China puts trust in AI to maintain largest high-speed rail network on Earth (www.scmp.com)
from ViXY_DBC@lemmy.world to world@lemmy.world on 12 Mar 2024 10:47
https://lemmy.world/post/13026807

#world

threaded - newest

lawrence@lemmy.world on 12 Mar 2024 11:10 next collapse

Interesting. They are using AI to identify problems before they happen.

owen@lemmy.ca on 12 Mar 2024 11:28 next collapse

Pretty reasonable. The title makes it seem like they have a hairbrained scheme…

Hamartiogonic@sopuli.xyz on 14 Mar 2024 08:13 collapse

Many problems have early warning signs. Just train the model to notice those signs.

At least in many industrial applications, the vibration of an electric motor or an axle is a good measurement. Also, the temperature of a ball bearing can tell you a lot. That’s just the basics though, because you can also train the model to look at fancier details.

cynar@lemmy.world on 12 Mar 2024 12:08 next collapse

This is a good example of how AI can be used well.

Current AIs are effectively fuzzy pattern detection and matching engines. This one can sift all the data coming in, and spot patterns that previously corresponded to problems. It then flags them for human interpretation.

The AI chunks the vast sea of data. A human is then involved to sanity check what it has found, and react accordingly. E.g. a pattern appears that often precedes a broken rail within a month. A human can check the subset of the data, and schedule a maintenance team a week later. Conversely, a pattern that leads by hours would require an immediate response.

voracitude@lemmy.world on 12 Mar 2024 13:03 next collapse

Exactly this. The number of fields machine learning breakthroughs can be applied in, just for processing huge collections of unstructured data, is truly mindboggling. We’re in for an interesting ride!

arin@lemmy.world on 12 Mar 2024 13:15 next collapse

But how did China do this when we banned the sale of nvidia gpus to China?

cynar@lemmy.world on 12 Mar 2024 16:12 collapse

Nvidia GPUs aren’t the only way to run a machine learning type system. They are just the easiest to use, currently. China has also been developing their own AI optimised chips. Though I don’t know much/anything about them.

Halosheep@lemm.ee on 12 Mar 2024 21:19 collapse

Pretty sure the other commentor was being sarcastic.

arin@lemmy.world on 12 Mar 2024 22:07 collapse

I still want to know

VirtualOdour@sh.itjust.works on 15 Mar 2024 14:46 collapse

Well it’s like the other person said Nvidia aren’t really doing anything magic, Cuda is great but it’s not the only option.

Kbin_space_program@kbin.social on 12 Mar 2024 13:25 collapse

Fun fact: Salesforce has been selling this exact service for ~8 years now. Its old tech.

It's their Wave Analytics package.

kugel7c@feddit.de on 12 Mar 2024 16:49 collapse

Also some application of similar tech has worked itself into industrial machines and factories over the last 10 years or so, it’s downright ubiquitous for anything that’s expensive and requires maintenance/ upkeep. Also it’s well intertwined with the ML tech we see consumer facing nowadays, the image recognition of 4+ years ago was made to recognize issues with materials, unexpected growing patterns, anomalies, as well as recognition and counting etc… before we got just point your camera and it’ll tell you what you’re looking at.

tsonfeir@lemm.ee on 12 Mar 2024 12:59 next collapse

Ahh yes, c(ai)de.

RagingHungryPanda@lemm.ee on 12 Mar 2024 14:20 next collapse

Really interesting overall, but the article is more like a tweet. I’d have liked to have read HOW it’s being used and gotten some more detail on it.

00x0xx@lemmy.world on 12 Mar 2024 19:22 collapse

This is a perfect use for AI to automate systems. The job of maintaining a train isn’t dynamic, it’s very predictable, AI shouldn’t haven’t a problem doing this task well.

isles@lemmy.world on 13 Mar 2024 20:11 collapse

AI is not needed for predictable tasks, it excels in finding predictions in a mass of data.

For example, one of the most dynamic systems we have is global weather and GraphCast is incredible at it.