Quote:
Originally Posted by jyl
But I think there will be more and more, and ultimately quite a lot, of jobs where AI can do part of the job with humans using the AI output, selecting from choices prepared by the AI, or supervising the AI. Fewer humans.
|
This is exactly what we are doing right now. We are currently using AI (machine learning, more accurately) to optimize the control system for our supercritical (cryogenic) hydrogen system. This system is really one-of-a-kind in the world and the controls architecture needed to maintain temperatures and pressures during thermal transients is very difficult to manage (or troubleshoot when things go wrong). When "things go wrong" we experience a pressure excursion that blows rupture disks and vents hydrogen to the atmosphere. We embarked on an AI-ML project to study the system and it ended up spitting out some PID control parameters that were wildly different than what we were previously using. We loaded those parameters and tested the system under controlled conditions and it performed wonderfully.
We would likely never have gotten there without the ML effort.