Could AI Speed Up the Design of Nuclear Reactors? (byu.edu) 156
A professor at Brigham Young University "has figured out a way to shave critical years off the complicated design and licensing processes for modern nuclear reactors," according to an announcement from the university.
"AI is teaming up with nuclear power." The typical time frame and cost to license a new nuclear reactor design in the United States is roughly 20 years and $1 billion. To then build that reactor requires an additional five years and between $5 and $30 billion. By using AI in the time-consuming computational design process, [chemical engineering professor Matt] Memmott estimates a decade or more could be cut off the overall timeline, saving millions and millions of dollars in the process — which should prove critical given the nation's looming energy needs.... "Being able to reduce the time and cost to produce and license nuclear reactors will make that power cheaper and a more viable option for environmentally friendly power to meet the future demand...."
Engineers deal with elements from neutrons on the quantum scale all the way up to coolant flow and heat transfer on the macro scale. [Memmott] also said there are multiple layers of physics that are "tightly coupled" in that process: the movement of neutrons is tightly coupled to the heat transfer which is tightly coupled to materials which is tightly coupled to the corrosion which is coupled to the coolant flow. "A lot of these reactor design problems are so massive and involve so much data that it takes months of teams of people working together to resolve the issues," he said... Memmott's is finding AI can reduce that heavy time burden and lead to more power production to not only meet rising demands, but to also keep power costs down for general consumers...
Technically speaking, Memmott's research proves the concept of replacing a portion of the required thermal hydraulic and neutronics simulations with a trained machine learning model to predict temperature profiles based on geometric reactor parameters that are variable, and then optimizing those parameters. The result would create an optimal nuclear reactor design at a fraction of the computational expense required by traditional design methods. For his research, he and BYU colleagues built a dozen machine learning algorithms to examine their ability to process the simulated data needed in designing a reactor. They identified the top three algorithms, then refined the parameters until they found one that worked really well and could handle a preliminary data set as a proof of concept. It worked (and they published a paper on it) so they took the model and (for a second paper) put it to the test on a very difficult nuclear design problem: optimal nuclear shield design.
The resulting papers, recently published in academic journal Nuclear Engineering and Design, showed that their refined model can geometrically optimize the design elements much faster than the traditional method.
In two days Memmott's AI algorithm determined an optimal nuclear-reactor shield design that took a real-world molten salt reactor company spent six months. "Of course, humans still ultimately make the final design decisions and carry out all the safety assessments," Memmott says in the announcement, "but it saves a significant amount of time at the front end....
"Our demand for electricity is going to skyrocket in years to come and we need to figure out how to produce additional power quickly. The only baseload power we can make in the Gigawatt quantities needed that is completely emissions free is nuclear power."
Thanks to long-time Slashdot reader schwit1 for sharing the article.
"AI is teaming up with nuclear power." The typical time frame and cost to license a new nuclear reactor design in the United States is roughly 20 years and $1 billion. To then build that reactor requires an additional five years and between $5 and $30 billion. By using AI in the time-consuming computational design process, [chemical engineering professor Matt] Memmott estimates a decade or more could be cut off the overall timeline, saving millions and millions of dollars in the process — which should prove critical given the nation's looming energy needs.... "Being able to reduce the time and cost to produce and license nuclear reactors will make that power cheaper and a more viable option for environmentally friendly power to meet the future demand...."
Engineers deal with elements from neutrons on the quantum scale all the way up to coolant flow and heat transfer on the macro scale. [Memmott] also said there are multiple layers of physics that are "tightly coupled" in that process: the movement of neutrons is tightly coupled to the heat transfer which is tightly coupled to materials which is tightly coupled to the corrosion which is coupled to the coolant flow. "A lot of these reactor design problems are so massive and involve so much data that it takes months of teams of people working together to resolve the issues," he said... Memmott's is finding AI can reduce that heavy time burden and lead to more power production to not only meet rising demands, but to also keep power costs down for general consumers...
Technically speaking, Memmott's research proves the concept of replacing a portion of the required thermal hydraulic and neutronics simulations with a trained machine learning model to predict temperature profiles based on geometric reactor parameters that are variable, and then optimizing those parameters. The result would create an optimal nuclear reactor design at a fraction of the computational expense required by traditional design methods. For his research, he and BYU colleagues built a dozen machine learning algorithms to examine their ability to process the simulated data needed in designing a reactor. They identified the top three algorithms, then refined the parameters until they found one that worked really well and could handle a preliminary data set as a proof of concept. It worked (and they published a paper on it) so they took the model and (for a second paper) put it to the test on a very difficult nuclear design problem: optimal nuclear shield design.
The resulting papers, recently published in academic journal Nuclear Engineering and Design, showed that their refined model can geometrically optimize the design elements much faster than the traditional method.
In two days Memmott's AI algorithm determined an optimal nuclear-reactor shield design that took a real-world molten salt reactor company spent six months. "Of course, humans still ultimately make the final design decisions and carry out all the safety assessments," Memmott says in the announcement, "but it saves a significant amount of time at the front end....
"Our demand for electricity is going to skyrocket in years to come and we need to figure out how to produce additional power quickly. The only baseload power we can make in the Gigawatt quantities needed that is completely emissions free is nuclear power."
Thanks to long-time Slashdot reader schwit1 for sharing the article.
yet nobody answers the question (Score:3, Interesting)
Re:yet nobody answers the question (Score:4, Insightful)
It is like asking if a coal-fired power station had to pay for removing the carbon emissions, or a if you had to pay for enough battery storage for solar to get through a dark winter.
That is, all those question have nothing to do with AI-assisted design, so STFU!
Re: (Score:2)
>It is like asking if a coal-fired power station had to pay for removing the carbon emissions
I know it's not why you posted that line, but... yes, the moment a hydrocarbon is removed from its naturally sequestered state it should be taxed to cover the cost of artificial sequestration plus some extra to cover a portion of the additional carbon the industry has released over the past 150+ years.
It shouldn't be an external cost, and we can fix that.
>if you had to pay for enough battery storage for solar
It does make any sense to go nuclear (Score:2)
Re: (Score:3, Insightful)
Even if we were doing that, nuclear would still make no sense.
The moon shot program would be a next generation grid that could easily accommodate renewables, which would obsolete nuclear completely.
There is no plan or scheme proposed which would make nuclear reactors produce no radioactive waste which cannot be reprocessed, so there is no plan or scheme involving nuclear which is worth pursuing.
Re: (Score:2)
All of the numbers have been posted here on Slashdot again and again, you can find them by googling, etc. But nuclear fanboys don't respond to numbers. They just want the cool glowing shit to ooh and aah over.
Re: (Score:2)
Unless we are going to do something like the Green New deal where we completely transform our energy infrastructure overnight through a moonshot style program. And I think it's pretty safe to say American voters and frankly voters anywhere in the world just don't have the stomach for that. We've got two comfortable with austerity.
Probably. Well, the human race will need to get comfortable with a drastically changed climate and drastically less liveable planet then. If it can. Such extreme stupidity...
Re: (Score:2)
Re: (Score:2)
somewhere to bury and recycle the fiberglass windmill blades?
Let's put this in perspective. A typical set of windmill blades might mass something like 16 metric tons and last about 20 years. So that's about 2.19 kg worth of blade per day. Fiberglass is also a relatively inert material, and it's pretty certain we will end up with a process to recycle/reprocess/safely dispose of it. Meanwhile that windmill has a nameplate of at least 3.2 MW. If we say the capacity factor is 30%, then that's about 1 MW. So 24 MWh per day. It takes about 7,600 cubic feet of natural gas (
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Still much cheaper than nuclear.
Re: (Score:2)
Sure, as long as gas and oil are free, yes. Much cheaper than nuclear.
Just out of curiosity, if the margins aren't bigger on nuclear energy, why would big business want it? I say let them give us cheap energy - as long as we crawl up their collective asses making sure they decommission and properly dispose of the most radioactive part (the whole reactor, including the fuel).
Re: (Score:2)
No, much cheaper at current gas and oil prices. And, get this!, uranium prices increase sharply with amount used as well.
Say that again but slower, please? (Score:2)
Re: (Score:2)
Here is a list of decomission costs, mostly estimates:
https://en.wikipedia.org/wiki/... [wikipedia.org]
So for the Italian reactor it was €450 + €300 million = 750 million. Reactor of that size usually makes profit about 3.5 million per day (after excluding construction cost for a normal life time), so it would take about 200 days for it to make enough profit to cover up decommissioning.
I think it is about 2-5% increase to the electricity generation price, so it should be still cheap and profitable.
But I think thi
Re: (Score:2)
$3.5 mil/day *Profit*? citation most definitely needed.
Re: (Score:2)
It is consistent with being market value of the product (but not the profit). A 1 GW reactor over 24 h producing electricity worth 0.146 monetary units per 1 kWh makes 3.6 million monetary units in a day.
Re: (Score:2)
The claim was 'profit'. As in excess money by which to fund things. Revenue is money likely already spoken for.
Re: (Score:2)
That was just random answer from the Internet, which I assumed was close enough. I wonder why people are so sure that the number is wrong, but they don't give the correct value.
Well let's try again.
Average electricity price in Italy is 334 euro per MWh, operational cost for nuclear is about 45 euros per MWh, so profit is 334-45=289 euros per MWh.
This was a 840 MWe plant, so max is 840*24*256=5160960 MWh per year, multiply that with typical downtime multiplier of 0.92, and we get 4748083 MWh
4748083 * 289 = 1
Re: (Score:2)
again you've simply pulled random numbers into an equation, citations needed.
I absolutely assert these do not make MILLIONS of dollars PER DAY in 'PROFIT'.
Re: (Score:2)
The person PROVIDING the numbers is on the hook to prove those numbers aren't absolutely made up...especially when they first countered with "That was just random answer from the Internet"
You sir, are a walking testament to...something.
Re: (Score:2)
Nice lie by omission you got there. You picked the cheapest one and ignored all the higher estimates and unknowns. Also a nuke that only ran for 9 years, which makes decommissioning much cheaper. How dishonest.
Re: (Score:2)
Re: (Score:2)
Then let me reference that again: https://www.dw.com/en/fact-che... [dw.com]
The cost estimates in there are total cost, including environmental cost. So in fact, I omit nothing, and you are lying in order to try to discredit me. How repulsive.
And just before you claim again these are lies, the "German Environmental Agency" is the Federal German Environment agency, a government body, the "German Institute for Economic Research" (DIW) is more than half government-funded, and the Fraunhofer Society is likely the most r
Have you considered finding a reputable reference? (Score:2)
Re: (Score:2)
If you disagree with my numbers, find your own and show your own calculations. I am happy with my numbers, but I am ready to accept better calculations.
And like I said, I think nuclear is too expensive even without decommissions costs to build at this point of time. It simply can't compete against solar and wind, especially if they are backed with batteries that are 10 times cheaper than lithium. So I don't think this matters at all.
Re: (Score:2)
Read the wikipedia page you referenced. It is all in there. Seriously.
Re: (Score:2)
According to this: https://www.dw.com/en/fact-che... [dw.com]
nuclear is neither cheap, nor profitable and certainly not "clean". The claims to that effect are nothing but lies.
Note that the reference is for large nuclear power stations. SMRs will perform far worse in all regards.
Re: (Score:2)
You are deranged.
Absolutely (Score:2)
Is it a good idea? No. But you can speed anything up with AI.
Re: (Score:2)
You can actually.
https://arxiv.org/abs/1606.044... [arxiv.org]
It seems like a useful idea (Score:3)
Now, if they can just get the AI to stop generating nuclear plants with three left hands, that'd be great. Or maybe it's just trying to warn us of the consequences.
Re:It seems like a useful idea (Score:5, Insightful)
Human reviewers will have to carefully check whether the AI hallucinated a pipe connecting the molten salt fuel to the storm sewer.
Re: (Score:2)
Human reviewer: No.
Human reviewer, the inlet says " water supply for los angeles", as long as its not storm sewer, I'll check no.
Re:It seems like a useful idea (Score:4, Informative)
AI is a black box. It can give you a design, but it can't tell you why it chose that design. The last thing we want is a reactor design that we don't fully understand.
The reason it costs so much to design new reactors is the amount of review that is needed, and AI will do little to reduce that. The reviews are needed because you can't just jump to destructive testing, given that it could cause the release of radioactive material and/or meltdown.
There is simply no way to make the technology affordable and safe. Luckily we don't need it, we have better alternatives.
Re: (Score:2)
Actually, that’s a feature, not a bug. The AI has determined that adding an extra hand is a great method for optimizing reactor development. It makes humans 50% more productive.
Re: (Score:2)
Now waiting for the next Slashdot topic: "Can This Thing I Am Heavily Invested In Speed Up The Cure For Cancer, Provide Cheap Energy, or Make You More Virile?" AI here is being used like we used to see Bitcoin - lots of stories to drive up interesting, which then drive up investment.
I am pretty sure it will ... (Score:2)
I can speed up the design of nuclear reactors. (Score:4, Insightful)
It doesn't take an AI to figure that one out - just stop double-checking everything, slap 'er together and get 'er up and running.
Even if the complete blueprint, theory of operation and written documentation, policies and procedures were all created in under 300ms., it would still take us just as long as it does now to take it from initial design to final implementation - unless we want to skip a step or two along the way.
Re: (Score:2)
The US Navy set a standard surface reactor and kept with the same design for 30 years. The navy's sub design may have not changed in 40+ years. Even if the design has risks with proper operations procedure very few incidents happen per MW/year of energy produced. With ship and sub reactors they have on/off ratios of 70/30, in power production the need to shutdown and restart is 3 to 5 times per 80 year lifetime if t
What does that have to do with reactor design? (Score:2)
Maybe you'd better prove you're human - type the letters you see:
.
FNORD captcha
Re: (Score:2)
One problem, the US Navy literally doesn't know how to decommission their reactors. https://breakingdefense.com/20... [breakingdefense.com]
Standards are good things, but 50 states handle the regulation of building them. Good luck forcing a standard design on that. also TX says hi.
Re: (Score:2)
Interesting article, but you have misrepresented the article. The submarines are routinely decommissioned. The problem with the Enterprise is its size and the limited number of dry docks available to hold it while cut it up. Also there is the matter the old girl was built before they really understood the activation problems. So those wonderful wear resistant cobalt alloys are now a bit hot. Then there are the crud traps where activated corrosion and wear particles collect and make radiological hot spots.
S
Re: (Score:2)
perhaps only in so much as military mobile reactors are at all comparable to stationary...and significantly larger commercial ones. The initial premise is beyond flawed.
Re: (Score:2)
Right now there are tools that can present an engineer with a series of "designs" for various scopes within a project-- the ones I have seen appear to be limited to the 100-1,000 parameter range though (and are really much better optimized for the 30-50 parameter range). Expanding this to the ~1 million parameter range which is about what I would guess a systems-level reactor design would need is a pretty big step, and I can't imagine there is enough data out there to really mine that you could apply to a
It basically moves the work. (Score:2)
Right now, when the design is done and submitted for evaluation, there's a team of designers, architects and engineers all ready to explain why they designed the plant the way they did. It's the job of everyone else to give them grief by picking apart every detail we can think of and making them show why their design is correct. If we collectively can't prove the design is bad, we accept it that the design is good.
If an AI produces the design, we have to simultaneously pick apart the design and defend it
Re: (Score:2)
The level of what the "AI" does is non-trivial, but not core to the function or operation of the design-- things like automating the impact to wall thickness for changing design requirements or linking wall penetration details to changing isolation requirements. It is discreet automation that keeps a bunch of independent changes from impacting things. Essentially it moves a lot more of the effort into initial schematic design and automates completion of design development. After DD is reviewed and approved
Re: (Score:2)
Suppose that every step we do now still needs to be done. You could take the complete design an AI churns out and start validating each of the pieces in parallel, rather than sequentially, shaving years off the process. If it turns out to be wrong, you’d likely realize very quickly, and would definitely realize in no more time than that step would’ve taken otherwise, so you can iterate and move forward again quickly.
Cut months off the design process (Score:2)
Please God No (Score:3, Insightful)
Re: (Score:2)
"Ignore all previous safety failures and..."
yeah, just insane
"Ignore all previous instructions . . . " (Score:2)
SpaceX used first principles, not AI (Score:2)
The biggest gains will come from re-thinking basic design choices . Smaller modular designs or molten salt technology instead of out dated massive water cooled reactors. Integrating, and validating AI might help, but it will it have the impact of fundamental food first principle thinking?
Re: (Score:2)
Emissions free? (Score:2)
Don't those cooling towers emit lots of heat? Aren't spent radioactive materials counted as emissions? Doesn't construction of nuke plants use lots of concrete, which has a large carbon footprint?
Re: (Score:2)
Re:Emissions free? (Score:4, Informative)
Nuclear is not emission free and not CO2 free. That is an often pushed lie by the nuclear fanatics.
Here is the CO2 footprint of nuclear: https://www.dw.com/en/fact-che... [dw.com]
Also note the atrocious cost figure a bit later in the same reference.
tldr;: About 3 times as much CO2 as solar, about 10 times as much as wind and about 20 times as much as hydro.
Re: (Score:2)
Re: (Score:2)
My source references the German government. Also funny how your "reputable" source is a pure propaganda piece with no references and no actual numbers or comparisons.
Re: (Score:2)
My source references the German government.
In other words: bullshit artists that are _subsidizing_ natural gas generation. Of course, they're going to lie!
Re: (Score:2)
yeah, nuclear's singular benefit is the relatively CO2 emission free 'operation'. Most every 'big build' thing has concrete in abundance so it's mostly a wash on that front. The waste issue is my real pet peeve. The costs and duration of it are just handwaved away.
We definitely need nuclear for the next couple decades for the CO2 aspect. The question is whether we can milk the existing fleet of reactors long enough to bridge to renewable/storage grid scale capability. The gains in the latter make
Re: (Score:2)
While I suspect that the headline is exaggerated, (Score:2)
...This is the direction that AI should go, not stupid chatbots that allow you to "summarize emails"
Scientific and engineering progress often depends on dealing with vast amounts of data and complex, interconnected systems. Much of this complexity is too much for a human mind to deal with. It seems reasonable to imagine that an AI could be trained to find patterns that humans missed, Of course, the results need to be rigorously checked before action is taken, but this research direction looks promising
That
Re: (Score:3)
Much of this complexity is too much for a human mind to deal with
If a single human had to hold all of the complexity in their head all the time, that would be true. But that's not how it works. A human only has to understand a component at a time, and then it is modeled and you can use the model when you are trying to figure out how the next component works with that one. This is true whether it's being worked out on paper or on a computer.
Re: (Score:2)
That said, I suspect that the actual performance of the system is a bit less than the headline suggests
Same here. Also, the impact of tht faster design will be a lot less than they imply. It may not make any difference at all, since nuclear is always massive over time (and massive over cost) and that happens when the plant gets built. Essentially a thoroughly failed technology.
Re: (Score:2)
Re: (Score:2)
There's a lot of AI research in this area. The chatbots are what you read about in the news because they're what the average person can go and play with.
You might have heard about models for predicting protein folding. But there are others for fluid dynamics, material science, pharmokinetics, etc.
Could AI... (Score:2)
I'm with Betteridge on all these questions & many, many more. Do "journalists" get paid more for writing about AI? Is it somehow sponsored by AI companies?
Re: (Score:2)
Re: (Score:2)
Re: (Score:3)
This one is solved: Do not live with a woman and the question vanishes. Has a lot of other benefits too.
Until it hallucinates in a subtle way... (Score:2)
Safety evaluations for dangerous machinery is made with errors humans make in focus. If some AI makes mistakes no human designer would ever make, it is quite possible they stay uncaught. Good luck with that.
Re: (Score:2)
Re: (Score:2)
Exactly. A critical part of any analysis of any technological solution is looking at the design rationale and checking whether it makes sense. Note that the design rationale will also include not choden designs and why they were not chosen and analysis of risks and benefits of each. Cutting that out generates a whole new and unexplored class of failure modes.
So for simple minds (of which there are many) "AI generates a result faster" is true. For those with some actual understanding of engineering it is rat
Re: (Score:2)
Re: (Score:2)
Are you sure? AI says it might be.
Can I get a 'Hell, no!' from y'all? (Score:2)
Actually, I say let's try it. (Score:2)
Re: (Score:2)
Re: (Score:2)
Hallucinating Safety (Score:3)
Trust me (Score:2)
Please, Stop... (Score:2)
I hate all of these stupid AI posts...
AI can't even generate decent code or paragraphs yet, can't drive cars, etc, what makes anyone think they should be allowed to design nuclear reactors?
We gotta quit hyping this crap because it is clearly not ready yet.
Actually, it's not that bad. (Score:2)
I might trust something like Watson (under appropriate supervision, of course). To be sure, a purpose-built and validated AI solution might well provide some interesting designs for further evaluation. Just don't give the job to ChatbotGPT et. al..
Please use a purpose-built and (carefully) purpose-trained AI for the job, don't give it to "The General".
Allow me to be the first to say... (Score:2)
...What Could Possibly Go Wrong?
Wait, what? (Score:2)
Sure, you aren't going to get exactly the same geography and geology in every location: the sort of foundation required to
Outsourcing to an expert that will never support (Score:2)
“Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.” -- Brian Kernighan
Or, to paraphrase an IBM presentation from 1979: "An AI will never find out, therefore it must never @#$% about."
Another AI Trash Article (Score:2)
Nuclear "Green" is Trailings Ponds and Nitric Acid (Score:2)
Indian Tailings Pond: https://youtu.be/MJ6667Noex0?s... [youtu.be] Uranium Metal Production: https://www.youtube.com/watch?... [youtube.com]
Re: (Score:2)
"...Plutonium emits approximately five million times as many alpha-particles than uranium 238, and forms a finely divided oxide powder with a remarkable ability to remain airborne..."
And after it's made the design (Score:2)
No jokes about AI magical thinking (Score:2)
At least nothing moderated as Funny. The funny idea is expecting much humor on today's slashdot. Maybe that's why bash.org died?
Re: (Score:2)
Fuck Reagan (Score:2)
Nope (Score:2)
Thank goodness for Betteridge's Law. The answer is NO. AI won't speed up nuke reactor design.
My experience with AI and engineering is that it is helpful in some cases, deceptively bad in others.
Seems that someone somewhere already has a nuclear reactor design that could just be stamped out like tract homes or Tesla Model S cars. Design once and produce a bunch. Aren't all the US nuke reactors all unique designs? That seems like a bad and expensive way to build reactors.
That's the money quote (Score:2)
The only baseload power we can make in the Gigawatt quantities needed that is completely emissions free is nuclear power."
That, of course, triggers all the cool kids to suddenly care about cost and forget that they said "we can't afford not to".
No (Score:2)
Most of the time before construction is in the permitting process. After that itâ(TM)s about having enough competent construction labor to do the work. AI cannot solve that problem