Preventing Stuck Pipe: A New, AI-Based Solution


Imagine this scenario: You work for an exploration and production (E&P) company on a drilling rig — typically in high-cost, high-risk environments. Your reservoir group has discovered a number of untapped targets to extract petroleum from. However, several mechanisms including wellbore stability, complicated well trajectory, challenging drilling fluid requirements, differential sticking forces, and unexpected changes in pore pressure increase the probability of the drill string getting stuck during operations. Suddenly, the drill string is stuck, drilling operations come to a halt, and your team must figure out how to free the drill string. Your drilling crew has encountered an unfortunately common — and costly — drilling dysfunction known as stuck pipe.

What Causes Stuck Pipe?

petroleum rig

Accounting for billions of dollars annually and up to half of the total well cost, stuck pipe events are one of the most costly drilling problems in the industry. It’s also a major contributor to non-productive time (NPT), which continues to comprise 20-25% of annual rig operating time. So, then, how do stuck pipe events occur?

There are generally two different types of stuck pipe: differential pipe sticking and mechanical pipe sticking. Both types can bring operations to a complete halt, but their causes are vastly different — and still result in costly repairs. Differential sticking occurs when the hydrostatic pressure of mud is greater than the formation pressure, pushing the drill string into a filter cake of permeable formation. When differential sticking occurs, it’s virtually impossible to move the drill string in an upward or downward direction, requiring operators to reduce the hydrostatic pressure by circulation. Mechanical sticking refers to other stuck pipe situations that are caused by factors other than differential pressure, such as wellbore instability and inadequate hole cleaning.

Both types impose significant amounts of unexpected downtime. For context, just two days of downtime means a loss of about USD $400K in today’s deflated market. Given this information, how can operators prevent stuck pipe events from occurring so they don’t have to deal with such exorbitant costs?

Current Solutions: How to Remove Stuck Pipe From Borewells

Regardless of how stuck pipe events occur, even the best freeing procedures are prohibitively expensive and require ample amounts of time. Depending on the type of stuck pipe, current solutions for how to remove stuck pipe from borewells include, but are not limited to:

  • Jarring downwards or upwards if the drill string has been pulled or run into a tight spot
  • Increasing pump rate to the maximum that’s operationally permissible to assist with hole cleaning
  • Increasing circulation rates to help enlarge the drilling hole
  • Reducing hydrostatic head and differential pressure by lightening the mud or circulating to a lighter fluid such as water or diesel
  • Applying various chemical approaches than can help dissolve the mud formation

Even worse, these solutions can potentially damage the equipment involved in freeing the drill string. In the past, drilling operators have leveraged various technologies such as physics-based modeling to help identify stuck pipe events. However, while physics-based modeling can provide a good foundation for using surface parameters to model torque and drag, physics models are unable to flag leading indicators of drilling anomalies with enough advance forewarning. Even just a few minutes of advance warning could be the difference between meeting drilling objectives and experiencing unexpected downtime.

New Solution: Machine Learning-Based Predictive Analytics

Today, there is a better, more efficient, and cost-effective solution to help prevent stuck pipe events before they occur. Outfitted with numerous sensors, drilling operations generate an abundance of historical and real-time operational data that enable operators to make better decisions and be more proactive in addressing all kinds of drilling challenges. However, the sheer volume of data is often far too great for human analysts to make sense of in the high-pressure environment and under the real-time constraints inherent to drilling activities. When dealing with impending stuck pipe events, accuracy and precision of just-in-time interpretations are critical.

A machine learning-based predictive analytics solution helps drilling operators better understand the state of their operations by unlocking opportunities within the data in real time. The basis of this solution is applying predictive machine learning algorithms to historical sensor data from assets used in drilling operations to build a baseline model of what normal operations look like. The normal behavior model is then used to analyze asset sensor data in real time, identifying and flagging any values that deviate from the established norm. These deviating values indicate anomalous behaviors that are likely to precede a production-impacting event such as stuck pipe.

We’ve leveraged this solution in previous deployments across the oil and gas industry to detect anomalous behaviors and prevent production-impacting events from occurring. One oil and gas supermajor turned to us to help maximize production potential and improve overall safety for its fleet of high-volume offshore platforms. Our machine learning-based predictive analytics solution helped the supermajor prevent unexpected failures in their fleet’s critical assets, including multiple glycol systems and export compressors that contributed to about 80% of the downtime on one of their marquee platforms. Overall, our solution improved production by up to 4%, or up to USD $30 million annually per platform, by increasing asset availability and uptime.

Drilling operators stand to gain similar benefits by adopting this solution to prevent stuck pipe events. In particular, a large drilling operator in the Middle East leveraged our machine learning-based predictive analytics solution to predict and prevent stuck pipe events. They deployed machine learning models across seven wells in different fields to help identify anomalies occurring during operations. This modeling approach mimicked the behavior of drillers using surface parameters, identified small deviations from normal behavior based on combinations of surface parameters, and leveraged automated machine learning to improve the model by exposing it to more data.

During this project, the models detected 79% of overall drilling anomalies, including stuck pipe events, with up to six hours’ advance notice. This allowed operators ample time to make corrections well before a stuck pipe event could occur, saving the company upwards of millions of dollars in associated drilling costs and downtimes.

In the case of stuck pipe, prevention is arguably far more economical than even the best freeing procedures. In a high-tech industrial sector like oil and gas, drilling operations are already outfitted with sensors that generate massive volumes of data. Machine learning-powered predictive analytics is the missing piece that enables operators to stay ahead of impending stuck pipe failures before it’s too late.

Learn more about SparkCognition’s solution for preventing stuck pipe.

Latest blogs

Abstract depiction of our Generative AI Platform created by DALL-E
Campbell LeFlore

What’s Inside our Generative AI Platform?

Perfected over ten years of real-world engagements serving many of the world’s largest brands in energy, manufacturing, transportation, utilities, financial services, and other industries, SparkCognition’s

Read More
Campbell LeFlore

The Top Challenges HSE Managers Face Today

HSE managers are the behind-the-scenes heroes enabling today’s high-throughput industrial environments to operate at peak performance. Their determined efforts ensure the well-being and protection of

Read More
SparkCognition is committed to compliance with applicable privacy laws, including GDPR, and we provide related assurances in our contractual commitments. Click here to review our Cookie & Privacy Policy.