Proper machine learning usage maximizes upstream workforce resources for optimal impact

Last year was characterized by a swift change in priorities regarding the political energy landscape. The term, “energy trilemma,” is one the industry has become overly familiar with, as it looks to balance the growing global population’s need for affordable, low-carbon and dependable energy supplies. These crucial drivers require abundant energy from a number of sources, at a time when the energy industry is still reeling from drastic headcount reductions as a result of the Covid-19 pandemic and a rapidly aging workforce. The mantra “doing more with less” is more true today in our industry than ever. 

Workflow trends. The upstream sector has certainly not been immune to this trend. The state of Texas, alone has 78,000 fewer upstream oil and gas employees than in February 2020, underscoring the challenge facing the industry. The upstream plays a vital role in addressing our global energy trilemma—whether maximizing fossil fuel production, accelerating geothermal asset development or identifying and monitoring subsurface carbon capture sinks long-term. Many of the seasoned workflows used today in oil and gas are equally relevant to low-carbon initiatives. However, transitioning these valuable workflows to be applied effectively to Carbon Capture Storage (CCS) and renewable energy efforts is challenged by the significant loss of experienced personnel. 

When striving to meet emissions targets, a healthy blend of new renewable energy sources and carbon management projects are necessary to ensure environmental goals aren't delivered at the exclusion of societal and economic requirements. However, as with many nascent and developing industries, personnel and investment often lag behind established industries. Efficient working practices are absolutely vital for growth in these sectors, as the hiring and training costs that companies would incur to onboard thousands of new workers would be ultimately prohibitive. 

Greater interest in ML. These persistent challenges across the energy industry amplify interest in Machine Learning (ML) and stoke the hope that the long-promised value will be realized in revolutionizing our industry, Fig. 1. ML’s automation promise, and resulting liberated resources, increased productivity and reduction of lost time drove the oil and gas Artificial Intelligence (AI) market to over $2 billion in value during 2019.  

The most beneficial application to ML to upstream operations isn't R&D- it's utilizing ML as a common tool that asset teams can depend on in their daily workflows, such as RockDoc RPML

Today, ML is no longer a new, unproven technology. There are ML success stories, as well as many cautionary tales stemming from failed projects and lower-than-expected returns on investment. It's often stated that 90% of digital transformation initiatives fail, and the implementation of ML and AI certainly falls into that figure. The subsurface has proven particularly difficult, when it comes to proving the value of these initiatives. The primary use of ML in the subsurface space has been to improve final deliverables, creating an incremental increase in the accuracy of models used in the decision-making process.  

However, these projects typically have limited scope to increase business practices across a company. Asset-specific models lack transferability across business units, while important prep work, such as data location, cleansing and validation often increases the amount of time taken to deliver a project. This doesn’t mean such approaches lack merit, but to truly move the needle and deliver the cost-savings and efficiency increases that ML advocates believe are possible, we need to alter our approach.  

Automating entire workflows—and even occupations—has been a lofty goal for ML and AI, and our current reality indicates these possibilities remain distant. A McKinsey paper estimates that only 5% of occupations may be entirely automated, but that 60% of occupations could potentially automate 30% of their tasks. It’s the automation of time-intensive and vital, yet highly repeatable, portions of workflows that energy companies hope ML can address.  

In the subsurface, trying to create an entirely automated model is an incredibly complex and involved task with huge amounts of trial and error as part of the iterative refinement cycle. However, solving some of the pre-model data cleansing, manipulation and model setup can be outsourced to be performed by guided, automated workflows, Fig. 1. Automating this preparatory work can free up subsurface specialists to create more value for their organizations through contributions to the decision-making process. 

Subsurface ML application. At Ikon Science, our Rock Physics Machine Learning (RPML) application was built specifically to advance this approach. RPML’s ability to optimize model and parameter selection, based on data inputs, allows geoscientists to instantly begin applying their subsurface specialist knowledge to reduce setup time and enable a single resource to work across multiple assets. The removal of subjective processing also reduces human input bias, leading to new levels of consistency across all business units. These advantages have important implications, including quickening the delivery of well service projects and relieving constraints on key personnel. Such results enable the energy industry to meet the world’s energy demands with a constrained workforce. 

Fig. 2. Using ML to automate lower-level workflows empowers workers to contribute their deep knowledge to critical decisions for better results in the field.

Focusing on repeatable tasks that are not specialized also allows geoscientists to productize ML, moving the discipline from the fringes of R&D into a common tool that asset teams depend upon in their daily workflows. 

Applied correctly upstream, ML has great potential to revolutionize our industry. Leveraging ML as a tool to automate lower-level workflows and empower—rather than replace—end-users can serve as a catalyst to foster faster decision-making and reduce bottlenecks. Automating repeatable tasks while optimizing models frees specialists to apply their crucial knowledge to companies’ decision-making protocols, yielding improved results in the office and oil field. 

About the Authors

Chris Hanton

Ikon Science

Chris Hanton is the director of Digital Transformation Solutions at Ikon Science, overseeing development and implementation of the next generation of software solutions. He has over 15 years of industry experience working with global operators, delivering subsurface solutions and technology to promote efficiency and productivity. Mr. Hanton has a MeSci degree from the University of Liverpool in Geology and an MSci degree from the University of Aberdeen in Integrated Petroleum Geoscience.