AI can help navigate the ‘new normal’ in oil and gas

The oil and gas industry can be transformed through automated drilling and real-time data

Unlike the oil shocks of the past, the COVID-19 crisis created an unprecedented situation where there was not only a supply shock but also a massive demand shock.

However, the decline in oil prices began a few years ago, driven by global efforts to transition to renewable and sustainable sources of energy.

COVID-19 has accelerated that trend. Certain estimates suggest that oil demand could peak as early as 2025, rather than 2040 as earlier reports suggested.

European oil majors such as BP, Shell, Total, Equinor, and Eni have been ahead of the curve when it comes to making the transition to renewables and alternative energy.

This changeover is challenging given the complexities we foresee in processes and cash flows at a time when the need to invest is more.

Therefore, innovation and digital transformation will have a big role to play here.

Reducing the cost of production, processing, and transportation by transforming their value chain and operations is the way to go.

AI will play a significant role in driving the transformation in the oil and gas industry.

Narrow view of data

Traditionally, oil and gas companies have structured themselves around achieving efficiency and cost optimization.

Therefore, they often operate in siloes with no single view of data and applications. This structure makes it harder to take on larger cross-functional Machine Learning problems.

AI adoption in the industry, so far, has been restricted mostly to localized point solutions that fail to create a broader impact.

Not surprisingly only 29 percent of energy enterprises say that their AI deployments have been working satisfactorily.

Here’s an example: Improving drilling efficacy and bringing down non-productive time (NPT) is a common use case for ML.

However, most algorithms address very specific aspects such as torque on mast motor for stuck pipes, rather than taking a comprehensive approach that uses data from across different rigs.

Cross-functional approach and data quality

While oil and gas companies have a wealth of data, a majority of it resides in the form of documents, reports, and scanned assets.

Therefore, cross-referencing, duplication, transformations, and merging of cross-domain data sets remains a huge challenge.

We now have access to unsupervised large language models like BERT and GPT3 that are equipped to make sense of unstructured organizational data through ML models.

Oil and gas majors are moving towards integrated data lakes managed by a cross-domain governance team.

New standards such as Open Subsurface Data Universe (OSDU), can help companies consolidate data operations. Companies are taking to integrated data lakes that are equipped to make cross-domain data accessible.

This data could be unstructured, transactional, real-time, IoT, or Petro-technical data.

We have seen instances where companies have made considerable progress on predicting the operating parameters for enhanced oil recovery through the integration of operational data for ML-driven analytics.

Merging Machine Learning Models with Physics and Engineering

Oil and gas operations are safety-critical; decisions arising from ML Models should be explainable and tested for consistency with the principles of Engineering and Physics.

For AI-powered operations to enable decision-making, aligning machine learning models with physics-based simulation models is key.

We’re seeing the rise of a differential programming paradigm that allows for the augmentation of data-driven AI with physics including,

This can help find a symbolic expression that matches the data trained on the neural networks. (Udrescu & Tegmark, 2020)

While the technologies are still nascent, it is evolving in this direction which can help the explainability of the ML models used in decision making.

It can also provide deep insights to help mitigate unplanned downtime, limit bottlenecks in process/workflow, and reduce safety incidents.

Looking through the crystal ball

European oil and gas companies are treading a fairly tricky path as they make the long-term transition from unsustainable fossil fuel towards an uncertain future dominated by renewables.

The COVID-19 pandemic has created its own set of challenges.

As the world moves towards a new normal of lower oil demand, we will see the emergence of unlikely alliances and a greater push towards driving process efficiencies.

As cross-functional silos melt away, there will be greater emphasis on integrated data and governance.

For instance, drilling efficacy will improve substantially through the use of integrated reservoir models as well as geological and geophysical interpretation models.

These models will be governed by real time data from wells on drilling and production, as well as past drilling reports.

We can also expect the emergence of new engineering workflows that are governed by reservoir and geotechnical interpretation.

We are likely to see the emergence of goal-driven automation in drilling, where the system dynamically calibrates and plans logistics including personnel, equipment, and consumables.

The role of AI in oil and gas will certainly be front and center in the future.

However, the industry will continue to demand more scale and functionality from AI adoption.

The ability to integrate physics and engineering technologies with data-driven analytics, machine learning, and automation will prove to be a huge success factor for oil and gas companies.

Balakrishna D R is SVP and the service offering head of ECS, AI, and automation at Infosys. His work covers the energy, telecoms, and media industries, among others, and is also responsible for driving both internal automation for Infosys and for providing independent automation services for clients.

References

Raissi, M., Perdikaris, P., & Karniadakis, G. (2019, February 1). Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics, pp. 686-707.

Udrescu, S.-M., & Tegmark, M. (2020, April 15). AI Feynman: A physics-inspired method for symbolic regression. Science Advances, pp. Vol. 6, no. 16.