Skip Navigation

Failure is not an option

Skip Side Navigation

Failure is not an option

Employees in a real-time center view live well data to help customers make better decisions on drilling parameters. (Source: NOV)


The drilling industry has changed considerably in the more than 150 years since the “Drake Well” was drilled, but the principle is still the same: crush rock to make a big hole in the ground. Wells have gotten exponentially deeper and more complex, necessitating significant technological advances to allow production of a commodity that is still the cheapest source of energy by far. Oil and gas drilling is also generally done in remote locations, like the middle of the ocean or desert, which makes connectivity a challenge when thinking in terms of data and information.

Other unique challenges have slowed down the drilling industry’s progress when it comes to taking full advantage of data or Big Data. Looking at other industries that were able to take advantage of their data, like the automobile industry, the clear differentiators are the volume of data and ease of access. When oil was at its peak in 2014, there were around 1,900 rigs operating in the U.S., which is a huge number for the oilfield; in comparison, Ford sells around 2,500 F150s per day. So, a data scientist wanting to build training data for machine-learning engines for failure predictions will probably want to do it on the F150s. Not only is the sample size significantly larger for the F150s, but they also generally operate under the same conditions, with variance only occurring as an outlier. While all drilling rigs are also effectively doing the same thing, crushing rock and pushing it up to the surface in Oklahoma is very different from crushing rock in North Dakota.

Sensors are necessary for data acquisition, which is central to the concept of Big Data. The challenge is that sensors must be able to survive the harsh environments and conditions—like high pressures or temperatures—encountered when downhole. The shock and load that the equipment a drilling rig sees is extreme, so companies that build IoT sensors for appliances that sit in a climate-controlled environment might be unable to produce sensors durable enough to use in the drilling industry. The other component of data acquisition is the human side of it, and there must be a lot of discipline and focus around how the data is entered. In research that NOV did in 2009 with drilling rig data from five drilling contractors and 100 rigs, there was a definite trend on the largest cause of downtime—it was marked as “Other,” while the second largest cause was “N/A” (see Fig. 1). Ideally, the people involved would have more accurately identified the downtime with a specific root cause, which would have enabled massive reductions in downtime. Fortunately, companies and software solutions are making it harder to input bad data, which has led to a substantial reduction of bad data on that front. The current generation of control systems are also smart enough to provide much of that data automatically without human input; that said, data quality is still something that should always be a focus point.


Fig. 1—Earlier research from NOV showed a clear trend in downtime cause, but ambiguity in the data made it difficult to understand the actual problems.


The above challenges notwithstanding, the past decade has yielded significant developments in machine learning, which are helping the industry better predict failures and build autonomous rigs. Anomaly detection helps companies manage exceptions to patterns, much like how credit card companies manage fraud. This makes having a data scientist looking at such data all the time unnecessary, as systems start to build patterns on spending habits and generate a human phone call when an anomaly is encountered. Similarly, control systems learn from supervised and unsupervised learning engines to understand patterns and from the feedback generated from anomalies to reinforce the models. Similar advancements in deep learning take full advantage of the Big Data infrastructures in place to correlate seemingly insurmountable chunks of data to build prediction models. Real-time technology centers are built around these technologies to monitor the rigs and their data to help reduce or prevent potential downtime

The key is understanding the challenges the industry faces and surmount them, and a significant one is connectivity. There usually isn’t a fiber line going to a rig in the middle of the ocean, so sending all the data that is collected back to where the Big Data infrastructure is located is always going to be a challenge.

Ultimately, everything is in place to make machine learning, artificial intelligence, and Big Data work, and using these advancements in technology is critical to improving reliability and uptime. When something fails on a rig, it costs millions of dollars a day in downtime, and worse, it could potentially damage the environment. Humanity sent a man to the moon and communicated with him 50 years ago—there’s no excuse not to make digitizing the oilfield and accessing data from remote areas a reality.