Reply to thread

Message

Sundog, I'm effectively retired and haven't worked on FCS development since the early 2000s, so I'm not completely up to date on stuff that doesn't make it into Flight or AvLeak, but within those limits it's plain there are a lot more FCS development projects due to the proliferation of UAVs, and also that that same proliferation allows much more experimental work. When you can throw an airframe together in a few months, and fly it without risking a test pilot, it becomes much easier to justify, and execute, experimental projects with radical configurations of both airframe and software. I'd be surprised if there aren't quite a few projects looking at neural-nets, but what would really interest me is whether anyone has managed to certify one, or even worked out how to approach that problem. A neural-net is a moving target WRT traditional testing, and if you can't demonstrate complete testing, how do you certify it?

<blockquote data-quote="DWG" data-source="post: 314752" data-attributes="member: 414"><p>Sundog, I'm effectively retired and haven't worked on FCS development since the early 2000s, so I'm not completely up to date on stuff that doesn't make it into Flight or AvLeak, but within those limits it's plain there are a lot more FCS development projects due to the proliferation of UAVs, and also that that same proliferation allows much more experimental work. When you can throw an airframe together in a few months, and fly it without risking a test pilot,&nbsp; it becomes much easier to justify, and execute, experimental projects with radical configurations of both airframe and software.&nbsp; I'd be surprised if there aren't quite a few projects looking at neural-nets, but what would really interest me is whether anyone has managed to certify one, or even worked out how to approach that problem. A neural-net is a moving target WRT traditional testing, and if you can't demonstrate complete testing, how do you certify it?</p></blockquote><p></p>

[QUOTE="DWG, post: 314752, member: 414"] Sundog, I'm effectively retired and haven't worked on FCS development since the early 2000s, so I'm not completely up to date on stuff that doesn't make it into Flight or AvLeak, but within those limits it's plain there are a lot more FCS development projects due to the proliferation of UAVs, and also that that same proliferation allows much more experimental work. When you can throw an airframe together in a few months, and fly it without risking a test pilot, it becomes much easier to justify, and execute, experimental projects with radical configurations of both airframe and software. I'd be surprised if there aren't quite a few projects looking at neural-nets, but what would really interest me is whether anyone has managed to certify one, or even worked out how to approach that problem. A neural-net is a moving target WRT traditional testing, and if you can't demonstrate complete testing, how do you certify it? [/QUOTE]

Verification