Can someone please explain where the innovation is here? I’m not very familiar with solar generation.
Given that we know the suns position and that of the mirrors, how is computer vision able to better aim the mirrors?
What am I missing?
Given that we know the suns position and that of the mirrors, how is computer vision able to better aim the mirrors?
What am I missing?
The article mentions seamlessly switching to solar when the sun comes out, and hints at having more precise temperature control. I wouldn’t know whether either makes much of a difference.
https://www.vox.com/science-and-health/2019/11/19/20970252/c...
Keeping the cameras from being blinded by direct hits during startup and adjustment might be a problem. They probably have to start with all mirrors off-target and bring them onto target one at a time.
[1] Probably normally natural gas fired burners/kilns
Sounds like a huge step forward to me. And Bill Gates is usually on track with his investments in transformative tech..
Heliogen claims to reach the desired temperatures purely through solar.
I'm trying think of anything weird that ML may be able to pick up like correcting for local atmospheric conditions, more even heating over the surface of the collector target, adaptive calibration for better aiming or ability to use cheaper components, etc.
https://en.wikipedia.org/wiki/ESolar
So... I’d guess that part of the eSolar approach is carrying over into Heliogen. The control system for using mirrors seems potentially more complex. I assume you want to direct light from an array of mirrors such that it optimally a single cell. Or so that it covers a number of cells, or such that it optimally avoids defective or low performing cells.
So with, some skepticism, I can kind of see an AI play here.
> The breakthrough in Heliogen’s technology starts with our patented closed-loop control system that makes our field of mirrors act as a multi-acre magnifying glass to concentrate sunlight.The HelioMax system is an industry first and a critical step in harnessing the power of the sun. Our ability to concentrate and capture sunlight allows us to create carbon-free, ultra-high temperature heat (HelioHeat) commercially for the first time....
They define the HelioMax system as
> HelioMax: An array of computer-controlled mirrors (heliostats) collects and concentrates sunlight.
So, obviously they are trying to not say anything useful about what their special sauce is. But they do seem to be claiming that it has something to do with how the mirrors are controlled. As you say, any modern system will use computer guidance, and the sun's trajectory is very simple and predictable, so it's hard to imagine what exactly they are doing special.
Maybe some sort of active correction for atmospheric distortion? I dunno.
This fits in with something I saw in another article, which said that Heliogen's system doesn't need months of calibration when installed before it can produce maximal output.
This suggests that they have some way of measuring the contribution of each individual mirror to the total output, and tweaking its position to maximize that, so that all they have to do is get each mirror near the right position and then they can quickly tune it in dynamically.
This reminds me of a film that the professor in APh 23, "Demonstration Lectures in Optics", showed us at Caltech in the early '80s. The professor [1] was also a researcher at Hughes, and the film showed a demo of a system they were working on there.
It consisted of a bunch of radiators. By adjusting the phase of the radiators to change the interference pattern they could get it so the radiation pattern had a strong lobe in one direction, which they could steer. In theory, all you had to do was crank up the power, and you've got yourself an energy beam weapon that will zap whatever you aim the lobe at.
But how to aim that lobe? That was the cool part. They would modulate the phase of each radiator, with each one having its own unique modulation frequency. Suppose now you've got a target flying around somewhere in front of the thing. It's getting hit by all the radiators, but isn't in the lobe. You have a sensor that can see the energy reflected off the target.
You analyze that reflected signal and look at the frequency components of its intensity variation. If a given radiator is at a phase that is trying to put the target in the lobe, you won't see much variation at that radiators phase modulation frequency. If a given radiator is at a phase that is trying to put the target in maximum destructive interference instead of maximum constructive interference, you will see its frequency in the reflection variation. You can use this to derive a feedback signal for each radiator to adjust its phase to try to make constructive interference at the target.
In the film it showed this system aiming against a dark curtain with no target. You could see a bunch of blobs of light just kind of drifting around aimlessly. Then they dangled a little aluminum model of the starship Enterprise in there, and pretty much instantly all the blobs of light from all the radiators converged on it.
Note that since the feedback is based on the contribution of each radiator at the target, it should be able to automatically compensate for atmospheric distortion.
I wonder if something kind of like this could be what Heliogen is doing? It wouldn't be able to be quite the same, because there wouldn't really be any noticeable constructive or destructive interference going on between the reflections from the different mirrors. But the general idea of modulating each radiator (mirror in this case) and then detecting that modulation at the target to provide feedback for controlling the radiator might work.
If you either had something that could detect small changes in heat over very short times, or small changes in light over very short times, you could put that at the target (maybe more than one, both in the target area and just outside it). Then modulate the mirrors, and look for corresponding variation at the target to figure out if the mirror is positioned right or not.
You probably would not need to do all the mirrors at once, which simplifies things. You could just scan through them, for each modulating it, determining the correction, applying it, and moving on to the next mirror.
If you have good enough computer vision, maybe that part about detecting small changes in light over very short times could be done by a camera looking at the tower.