AI/ML in Oil and Gas — Five macro trends to expect (2020 edition)

I wrote this story on LinkedIn in 2017 where the depressed oil prices were a new experience to the oil industry. In 2020, the demand for oil & gas has been slashed due to COVID-19. Renewables pose a stiff challenge to the once dominant fossil-fuel industry. Nevertheless, oil & gas is expected at least over the next decade to provide energy to a major part of the world. It will not however, be — “business as usual”. Each aspect of oil production, exploration, and marketing will need to streamline and depend on less manpower. These are the 5 major areas where I see movement in.

Surface robotics and autonomous systems (AI)

This is a much more mature field now. UAVs — Unmanned aerial vehicles (drones to rest of us) routinely inspect remote oil fields. Powered with AI, these drones can now take readings from analog gauges and convert them to digital signals and transmit them to cloud databases.

Equipment health monitoring for rotating machinery like turbines are a commodity technology now, provided by several companies like C3.ai, SparkCognition etc. IoTs are extensively used to send equipment malfunction, production anomalies to the cloud so that a central team can look at it. What required a dedicated field staff to perform just 5 years ago can be performed with a fraction of the staff!

There are still a few challenges ahead

1. Detecting spills — This is unlike a typical image detection application where we have access to vast corpus of data e.g. ImageNet. It is still very hard to differentiate between an oil spill, a chemical spill, and water spills (rain sometimes creates puddles that get mistaken for spills)
2. Detecting hydrocarbon leaks — Through infrared imagery, we can pin-point locations of hot-spots and potential leaks. It is really hard however, to differentiate between hydrocarbon leaks and exhaust from machinery e.g. generators

Production optimization (ML)

This again remains as relevant as it did in 2017. The number of algorithms that address this application are limited as we operate in spatially sparse but data rich environment.

For unconventional oil & gas applications, we’ve made huge strides in optimizing wells on artificial lift — think the pump jacks you see while driving through West Texas, wells on electrical pumps, plunger lift etc. Gas-lift is still an application that could use algorithmic advancements. Multiple players like Kelvin, Abyint operate in this space.

What is missing?

1. Unconventional wells are still not as profitable as the industry thought them to be. A big reason is the rapid decline of oil production — A well could lose 6% production each month, reaching a steady 10–20% of original oil production within a year. If we can reduce the decline even by a few percentage points, the impact on profitability is huge! This requires algorithms to tie routine field decisions to underling profit-loss calculations. This does not exist since the data is in non-standard format e.g. PDF documents, non-standard databases
2. The remote offshore poses a larger challenge. Technologies like digital-twin are good to hear about but do not provide any tangible benefit. To make things worse, remote oil platforms often have a few wells, reducing the amount of data necessary to drive optimization without physics-driven models

Knowledge extraction (AI)

Since 2017, large oil producers have delegated significant resources to digitizing old reports, leading to interesting applications

1. Risk analysis of new projects based on past experience — Often developing a new oil field or drilling an exploration well is a risky undertaking. Companies spend vast amounts of time and resources to de-risk these applications. Having ready access to a vast corpus of old project documents, emails provide a never-before way to reason through scenarios.
2. Feeding simulation models automatically based on old reports — It takes a lot of time to feed large reservoir simulation models. It takes an even larger time to change the underlying concepts. These are now possible through templatizing geological features.

What’s missing (but being worked on)?

Companies like Nesh are working on novel ways to keep oil & gas knowledge from depleting due to retirements or layoffs. These use answering systems, recommender systems together with Alexa-like speech driven devices.

I’m sure I’ve missed a few more applications that I plan to get to in future articles.