Not dismissing this press release, but at least some scale of progress should be used to compare against. Or state where the economical/industrial/regulatory innovation stands.
TL;DR Traditional CSP is aligned once, Heliogen aligns continuously. This allows them to correct for individual panel changes in warping, settling, weather, dust, etc. This approach also allows them to use cheaper hardware.
> Rather than assembling large, complicated, curved heliostats (mirrors) on site, eSolar used small, flat, prefabricated heliostats of only about a square meter. They were cheaper, faster and easier to set up, more modular, and easier to replace.
> Gross’s key insight was that he could replace a lot of the material and labor involved in CSP with computing power. (Or, he could replace stuff with intelligence.) Rather than make bigger, more complicated mirrors, he made small, simple ones and controlled them with software, so they stayed aligned more precisely and produced more power.
I did, their entire premise is invalid. They are either very bad at math, or engineering, or both.
The metal part, screw drive, and simple pulleys and etc are nor hard, nor expensive. Saying it otherwise is purest bs.
The limits of precision of theodolite are more than enough position for precise initial reference.
Existing solar thermal farms were not designed for generation of process heat in mind to begin with, so them saying that they suck at that because of their "design" is a PR trick.
Existing thermochemical processes are profitable enough to burn oil for them, and if one minds anything cheaper, they will first consider CHP colocation before thinking of anything as weird as this.
Even if the surface temperature of the solar collector focus point is 1500C, the molten salt flowing on the other side of that wall must be at a significantly lower temperature in order to transfer heat quickly enough from the small area. Then you will have a thermal profile in the wall, say from 1500 to 1200 C, and furthermore a thermal boundary layer in the fluid, say from 1200 to 800 C.
Molten salt towers don't heat above 800 iirc. If they were designed for storing more energy it would come from more mass. It goes without saying that the benefit of salt is reduced once it becomes gasified.
In a technical sense, that would depend in part on the salt. If a salt with a higher melting point existed then it could be stored at a higher temperature.
I’m not meaning to “correct” your statement, because the actual practicalities of these systems Is not my expertise and there may be other much more significant factors to consider
It's a solar furnace / oven not a solar power tower.
Odeillo's purpose is to heat things very high and very quickly, not to produce power. It's a tool for materials research not a solar power demonstrator.
Granted. Size and lightweightness are definitely a point - that could be emphasized as a major argument - even the ratio with size & power capacity & heat intensity.
Bill Gross, Heliogen’s current CEO previously founded eSolar. eSolar proposed using an array of Heliostaticly controlled mirrors to direct light toward a solar cell:
https://en.wikipedia.org/wiki/ESolar
So... that part of the eSolar approach is carrying over into Heliogen.
Last time this was posted the “AI” component of the play was less clear... But with an array of mirrors, in particular being to generate heat in some volume I guess a more complex feedback/control system may be required.
What exactly do you need AI for except for buzzwording? It is not as if the sun/earth trajectory or rotation speed is random. You can precalc with ease at every point which is the best position for each mirror.
The difficult problem is not knowing where the sun is but knowing where the ground is. Since all the panels are separate structures not on a single rigid slab, ground "settling" can tilt them fractionally in unpredictable ways.
It sounds like what they have is machine vision as a novel input into a conventional control loop. For one panel at a time, wiggle it about and look how the light spot changes. Find the local maximum for tightest spot. That's now "locked" in the PLL sense and should stay locked for days or weeks.
Just align one mirror at a time. The one being aligned vibrates at some frequency, and detectors around the target just measure that modulated signal. So just keep cycling through all the mirrors.
From my comfy armchair, with zero experience in this field, it seems like you could use star tracking to very accurately predict the angle of the sun (since the sun is effectively fixed in position relative to the background stars over the timescales you'd care about). Spotting the target is a little harder, but the same measurement could be made by sticking some reflectors in known positions around the target.
I'm sure there are tons of considerations that aren't occurring to me (for instance, thermal expansion likely contributes substantially to making it a moving target (literally)), but I really don't see where "AI" comes into it...
Figuring out the sun's location is the easiest part of the problem. The hard part is that each mirror is a big, floppy slab that moves with wind, temperature, gear backlash, mechanical wear, rain/hail, and ground settling.
The so-called "AI" here is really just a clever way to use closed-loop control so that the above factors don't matter much.
(Side note: A large amount of "AI" these days just amounts to changing open-loop mechanisms to closed-loop, often by using computer vision as the state sensor. 3D printers, for example could become much more precise with no mechanical improvements if they did this. Probably some have already done so.)
Many years ago, I implemented star tracking for telescope alignment for auger.org. An actual basic version using libnova et al was pretty straightforward and a matter of days to prototype if I recall correctly (it was a project for a master student, so I had to do most of the work over a longer period of time). I think the accuracy was around a decimal fraction of a degree then. But the cameras of that thing have only 440 1.5° pixels (photomultipliers that sample at 5-20M samples/s). A higher resolution camera could probably do much better assuming you have a very accurate computer model of the optical properties of your telescope/lenses/camera.
> The difficult problem is not knowing where the sun is but knowing where the ground is. Since all the panels are separate structures not on a single rigid slab, ground "settling" can tilt them fractionally in unpredictable ways.
Ground vibration might also be a problem, in applications where this is being used to generate heat for some heavy industrial application. Presumably in that cast said heavy industrial application would be taking place nearby, and so could disturb the ground under the mirrors or tower.
They could be said to be naturally intelligent, yes. However they only have to point at one target, the sun, while the mirrors have to point halfway between two targets, the sun and the tower.
You can replicate this behavior with four solar panels mounted on a motorized platform. The panels should be at a slight angle, with a peak in the middle... then a voltage differential top-to-bottom activates the vertical motor, likewise the horizontal motor.
I'm not sure how to make such a simple system for mirrors pointing at targets, but I suspect "AI" and "cameras" are overkill.
Gyroscopes drift. Every system using gyroscopes that needs precision has some external tracking system as another reference (e.g. GPS, star tracking). In this case, that external system is a camera on top of the tower.
Yes, you can precalc, and then you need very precise actuators to move the mirror to that precalced position.
You also need precise surveying of the ground site to know its geometry.
The precise actuators and surverying are expensive.
Their approach is using cheap actuators in a closed-loop system where a computer vision controller (using the four cameras on top of the tower) moves the mirrors until it gets it right. Cheap actuator slip and lag is terrible, but with the feedback of the closed-loop, it's doable.
This is an approach that's now allowing hobbyists to build cheaper motorised telescopes, robot arms and so on.
That did strike me as weird. Although they say "advanced computer vision software" so maybe they do some workaround and take the position from a camera rather than pre-calculations? Can it be the case that a few thousandths of a degree matters here and that's beyond our generalised model?
I'd imagine that the "few thousandths of a degree" is something you'd get worse with AI than with a mathematical model. AI methods introduce noise; with an explicit model you can keep adding complexity directly, until it gets more accurate than your measuring system.
I suspect that the "AI" they mention is mostly photogrammetry - being able to tell the position and orientation of your mirrors from a video stream - in a feedback loop with motor controllers and temperature sensors. Use of cameras would make it close enough to "AI" that the buzzword would stick.
>mostly photogrammetry - being able to tell the position and orientation of your mirrors from a video stream - in a feedback loop with motor controllers and temperature sensors.
This. Modern photogrammetry uses computer vision techniques like deep neural nets.
It's an amazing advancement.
Imagine this problem: having to place a ground vehicle like a wheeled robot at an arbitrary coordinate in a room. The mirror positioning problem can be reduced to this one. Traditionally, you'd do it by building a very nice wheeled robot with non-slip wheels, wheel rotation angle sensors, fraction of a degree stepper motors, all very nice to work with, precise, and expensive. You could even solve this in an open-loop system with good enough actuators.
Today, just stick a camera on the ceiling - or on the robot, if want a more challenging problem - and use CV to move a cheap RC car where you want.
The Terminator will jangle and jiggle, instead of being built like a precise Swiss watch. But you'll get ten of them for the same price!
> This. Modern photogrammetry uses computer vision techniques like deep neural nets.
I'm surprised to hear that. What good do DNN do for photogrammetry? Aren't they too unreliable?
I only have university course-level exposure to this, most of it spent learning how to compute coordinate systems and transformation matrices from points on a set of pictures taken from different position. I guess DNNs could help with object isolation and labeling (identifying same objects on a set of pictures taken from different perspective)?
The Related work section in the DF-SLAM paper is great.
As you intuited, DNNs work great for tracking visual landmarks (better than traditional computed features) between frames. They can also estimate depth from a monocular camera, which is very useful.
A more out-of-left-field idea is training a CNN to directly regress the camera pose (position and orientation) from an image:
I would imagine that it’s not the suns location that is the primary concern, but where you are directing the light.
The light is being reflected and focused by mirrors. That light is being directly onto some volume of material. So, in this case, they want to get the material above some critical temperature (e.g. so it melts). I imagine that uniformly heating the whole volume may not be the best approach, or that you may be able to work with smaller volumes depending on your expected solar input etc.
So I can see a little more justification here than the case where solar cells are just being pointed at the sun.
I think the point of calculating using imaging vs a model that requires layout and flat land etc is this would be cheaper to build. You have to build one collection tower/ target, and the mirrors can aim themselves. This greatly reduces the grading, measurement etc and makes installing on existing land without needing to flatten at all, or to some crazy tolerance, much easier. I can see this used in rolling hills, with less impact etc possible.
Very skeptical that this is a practical means to supply 1500C to an industrial process. Do they plan on piping heat via liquid metal, or what? Do you shutdown on cloudy days and sen everyone home?
Since thermal energy is only lost at boundaries, the larger the amount you have, the smaller the net surface area per unit of volume.
This is how you can store huge blocks of ice well into the summer (e.g. https://en.wikipedia.org/wiki/Ice_house_(building)), but also how a giant underground mass of molten salt that's not in physical contact with the surrounding ground could store a lot of heat for a long time.
I'd love to see something like this put into use for smelting titanium. We really need to start embracing it as a building material as it doesn't corrode or break down like wood, steel, aluminum does...plus it's light and strong which would allow moving houses and vehicles made from it much less energy intensive. It would be great to invest the energy into building materials that can last instead of continuously having to fight corrosion.
I wonder who Heliogen hired to do their PR. Must be a really good (or expensive) campaign. There is nothing new in high temperatures and the AI part is pure buzzwords.
Inertia must be incredibly strong. Otherwise, why did it take so long to come up with something that strikingly simple? The last time someone allegedly had an idea like that was around 212 B.C., in the defense of Syracuse.
> Inertia must be incredibly strong. Otherwise, why did it take so long to come up with something that strikingly simple?
It didn't? There are both solar ovens and solar power plants in active use.
The issue's mostly that it's historically been fairly inefficient and expensive (compared to other electricity production sources). Prices have been coming down through improved efficiency and economies of scale, the plants being built these days are bid under $75/MWh and construction has been ramping up a lot especially in high-irradiation locations e.g. Morocco onlined their first station in 2017 (510MW CSP, 72MW PV) and is already building a second one, and Dubai is apparently aiming for 5GW installed by 2030.
Because of oil lobby mostly. The idea is nothing new, it was just easier to burn oil instead of building new industry from ground up and going against all existing oil companies.
That "oil lobby" conspiracy theory is such nonsense. There are several developed countries with minimal oil reserves and no domestic oil companies. If alternative energy industries were really practical before then those countries would have built it.
The technology is fundamentally simple, but not practical on a small scale. There are plenty of large, yet simple projects which nobody cares to invest in because there are better (including more expedient, or proven) alternatives.
CSP requires a large investment compared to PV solar, as it needs to be done at scale to make coolant systems, towers, turbines and generators cost effective. It needs its own large allocation of land, rather than be distributed. It also, like PV solar, has poor temporal availability, so compares badly to fossil fuels, nuclear and hydro.
It has potential to be cheaper than PV, and integrate well with thermal energy storage. It could also be more efficient, if the temperature can be increased (due to Carnot efficiency). However, the convenience and falling prices of PV panels may prevent CSP becoming popular.
Is 6000 birds a year a lot? It's a serious question, I have no idea what fraction of birds in the area that is.
Is it comparable to the normal fluctuation in bird mortality in the area or is it so large that it has a disturbing effect on the population?
Of course fewer bird deaths would be good, and zero would be better.
But these bird deaths are noticed because of where they occur. What is the 'background' level, the death rate in the same place without the solar plant?
The Ivanpah installation is on a known migration route, presumably most other such installations will cause fewer bird deaths simply by not being on such a heavily used route. The article was a bit light on background information and I don't know how to find it.
The Mojave desert wilderness areas are not overrun with domestic cats.
6000 sounds like a lot to me. I have five acres near joshua tree, and while there is wild life the only stuff that seems to be abundant are some rodents, rabbits, and reptiles. It's a fragile ecosystem. There used to be tortoises supposedly, but the roads wiped them out.
For wind turbines at least there's a bit of a difference, as those kill larger birds than cats, in smaller numbers, but those birds also have smaller populations. So just looking at the numbers without knowing the birds and their total populations, makes things hard to compare or judge.
But aren't most larger birds predators that are highly habitat/food supply constrained? That kind of population could be quite resilient to moderate increases in mortality.
(I might be assigning far more importance to predator-Peru cycles than is warranted though, I was reading about what I now found is called https://en.wikipedia.org/wiki/Lotka%E2%80%93Volterra_equatio... at a very young age and the concept stuck for life)
Sure, thought needs to go into the location of these things; avoiding migration routes and habitats of endangered birds. Using scarecrows and possible trained predator birds (presuming these can be trained to avoid the danger zone!) etc.
Of course conventional industry doesn't get a free pass here either, treatment and storage ponds, climate change, habitat destruction etc.
"started operating in 1970", "Temperatures above 3,500 °C (6,330 °F) can be obtained in a few seconds".
See also https://en.wikipedia.org/wiki/Solar_furnace
Not dismissing this press release, but at least some scale of progress should be used to compare against. Or state where the economical/industrial/regulatory innovation stands.