3 ms·
The common reaction to "this process is too slow" is to ask the question, "what if we made it fast"? As I mentioned the last couple of times this topic came up
by hinkley 4y ago
The common reaction to "this process is too slow" is to ask the question, "what if we made it fast"?
As I mentioned the last couple of times this topic came up, I am concerned that the answers to that question that have already been offered are ignoring how the statistics of scalable system work and particularly where they overlap with the Rule of Large Numbers.
It's been suggested that if we want to go fast what you do is make a bunch of small reliable reactors and then co-locate them, but if you make something that's 10x as reliable, the assumption is that having 10x as many of them results in the same failure rate, and the same consequences. But that's not quite how statistics work, or for that matter equipment.
When you have 100 of something, there is always preventative maintenance or repairs going on. We have storage array designs specifically to deal with this problem in data centers by just letting equipment fail and then replacing whole units, because the labor involved with reacting to each event is simply too high. That's not particularly unique to IT. That's just a fact of large scale operations. You keep spare tools because some of them are going to break every week. You swap them out and -maybe- someone repairs them, or cannibalizes them.
With mechanical equipment, maintenance materially affects the reliability and safety of the job site. Suspending maintenance and repairs because you've been pre-empted by the failure of other equipment leads to risks stacking up higher and higher. Can I do scheduled maintenance on Reactor 3 if Reactor 50 ate itself yesterday? At some number of reactors, every plant has experienced one catastrophic failure. What do you do with that? What can you do with that?
- angry_octet 4y agoExactly correct. The techniques for repair of radioactive equipment are still in their infancy, and seem to progress only with each accident (due to their inherent costs). Robots die in minutes in a hard radiation environment, and we don't make any new microprocessors with rad hard techniques (though maybe RISC V will change that). From a safety critical analysis POV there has been a huge change in technical capacity over the last 20 years, but no one in the field would say we're close to strong safety, especially now in regards to software intensive systems. IoT just isn't going to be the same with radiation, you can leap from a correct state to an unsafe one at any point, dozens of bits being flipped every second, transistors misbehaving, faults between silicon layers... Completely unrelated logic is in the same fault compartment of critical trusted logic. So if you have 50 reactors, that's 50 control systems, plus the shared infrastructure (water/electricity/data/fuel in and out, fire suppression). If you remove one for refuelling you're moving/vibrating a radioactive system with lots of pipes which are now fragile due to radiation. All the stress analysis techniques we use on planes are very hard to use on plants due to lack of access. It seems to me like we're just beginning to understand the complexity of SMRs and, contrary to the PR, they will take a long time to mature. Everything about nuclear is nonlinear and coupled. Production line techniques are all about equipment that operates in the known region, and the consequences are not too bad when it departs from that because the operator is at fault.
- radu_floricica 4y agoAs far as probability of failure, having a large number of small powerplants raises it to basically 1. But as far as consequences go, it's a very different story. If you have a 10 kg rock dropping on your head, you're very likely dead. For 10 kg of gravel you've very pissed, and for the same amount of sand you're... possibly having fun? We don't care that much about the probability of failure, because as a society we've already decided we REALLY don't like nuclear disasters. I don't think you could convince anybody that going from 1.5% to 0.5% is good enough. But lowering the maximum impact of a failure is a different game - it just means what we now call "disaster" becomes not just unlikely, but impossible.