Power plants are both crucial and mercurial; the whims of wind speeds, light availability, ambient temperatures and equipment failures can, under the wrong circumstances, send grid operators scrambling to shore up losses in power production. But as simulations and predictive technologies like AI become more robust, our understanding of how, when and why power plants experience downtime is rapidly increasing – and, as a result, so is our ability to minimize those downtimes. Today at SC21, Nvidia has announced that it is working with Siemens Energy to develop digital twins aimed at predictive maintenance of power plants.
Siemens Energy (formerly Siemens Gas and Power) emerged last year amid a broad structural reorg of the German multinational’s energy strategy. Natural gas turbines remain a major portion of the company’s portfolio, though – and specifically, combined-cycle natural gas plants, which run a natural gas turbine, siphon off the waste heat that would normally be lost to the atmosphere and use that waste heat to power a secondary steam turbine.
The energy production from this secondary turbine, and the resulting efficiency gains for the plant generally, can be substantial – Siemens Energy estimates that using combined-cycle technology improves their plants’ efficiency by over 60 percent. But there’s a downside: the heat recovery steam generators (HRSGs) that enable the combined cycle plants can also introduce corrosion, which in turn means maintenance, downtime, lost revenue and lost efficiency. Currently, the HRSGs have an average planned downtime of 5-6 days per year, during which time the pipes are regularly checked for corrosion – and Siemens Energy estimates that reducing that downtime by just 10 percent would annually save the industry $1.7 billion.
Enter Nvidia. The company has been using digital twins – meticulous recreations of real-world objects and processes, powered by Nvidia’s real-time simulation platform, Omniverse – to help organizations and companies develop and operate their tools. Ericsson, for instance, uses Omniverse-powered digital twins of cities to study how best to propagate 5G signals, while BMW is using digital twins to mirror its “factory of the future.”
Siemens started with Nvidia’s Modulus framework for physics-based machine learning models, using the framework to model the dynamics of steam, water, temperature, pressure, pH and more in the pipes of Siemens Energy’s HRSGs in real time. (The Modulus portion of the work was, itself, run on AWS P4d instances powered by Nvidia’s A100 GPUs.) Then, Omniverse stepped in for the visualization and broader simulation of those dynamics.
The companies hope that this real-time, physics-based simulation of water and steam in HRSG digital twins will help Siemens Energy predict corrosion and mitigate downtime in the real world by reducing the frequency of necessary corrosion checks. Previously, simulating the physics involved in these plants had been onerous at best, prohibitive at worst: eight weeks per HRSG for a portfolio of over 600 units. Reduced-order models, meanwhile, had exhibited low accuracy. Now, Nvidia says, accurate digital twin simulation can be conducted in just hours.
“Nvidia’s work as the pioneer in accelerated computing, AI software platforms and simulation offer the scale and flexibility needed for industrial digital twins at Siemens Energy,” said Stefan Lichtenberger, technical portfolio manager at Siemens Energy, in an interview with Nvidia’s Richard Kerris. The digital twins are part of an ongoing partnership between Nvidia and Siemens, with Nvidia also recently providing its Triton Inference Server for Siemens Energy’s automation and power plant inspection efforts.
Nvidia’s digital twins tech is certainly having a low-carbon moment, between the Siemens Energy partnership and the announcement last Tuesday of its (somewhat more ambitious) in-development digital twin of Earth for climate change prediction. The digital twin will be run on a new, Nvidia-built supercomputer called Earth-2 (or E-2 for short). “All the technologies we have invented up to this moment are needed to make Earth-2 possible,” said Nvidia CEO Jensen Huang during his keynote. “I cannot imagine a greater and more important use.” Not much else has been revealed about E-2, but we can expect more information on it at the spring GTC event in March.
Header image: high-fidelity water and steam flow inferred by Nvidia’s digital twin.