3 ms·

It’s hard to adjust one mirror at a time, because 1000 other mirrors are reflecting at the same spot. And you can’t put a camera on the target to look back at t

by tlb 4y ago

It’s hard to adjust one mirror at a time, because 1000 other mirrors are reflecting at the same spot. And you can’t put a camera on the target to look back at the mirrors, because it would get melted. And some errors are harmless, like aiming too high, but aiming too low can melt the tower. And you have to do all the aiming with cheap linear actuators that aren’t very stiff, so wind affects everything. I dunno about AI, but it is a problem that programmers have struggled with so it might be worth a try.

manholio 4y ago
> And you can’t put a camera on the target to look back at the mirrors, because it would get melted. So just put it behind a heavy metalized filter that reflects over 99% of the incoming energy? And suspend the whole assembly about 1 meter in front of the target on a metalic post so it can cool down in the air. Or rather, a number of such assemblies so you can maximize incident energy over the entire target. Optimizing that does not sound like a deep learning problem, just a classic computer vision algorithm.
- tlb 4y ago
  Maybe. Just remember that an extra 1% of efficiency is worth 1% of the billion-dollar project cost, so there’s pressure not to have sensors blocking even a tiny part of the view.
- londons_explore 4y ago
  Or just put the camera somewhere else. Then point one mirror at a time at the camera (therefore, it only seex 1x sunlight) to do angle calibration, and then aim that mirror back at the proper target and calibrate the next.
zardo 4y ago
If actuators to keep the mirrors oriented correctly really do blow the budget then this sounds like a dead end.
- tlb 4y ago
  It's not black & white. You want to use the cheapest actuators you can get away with. If clever software lets you use less precise actuators, it's a big win.
buescher 4y ago
You could have position feedback on the mirrors and periodically calibrate the transfer function of mirror position to focus point at night.
- always2slow 4y ago
  I bet the mirror will heat up during the day and your night calibration will be off.
  - buescher 4y ago
    Good point.

by tlb 4y ago

manholio 4y ago

> And you can’t put a camera on the target to look back at the mirrors, because it would get melted. So just put it behind a heavy metalized filter that reflects over 99% of the incoming energy? And suspend the whole assembly about 1 meter in front of the target on a metalic post so it can cool down in the air. Or rather, a number of such assemblies so you can maximize incident energy over the entire target. Optimizing that does not sound like a deep learning problem, just a classic computer vision algorithm.

tlb 4y ago
Maybe. Just remember that an extra 1% of efficiency is worth 1% of the billion-dollar project cost, so there’s pressure not to have sensors blocking even a tiny part of the view.
londons_explore 4y ago
Or just put the camera somewhere else. Then point one mirror at a time at the camera (therefore, it only seex 1x sunlight) to do angle calibration, and then aim that mirror back at the proper target and calibrate the next.

zardo 4y ago

If actuators to keep the mirrors oriented correctly really do blow the budget then this sounds like a dead end.

tlb 4y ago
It's not black & white. You want to use the cheapest actuators you can get away with. If clever software lets you use less precise actuators, it's a big win.

buescher 4y ago

You could have position feedback on the mirrors and periodically calibrate the transfer function of mirror position to focus point at night.

always2slow 4y ago
I bet the mirror will heat up during the day and your night calibration will be off.
- buescher 4y ago
  Good point.