In Stay Sharp Episode 154, Operational Digital Twins: Closing the Speed Gap, co-hosts Juliann Grant and Jonathan Scott welcome Prithvi Singh Chauhan from Geminus AI for another installment of Stay Sharp’s series on Digital Twins. In his career, Prithvi has supported large scale energy systems as both a working petroleum engineer and a machine learning engineer, which makes him the ideal guest to delve into how AI and machine learning play a role in operational digital twins (ODTs).
Developing AI Concepts
Prithvi starts by defining three AI terms, though he cautions that the concepts have been building for a while, and it’s just the buzzy names that are new. Scientific AI is a model with physics built in, meaning it’s first trained on physics equations and then on actual field data.
Prithvi Singh Chauhan, Machine Learning Engineer, Geminus AI
Multimodal AI can consume multiple data sets in any format, from simulator output and chemical lab results to engineering drawings and more. And world AI—probably the most intriguing if not the most trustworthy—is an AI system that builds its own representation of the world by observing the data and environment around it. Prithvi says, “Everyone in this space is chasing the same destination, which is a model that can understand the whole environment well enough to reason about it.”
An ODT and AI
Before AI, Prithvi explains, the back office of an unglamorous, traditional energy company would use industry-standard, steady-state simulators and models to design and engineer oil flow and hand that off to the operations team. Every few months, someone would tune the model against field data so it roughly matched the asset. “It behaves like a very knowledgeable but very slow colleague,” he says. But while the simulators have excellent physics built in, their performance falls short because they were never designed to work at the rate that matches the asset’s operational speed.
That’s the challenge AI can solve: making the operational digital twin fast enough and connected enough to the assets running in the field to keep information fresh instead of stale, connected instead of siloed, and most of all, flowing at the pace of operations. That’s what is bridging the gap between operations, simulations, and models. The digital twin infrastructure has existed, but the computational power, graphics processing unit advances, and the AI revolution are combining to dramatically accelerate the data generation and reconciliation that’s possible. “You can reconcile them against the live data continuously now instead of quarterly calibration,” Prithvi says. “Then each domain can have a fast model, and you finally connect them to these high-frequency data sets and optimize it. That’s where the value has been hiding.”
AI + Operational Digital Twins + Connected Data = Engineers spending less time wrestling with models and more time deciding what good looks like.
The Future is Science Fiction
Prithvi admits that as fast as technology is moving in the energy industry, the industry’s willingness to use it isn’t keeping pace. “That gap tells you where the frontier actually is, the gap between industry’s adaptiveness versus the technology,” he says. What have been adopted industry-wide today are fast, physics-grounded models of a single system. In the last couple years, those are starting to be connected instead of being isolated problems—and the connectivity means they can optimize your assets or process. The next steps Prithvi sees the industry taking is adopting models that maintain themselves and predict their own uncertainty percentage. He explains, “The leading edge is where the model is watching its own predictions against the live data, catching the drift between what is happening in live data versus itself, and recalibrating it without anyone commissioning that work.” Plus then it’ll tell you how confident they are in their own work, all of which can dramatically change how engineers use AI.
Where Prithvi thinks the industry is headed is toward a model that’s pre-trained across hundreds of assets that you can fine-tune to your assets in hours. This means, “you stand up a twin for equipment nobody has ever modeled because it’s seen a thousand cousins of it around the world.” To him, the frontier is the industry trusting an AI model that is grounded in and understands physics. He also suggests reframing how we all think about AI’s evolution. “Stop asking whether AI can replace a simulator,” he says, “and rather ask what a simulator could do if it could learn like AI.”
Learn More About AI for ODTs
The full podcast contains more discussion and details on a variety of topics, including why Prithvi says the two halves of his career were the same problem from different directions, why he describes an AI-enabled digital twin as a simulator with an engineer in front of it, examples from the oil and gas industry that illustrate how vital this AI evolution is, and more.
Be sure to check out Stay Sharp Episode 154: Operational Digital Twins: Closing the Speed Gap, and join us each week for a new podcast.
Follow the Razorleaf Podcast, Stay Sharp in Digital Engineering on:
