automated drilling systems can improve drilling efficiency and reduce costs.

            Optimized well planning. AI may also revolutionize well planning by enabling more accurate and optimized drilling paths, according to WWT. By analyzing geological data and drilling parameters, AI algorithms can generate detailed 3D models of the subsurface, which can help companies plan more efficient drilling paths. These optimized drilling paths can reduce drilling time, minimize the risk of drilling problems, and ultimately lead to significant cost savings.

          Improved safety. Finally, WWT says that AI can improve safety in the drilling industry by providing real-time monitoring and analysis of drilling operations. AI-powered sensors can detect potential hazards, such as high levels of gas, and alert workers to take necessary precautions. AI can also analyze worker behavior to identify potential safety issues and suggest corrective actions. By providing real-time safety monitoring and analysis, AI can help companies prevent accidents and improve worker safety.

As AI technology continues to evolve, WWT says that it believes that the possibilities for its application in the drilling industry are endless. In the future, the company says that the industry can expect to see even more advanced AI-powered drilling systems that can make drilling even more efficient, safer, and cost-effective. For example, AI algorithms could help predict well productivity or optimize the production process based on real-time data. AI could also enable real-time drilling data analysis and interpretation, helping companies make informed decisions about drilling operations.

Research efforts 

Researchers say that there are several data parameters in the drilling process that can benefit from AI and machine learning applications. These include:

  • ROP prediction 
  • Drillstring vibration prediction
  • Lost circulation prediction
  • Pipe sticking incident detection 
  • Gas influx detection
  • Drillstreng washout detection
  • Abnormal drilling detection
  • Drilling fluid design 
  • Integrated drilling optimization. 

There have been a number of studies over the past few years that have refined and advanced these concepts using real-world data. A select few of these efforts are highlighted below.  

Lost circulation prediction 

In 2021, several experts from Saudi Aramco’s drilling technology team, recognizing that drilling fluid lost circulation incidents (LCIs) are major sources of non-productive time, proposed that machine learning and deep learning classification algorithms could be used to process time-series data, and achieve early detection of such temporal phenomena. In their IEEE paper, “Deep Learning and Time-Series Analysis for the Early Detection of Lost Circulation Incidents During Drilling Operations,” the writers noted that LCIs are typically monitored at the rig site by observing drilling fluid levels in the fluid tanks. They pointed out that this manual process often entails many inefficiencies, including missing the occurrence of LCIS, or late detection. By performing a large-scale analysis of the surface drilling and rheology data obtained from historical wells with LCIs, the authors concluded that convolutional neural network models performed best in the early detection of LCIs. They noted that although the model was derived to detect severe/total fluid losses, the model that they developed was able to detect signs leading to seepage or partial losses.

Pipe sticking incident detection 

In 2020, Haytham H. Elmousalami and Mahmoud Elaskary of the General Petroleum Company of Egypt set out to develop a reliable classification model for stuck pipe incidents. They examined the possibilities for using machine learning algorithms for efficient predictive analytics, optimization, and decision making. To this end, they analyzed datasets from several drilling jobs carried out in the Gulf of Suez. Data from these wells was collected to create a real-world dataset for analyzing machine learning performance. In their paper “Drilling stuck pipe classification and mitigation in the Gulf of Suez oil fields using artificial intelligence” (Journal of Petroleum Exploration and Production Technology: March 2020), Elmousalami and Mahmoud Elaskary showed how the generation of stuck pipe classifications could be automated using machine learning algorithms, and how this in turn could help mitigate stuck pipe incidents. Twelve machine learning techniques were used to create stuck drilling pipe classifications using artificial neural networks, logistic regression, and ensemble methods such as scalable boosting trees and random forest. Out of twelve AI techniques, their work found that the most reliable algorithm was extremely randomized trees (extra trees), which produced 100% classification accuracy based on the testing dataset.

Real-time gas influx detection 

In 2019, researchers with the China University of Petroleum in Beijing examined the possibilities for automatic gas influx detection in offshore drilling based on machine learning techniques. The driver for this research effort were the influxes that had occurred during exploration drilling activities in the South China Sea, and the threat that they posed to well control strategies. The researchers noted that the traditional detection method had been to analyze the mud-log of mud tanks, a process that was always slow, with a time lag between the gas influx occurrence and the influx detection. In their paper “Automatic Gas Influxes Detection in Offshore Drilling Based on Machine Learning Technology,” the researchers proposed a methodology that allowed for real-time gas influx detection using artificial intelligence and data analytics. To develop the methodology, twelve mud-log parameters were collected for 208 influx incidents on 62 drilled wells in the South China Sea. Data analysis was performed on these incidents, using an artificial neural network, to develop a gas influx warning system model. From that model, a new comprehensive real-time gas influx detection method was developed. The research was one of the first attempts to develop a real-time gas influx detection system utilizing data analysis from mud logging in tandem with artificial intelligence. As detailed in their paper (SPE-198534-MS), this methodology was successfully applied to a gas field in the South China Sea with accuracies up to 95%.

Abnormal drilling event detection 

In 2019, a group of researchers at China University of Petroleum (Beijing) employed machine learning to tackle the problem of detecting abnormal drilling events. They analyzed the logging data of several Bohai Sea oil wells in order to better assess the downhole conditions of these wells. As detailed in their paper “Combining Drilling Big Data and Machine Learning Method to Improve the Timeliness of Drilling (SPE-194111-MS),” the team used artificial intelligence to review 200 sets of logging data from the Bohai Sea oil wells. The researchers found that the use of AI and machine learning could significantly reduce the amount of time traditionally required for assessing drilling conditions, as compared to manual reviews of the logging data. With the model they developed, the researchers found that they could accurately determine the working conditions of a well, and whether there had been some type of downhole impediment, with a relative error rate of less than 5%.

Equinor’s AIM project 

More recently, Equinor has been using AI to improve its well planning process. To that end, the company says that it is implementing a project called AIM – Artificial Intelligence Maturation. Officials say that previously, planning for even just one well was a months-long process, and required many time-consuming manual tasks. With AIM, Equinor says that it can quickly develop multiple field development scenarios with thousands of well alternatives. The AI can auto-generate recommended well paths from which engineers and other specialists can choose. Equinor says that “the expert is still best at making the interpretations,” but says that the computer can increase data coverage, build new models for prediction, and help validate concepts and ideas better and faster than before. “If we can better represent the complexity of well planning including the available slots, the targets, and all the constraints, then this means value creation and efficiency with our precious time,” says Subhro Sinha Roym subsurface project leader with Equinor. “For me, in AIM, that means working across subsurface and drilling and well challenges – I have to understand each other’s data and knowledge in order to get the cheapest and least risky well paths.”

Challenges ahead 

While the opportunities for improved drilling efficiency through the use of AI are promising, several challenges remain. Most of these challenges relate to data quality, model accuracy, and reliability. There are questions about AI’s ability to automatically – and accurately – process multi-source and multi-scale data. Additionally, there are several untested research frontiers in the arena of AI-informed drilling and completion. Researchers say that future studies will likely focus on the fusion of data-driven and physics-based models, small sample learning, uncertainty modeling, and the interpretability and transferability of intelligent algorithms.