Comment by drzaiusapelord | Hacker News Reader

Comment by drzaiusapelord | Hacker News Reader

Parent Full thread

I guess my point is, can we do a full simulation of every screw, material, plumbing, liquid dynamics, weather dynamics, etc and augment those with known fail scenarios and other fuzzy data to build out a real world KSP that predicts fails reliably? We should understand how things like corrosion and condensation work on a rocket engine. Considering the low cost of incredible amounts of CPU power, granular level simulation is possible on a certain level today if someone wanted to create it. We certainly see this kind of thing with stealth technology, where we can simulate every permutation of near every radar photon hitting the various surfaces of planes with various materials, scenarios, temperatures, etc.

I imagine this level of simulation might not be entirely feasible yet. Maybe for the lack of trying or budget. In a growth industry or one powered by both commercial and technical pressures, it may be difficult to sit down and build something like this out. From a more practical point of view, it may make sense to just let things explode than spend years running expensive simulations instead of building things, launching, and collecting paychecks.

I think a critical part of his point is that what you build and what you model are neccessarily two different things. You can totally simulate it, and then when the rocket inevitably fails, you can go "aha! This bit deviated from the simulation!" But you can't feed that forward beforehand to prevent the failure, since you can't expect random acts of poor workmanship or crafting. You can only prepare and hope (and you can do that with shockingly high confidence - but rockets are more than equally shockingly complex.)

Part of what makes the idea of bringing the first stage back to the pad so important is that we so rarely get to use the same engines multiple times in the field (where all the really nasty reality checks are done). Being able to reuse stages allows us to far better model how they will perform in the future. Otherwise, we're using test beds to feed parameters into sims to inform our launches; it's good practice, but more physical evidence is always better.

This.

Engineering is what hopefully guides reality up the correct branch of a theoretically possible tree.

You can simulate most of each one of those branches. But what are you going to do with a million simulation results? How does that guide your course of action? What do you do differently?

If this was an engineering or assembly defect, the answer is always going to be "Don't do that next time." If it was a design defect, then the part wasn't simulated (unlikely) or our understanding of how it operated in this design was incomplete (more likely).

The trick with rocket science is that the design tolerances are by necessity very tight. Physics dictates this with chemical propulsion. Every part you over-engineer in a weight-increasing way decreases the weight available for payload. And there isn't very much weight there to start with...

The issue is that rocket physics (essentially Newton's laws of motion) are for the most part first- and second- order, linear, ordinary differential equations with exact solutions. The plumbing, on the other hand, is governed by fluid dynamics, heat transfer, and the like. These are systems of complicated partial differential equations without exact solutions, requiring numeric methods.

>I guess my point is, can we do a full simulation of every screw, material, plumbing, liquid dynamics, weather dynamics, etc and augment those with known fail scenarios and other fuzzy data to build out a real world KSP that predicts fails reliably?

Frankly, no. You're wildly underestimating the complexity of the system.

The radar technique is possible because of the trivially-parallizable finite-difference time domain method, which most certainly does not simulate every photon (this may have been a rhetorical flourish on the part of wherever you learned this; FDTD curiously has zero error term). Nor would it simulate a wave hitting from every direction; each of those would be a separate run of the simulation.

no, we can't. we currently do not have the CPU power to do molecular simulations of fluids and solids (beyond the grain level) at large scales, that's why we still have supercomputers working at simplified versions. additionally, we don't have 100% fidelity in knowing environmental conditions, what the weather is doing at that exact point, the exact velocity/temp field of the atmosphere (which is constantly changing) at the molecular scale, which would be essential for a 1:1 simulation.

https://en.wikipedia.org/wiki/Direct_numerical_simulation

A lot of simulation already happens but simulating every imaginable scenario is impossible.

Reply on news.ycombinator.com