Redefining the oil and gas industry through machine learning

The challenge within the oil and gas industry is that of complexity and therefore cost – the work is often in tough environments (underground or even offshore) and delays/ downtime can be very costly. Some industry analysts size the cost saving opportunity from digitization at $73B within just the upstream segment¹. Data collection is thus a priority in the industry and there has been a proliferation of sensors in equipment to collect more data than ever. However, only a small percentage of the data collected are analyzed today². This is because interpretation is difficult, time-consuming and sometimes repetitive. Schlumberger, as the largest oilfield services company, is trying to solve this issue in the short term by using machine learning to augment human decision making and create better processes.

An example of machine learning being used to improve processes in the short term is in drilling. Schlumberger is using machine learning techniques to predict drilling events such as severe vibration or fluid loss based on data inputs ranging from surface data to the wellbore geometry³. As a consequence, drillers are aware of these events before they occur and able to adjust operations accordingly real-time. The AI-augmented process in drilling is thus able to save costs by reducing drilling times by preventing disastrous events, but also reduce machine downtime through implementing predictive maintenance based on actual wear-and-tear instead of pure scheduling⁴.

Another area where Schlumberger is using machine learning for near-term process improvement is in well completion – going from a drilled well into a producing one. Traditionally, less than 5% of fractured wells in North America are designed through advanced simulation due to the skill and processing time required⁵. Schlumberger is exploring machine learning methods that are able to create models within a matter of hours that can predict the productivity based on a change in the completion design. Engineers are now able to quickly evaluate across a wider set of options which fundamentally changes the process to well completion design optimization.

In the longer term future, Schlumberger’s aim is to integrate processes across both well planning and execution with back end systems such as inventory management and logistics⁶. Machine learning enables this integration through driving automation and data-driven decision making. The role of the humans will be less on executing processes but evaluating further improvements and interpreting data. Machine learning thus enables process improvements to shift from a mindset of reducing processing time and uncertainty but towards generating new ways of working.

Many of the machine learning methods outlined above have dealt with structured data. Further areas of exploration for Schlumberger can be using machine learning to use of unstructured data through machine vision and natural language processing. An application in machine vision for process improvement is facilities inspection. The current processes require workers to visually identify any errors which can have subjectivity. Combining machine learning and computer vision in this area will allow Schlumberger to improve on the existing facilities inspection process by increasing safety and reducing the cost of having people on-site. Natural language processing advances can allow Schlumberger to use machine learning in improving the process how workers can query and access the existing database of knowledge. Instead of going through manuals or relying solely on personal experiences, workers can ask the system for the best answer according to the accumulated experience of the company and the current conditions. The reduction in complexity for the workers will improve productivity by allowing focus on value-add activities and allows for better decision-making regarding potentially high capex decisions.

Pathways to Just Digital Future

Watch this tech inequality series featuring scholars, practitioners, & activists

In conclusion, machine learning is beneficial for Schlumberger to tackle complexity, especially in the abundance of data, within many of its processes. Machine learning thus does not only deliver step change improvements in existing processes but can create new ways of working by reducing both the time and cost to do certain simulations or processes especially for the upstream business. Some open questions are in the area of competition and collaboration. While Schlumberger has historically been one of the leaders of technological innovation in the industry, aside from proprietary data, how will they differentiate themselves from other service firms or startups tackling many of the similar issues? And within the data context, how likely is it that existing customers will share their data openly, especially now that data is a competitive advantage in the age of machine learning?

Word count: 724 words

Footnote

1. Shadia Nasralla, “Digitalization can save oil upstream business $73 billion a year: Woodmac”, Reuters, November 12, 2018, [https://www.reuters.com/article/us-oil-digital-savings/digitalization-can-save-oil-upstream-business-73-billion-a-year-woodmac-idUSKCN1NH0QR ], accessed November 2018
2. Ed Crooks, “Drillers turn to big data in the hunt for more, cheaper oil”, Financial Times, February 12, 2018, [https://www.ft.com/content/19234982-0cbb-11e8-8eb7-42f857ea9f09], accessed November 2018
3. Jie Zhao, Yuelin Shen, Wei Chen, Zhengxin Zhang, Sonny Johnston, “Machine Learning-Based Trigger Detection of Drilling Events Based on Drilling Data”, Society of Petroleum Engineers, October 4, 2017, [https://www.onepetro.org/conference-paper/SPE-187512-MS], accessed November 2018
4. Schlumberger, “Operational Workflows Enable End-to-End Digital Solutions”, E&P Magazine, April 30, 2018, [https://www.epmag.com/otc-extra-field-development-operational-workflows-enable-end-end-digital-solutions-1697821], accessed November 2018
5. Piyuh Pankaj, Steve Geetan, Richard MacDonald, Priyavrat Shukla, Samir Menasria, Han Xue, Tobias Judd, “Application of data science and machine learning for well completion optimization”, Offshore Technology Conference, April 30, 2018, [https://www.onepetro.org/conference-paper/OTC-28632-MS], accessed November 2018
6. Kelli Ainsworth Robinson, “Next-level automation to go beyond discrete processes, linking data and workflows with digital technologies”, Drilling Contractor, Feb 2, 2018, [http://www.drillingcontractor.org/next-level-automation-to-go-beyond-discrete-processes-linking-data-and-workflows-with-digital-technologies-45375], accessed November 2018