Hacker News Comments on
SpaceX Merlin (& Raptor ) Engine R&D, GPU-Powered
Burn Hard Zen
·
Youtube
·
4
HN points
·
6
HN comments
- This course is unranked · view top recommended courses
Hacker News Stories and Comments
All the comments and stories posted to Hacker News that reference this video.SpaceX has done critical work on GPU accelerated computational fluid dynamics for modeling rocket engine combustion: https://www.youtube.com/watch?v=txk-VO1hzBYAnd saying this is about money and resources, and not focus and management, ignores the fact that SpaceX spent less than ULA and Blue Origin to develop 2 rocket engines, 3 rockets, three different recovery systems, and 2 different capsules. Blue Origin has been spending $1 billion a year of Jeff Bezos's money, and despite starting before SpaceX, still hasn't reached orbit or beaten the Falcon 1.
Rocket Lab's Electron made it to orbit with $210m of total funding, and they built their own engine with an electric pump. Far smaller than many other companies.
The reason why SpaceX is so successful is because Elon Musk isn't in it a vanity project, or to make money taking a slice of the LEO sat launch market. He has a vision, call it lunacy, of getting huge numbers of people to Mars. Working backwards from that goal, informs everything they do, which is why they won't settle for an incremental improvement on what ULA or Arianne is doing.
It's like saying if you gave equal money in 2001-2006 to someone to build better mobile phones, they would have produced an iPhone, because most of the components were there. I worked in the mobile market at that point, and I can tell you most people were chasing Blackberry, or imagining slightly better versions of say, an Nokia Communicator 9000, iPAQPhone, or OQO. The iPhone took a vision, and someone willing to tell the naysayers and corner cutters that he doesn't care about their objections, this is what they're doing.
For TEN YEARS on USENET, I read sci.space as people like Henry Spencer and other luminaries decried reusability. They rightly said that anything wings, or landing legs, or extra strengthening and margin you add to have repeated flights comes out of your payload. The major consensus of the time would be to tell Musk his ideas won't work and the physics say disposable rockets are the most efficient.
But marginal cost matters more than Isp or payload fraction, and if you have a system where the majority of your costs are amortized over many launches, lower payload is less relevent, you just do more launches.
Ultimately, SpaceX will prove the naysayers wrong, as surely as he proved the establishment and ULA's allies wrong.
GPU acceleration of dynamically scaling cell-based fluid dynamics simulation. Video lecture / demo from 2015; 45 minutes.I don't know if it was state-of-the-art but it looked impressive to me as a layperson.
To model Raptor's hypersonic turbulent combustion SpaceX used an internally developed simulator, which uses wavelet compression to vary resolution across many orders of magnitude in both time and physical dimensions:https://www.nextplatform.com/2015/03/27/rockets-shake-and-ra...
Here is a fantastic talk from the NVIDIA conference:
Someone in the FEA/CFD/Simulation industry upon seeing the SpaceX video on their capability indicated that no commercial product he was aware of could match what they were doing. Firewall won't let me confirm this link is correct:https://www.youtube.com/watch?v=txk-VO1hzBY
SpaceX has best in class hypersonic combustion CFD capability that runs on a GPU. Written in-house.
⬐ BalgairTo be fair, it is a VERY hard problem. 'Most' airplanes are designed in a laminar flow situation. Meaning that turbulence is just left out. This makes the calculations much easier, as you can set boundary conditions. In super-sonic conditions, the air is no-longer in-compressible, so the boundary conditions change and you have to add more variables to the equations. In hyper-sonic flow, the air itself is now a variable, as the molecules are split and recombine along the wing in ways we have not yet even empirically measured. This ads a complexity to the equations that is very large and the 'educated guess' of those chemistries on the wing becomes very important.⬐ phkahler...as the molecules are split and recombine along the wing in ways we have not yet even empirically measured.I'm sure one can speculate. In the video they do talk about all the possible intermediate species, so one might have to add air molecules (which can break apart and add to the reactants) to the mix. But yeah, lots of unknowns. Doing experiments and comparing to the simulations should help figure it out. The problem has traditionally been an inability to do the simulations, but that's possible now.
I came across a nice video/presentation from some guys at SpaceX on using GPUs to model and visualise the fluid flows inside rocket engines (specifically the Raptor engine) and around spacecraft during reentry, as well as a good explanation of how they do it efficiently using adaptive grids. Their explanations are quite approachable as well.
⬐ BromsklossDo they have code available?⬐ tankenmateVery unlikely, this software is an enormous commercial advantage. Not so much the cost in and of itself of the software but for the advantage of speeding up development of engines, etc through shortened iterations. SpaceX developed this because they couldn't find any reasonable solutions on the market already. I suspect that reasonable not only included cost and availability but also the ability to adjust the product to their needs.⬐ tobmltNote the citations at the end of the video (44:30 mark ). Code from some of the original research the spacex stuff was based on, adaptive wavelet collocation methods, used to be online in a bitbucket repo here: https://git.plgrid.pl/scm/tmh/wavelet-original-code.gitUnfortunately it seems to have been taken down. A shame - it was really everything! But as a huge CFD library research code it is a bit hard to sift through anyway.
I'd recommend the papers then. Stuff like this, from these guys (Jonathan Regele and Oleg Vasileyev), led to the Spacex software: https://www.researchgate.net/publication/239395184_An_adapti...
An earlier starting point might be something like this: http://ms.mcmaster.ca/kevla/pdf/articles/Kevlahan_SISC_v26.p...
This line of work dates back to the mid 90s as you will see from the sources.
Science has yet to discover the ultimate physical laws that govern the universe so 100% simulation is still impossible even in theory. In practice high fidelity simulation is still hard. Here's a video about the state of the art in combustion chamber simulation and what it's current state is: