AI/machine learning

Cognitive Computing: Augmenting Human Intelligence To Improve Oil and Gas Outcomes

The goal of cognitive computing is not to eliminate humans but allow highly skilled professionals to spend time doing what’s most valuable for the company.

twa-2014-3-forumhero.jpg
Source: Getty Images.

The severe downturn in oil prices over the past 3 years has made business transformation complicated for the upstream sector. Early in the downturn, companies were focused on cutting costs and restructuring for future growth. Many organizations are now taking advantage of gradually increasing oil prices to launch digital transformation efforts.

According to the International Data Corporation (IDC) 2018 Oil & Gas Predictions Report, by 2020, 80% of large oil and gas companies will run their business with help from cognitive/artificial-intelligence (AI) agents. The report found that 62% of users say outcomes from cognitive initiatives exceed their expectations.

Cognition quite simply refers to thinking—and cognitive systems such as IBM Watson can understand, reason, learn, and interact with us. Cognitive systems excel at understanding natural language, pattern identification, and knowledge location, and have endless capacity. This allows humans to focus on interpreting, analyzing, and adjusting designs, plans, and activities, and make decisions based on the data provided.

For example, a company’s cognitive system can focus on finding geohazards before drilling offshore. The system does 6–8 weeks of manual research in seconds, identifying specific geohazards buried within tens of thousands of pages of drilling reports, and dynamically converts text into easy to understand tables and graphs highlighting areas of interest. The goal is not to eliminate humans but allow highly skilled geoscientists and drilling engineers to spend time doing what is most valuable—defining the safest, most cost-effective drilling plan. This is why IBM prefers to reference AI as augmented intelligence because it augments or improves upon the expertise, capability, and potential of the decision makers and teams.

Cognitive systems differ from traditional programmed systems that provide predetermined outcomes based on specific rules. They consume all types of information from structured to unstructured and historical to real time. These technologies include but are not limited to:

  • Natural language processing
  • Predictive analytics
  • Recommendation engines
  • Robotic automation
  • Machine learning systems

Cognitive systems are adding value to oil and gas companies around the globe in multiple functions. Here are a few samples.
Near-real-time analytics identifies underperforming wells. A global oil and gas company set out to improve its rate and phase calculations using analytics to optimize oil production and maximize revenue streams. With near-real-time data from well sensors, the analytics solution rapidly executed a set of fluid rate and phase calculations to detect subtle changes in pressure and temperature. An imbalance/out-of-tolerance triggered an automated alert to the operations center, allowing the company to make adjustments, as necessary, quickly. This led to $11 million uncovered in revenue opportunities, 99% faster execution of rate and phase calculations, and 97% accuracy in detecting underperforming wells, allowing the company to make adjustments.

System responds to natural language questions. An Australian oil and gas producer tackled the challenge to make 30 years of project data available to employees using natural ­language questions. The company had more than 30,000 documents relating to their historical projects, which would take an employee over 5 years to read. The company trained the cognitive system on approximately 1,200 questions across a wide variety of project and engineering-related topics. When a junior engineer could not get the answer from the senior engineer, he asked the system the same question in natural language. “What is the maximum backpressure on the high-pressure flare line?” The system returned a set of ranked answers and he found the answer in the top 3, with supporting information/data—avoiding the need to run an expensive simulation. All employees can now access a vast amount of knowledge, changing the approach and pace of learning.

Adding unstructured data increases model sensitivity. A US oil and gas producer needed to reduce nonproductive time (NPT) while drilling. The ability to predict trouble during drilling could significantly reduce NPT. The team developed predictive models based on available structured data such as hookload weight and depth. Available unstructured data such as drill reports were analyzed to identify concepts that aligned with drilling events. The predictive model and text insights were aligned in time series to improve the prediction of the event. The model incorporating unstructured data outperformed the model based only on structured data with a modest improvement in accuracy and a more substantial improvement in sensitivity.

Maximizing Investment

When it comes to accelerating their organizations’ digital transformation, companies are looking outside the normal one-off proof of technology or project approach. An approach gaining momentum is joining various consortia or innovation programs.

The technologies enabling cognitive computing are advancing rapidly and many organizations are already seeing benefits. While many are starting their cognitive journey, industry leaders are accelerating their pace. A recent study found that 94% of oil and gas executives familiar with cognitive computing expect it to play a disruptive role in the industry.

Scott Kimbleton, SPE, is an associate partner in IBM’s Chemicals & Petroleum Center of Competence. He has 15 years of experience delivering analytics and cognitive solutions, with an emphasis on chemicals and petroleum. He has authored multiple white papers on data management, and his responsibilities include execution of the Watson for Natural Resources Innovation Program in the US and cognitive enablement for key global clients.

John Matson, SPE, a petroleum engineer, is an upstream oil and gas expert in the Chemicals and Petroleum Center of Competence. He has 40 years of experience in the industry. Prior to joining IBM, he worked in progressively senior operational and managerial roles with Mobil, Halliburton and Berry Petroleum. He has authored numerous publications on the digital oil field and is a former board member of Louisiana Tech University’s Engineering and Science Foundation.