Data mining/analysis

Halliburton’s Chief Data Scientist Speaks on Analytics Projects That Are Solving Big Problems

The oil and gas industry has a lot to gain from the adoption of big data analytics as recently highlighted examples from major service company Halliburton demonstrate.

2017-jpt-gettyimages-182208553.jpg
Pumps, bits, and drilling are among the areas the world’s second largest service company is working to improve through the use of big data analytics.
Source: Getty Images

As big data analytics becomes more intertwined with the operations of some of the biggest names in upstream, this tech space is beginning to deliver the industry some compelling applications. The most recent developments were highlighted in a discussion led by Satyam Priyadarshy, Halliburton’s chief data scientist, who declared that big data analytics “will transform the industry.”

That transformation is to be realized by using analytics to identify and eliminate hidden inefficiencies. According to Priyadarshy, that means that solving known problems is not enough; to achieve a higher order of innovation, oil and gas companies must discover and answer problems they never knew about in the first place.

Speaking at the first SPE Data Analytics Study Group in Houston, Priyadarshy explained how Halliburton analyzed data from 5,000 wells and found that a specific pump model was suffering from an inordinate amount of downtime. Hidden within the data was the simple fact that the pump was not well-suited for one particular climate zone.

“Now we can actually predict that a certain pump, of a certain manufacturer, should never be sent out to a certain area,” said Priyadarshy, noting that the cost of the analytics platform has already been recovered by the revenue earned from keeping the pumps running longer.

Priyadarshy described another analytics program that, when told a specific cutter on a drill bit has failed, will show the drilling crew which of the next cutters is most likely to fail next. Presumably, such information could influence a driller to change tactics to extend the life of the bit, and therefore reduce the number of time-consuming trips out of a well to retool.

The company is also building an “Earth model” that will simulate the Earth’s subsurface using industry data from the fields of geochemistry, fossil research, and geology. Priyadarshy said Halliburton has 150 PhDs working on this initiative.

Fixing Stuck Pipe

In regard to solving big problems, Halliburton may be close to solving the drilling issue known as stuck pipe. Among the oil field’s oldest sources of cost overruns, stuck pipe involves a situation in which a drilling string cannot be pulled from the borehole without leading to damage.

Using an historical data set as an example, Priyadarshy said analytics could have saved one anonymized operator USD 17 million out of the USD 22 million it spent on stuck pipe issues during a multiwell drilling campaign. “That is the value of the data that was just sitting in somebody’s warehouse,” he said, emphasizing a point that though the industry collects big data, it rarely uses it.

Priyadarshy said the program took only 6 weeks to train and works by analyzing what he called “drilling language.” This variation of natural language machine learning takes the unstructured text data found in the human-written drilling reports and connects the dots in a way that is seductively simple.

In this case, the dots are represented as three categories on a bar chart: symptoms, actions, and the event—a stuck pipe. “From that, we can calculate how much money [the operator] lost because of certain actions and symptoms,” said Priyadarshy.

The next phase of this project is to build a predictive model that begins looking for symptoms and actions before the drilling report is even written up. Priyadarshy said this will involve creating a way for the drillers to input more meaningful observations than are currently included in a typical report.

Get Educated and Find Help

Priyadarshy also spent time remarking on the factors he sees holding back the industry from a faster uptake of analytics. Of his biggest concerns is that too many people in the industry simply think of big data as “a lot of data,” which he called a “fundamentally flawed definition.”

His definition: “Big data is about all the data of the business, which means the historical, current, and future data that you will generate.” This understanding falls in line with a concept that might be described as total-analytics, in which companies allow the entirety of the their data to be placed under the microscope in order to drive out wasteful workflows and unnoticed problems

“The reason [industry workflows] are inefficient is because whatever inefficiencies we knew about, we have taken out, and beyond that, we can’t see because we haven’t yet looked for them,” he explained, before adding later, “The only way you can actually be innovative is by looking holistically at the problems of the industry through big data analytics and by building new intelligence that is based on the hidden inefficiencies.”

Aside from the steep learning curve, the oil and gas business is facing a major challenge with the talent gap. Priyadarshy, who joined the industry only 3 years ago after spending most of his career in the academic and software arenas, cited figures from a McKinsey Global Institute report that estimated the US will face a shortage of 250,000 data scientists by 2024.

In addition to the shortage, he noted that the oil and gas industry is competing for this talent with startups and large tech firms in places such as California, New York, Washington, D.C., and Austin.

Priyadarshy, who is in the midst of building the fourth data science team of his career, strongly advised that when upstream firms acquire data scientists, they should not micromanage them. “If you want to grow a team, you have to really know how they think and give them their space,” he said. “The moment you encroach on that space, they will move. That’s the simplest thing I can tell you.”