Darwinian spacecraft engine to last twice as long
Darwinian spacecraft engine to last twice as long
http://www.newscientist.com/article/mg2 ... -long.html
SPACE agencies may one day have Charles Darwin to thank for the longevity of their spacecraft. The life expectancy of a popular type of ion engineMovie Camera has been almost doubled using software that mimics the way natural selection evolves ever fitter designs.
Electrostatic ion engines are becoming popular in space missions. Instead of relying on burning large amounts of heavy liquid propellant for thrust, they use solar power to ionise a small supply of xenon gas. A high voltage applied across a pair of gridded electrodes sends the positively charged ions rushing at high speed towards the negative electrode. Most ions pass through the grid, generating thrust.
However, some ions collide with the grid itself, causing it to gradually wear out, says Cody Farnell, a space flight engineer at Colorado State University in Fort Collins. Simulations suggest grids in a typical NASA engine will last 2.8 years - but Farnell wondered whether changing the grid's design could extend its lifespan.
He used evolution-mimicking software called a genetic algorithm (GA), and started by instructing the algorithm to randomly generate values corresponding to the geometry of the grid and the voltages applied to it. These values can be thought of as analogous to genes.
Each combination of values was then fed into a simulator to give an idea of the grid's performance and its expected lifetime. If the performance was promising, the "genetic material" was subjected to further random changes, or mutation, and this process was repeated until no more improvements were forthcoming.
After 100 generations, the GA spawned a geometry/voltage set that boosted the ion engine grid's lifetime to 5.1 years - at least in the simulator (Journal of Propulsion and Power, DOI: 10.2514/1.44358). Factors optimised included grid hole diameter, hole spacing and the thickness of the grids. The engine could be improved further, says Farnell, by evolving the other parts too.
"After 100 generations, the algorithm spawned a design that almost doubled engine lifetime"
Propulsion engineer Steven Gabriel at the University of Southampton in the UK is developing a triple-gridded ion engine. He says Farnell's work could have major implications if real-world tests of his design match the simulations.
SPACE agencies may one day have Charles Darwin to thank for the longevity of their spacecraft. The life expectancy of a popular type of ion engineMovie Camera has been almost doubled using software that mimics the way natural selection evolves ever fitter designs.
Electrostatic ion engines are becoming popular in space missions. Instead of relying on burning large amounts of heavy liquid propellant for thrust, they use solar power to ionise a small supply of xenon gas. A high voltage applied across a pair of gridded electrodes sends the positively charged ions rushing at high speed towards the negative electrode. Most ions pass through the grid, generating thrust.
However, some ions collide with the grid itself, causing it to gradually wear out, says Cody Farnell, a space flight engineer at Colorado State University in Fort Collins. Simulations suggest grids in a typical NASA engine will last 2.8 years - but Farnell wondered whether changing the grid's design could extend its lifespan.
He used evolution-mimicking software called a genetic algorithm (GA), and started by instructing the algorithm to randomly generate values corresponding to the geometry of the grid and the voltages applied to it. These values can be thought of as analogous to genes.
Each combination of values was then fed into a simulator to give an idea of the grid's performance and its expected lifetime. If the performance was promising, the "genetic material" was subjected to further random changes, or mutation, and this process was repeated until no more improvements were forthcoming.
After 100 generations, the GA spawned a geometry/voltage set that boosted the ion engine grid's lifetime to 5.1 years - at least in the simulator (Journal of Propulsion and Power, DOI: 10.2514/1.44358). Factors optimised included grid hole diameter, hole spacing and the thickness of the grids. The engine could be improved further, says Farnell, by evolving the other parts too.
"After 100 generations, the algorithm spawned a design that almost doubled engine lifetime"
Propulsion engineer Steven Gabriel at the University of Southampton in the UK is developing a triple-gridded ion engine. He says Farnell's work could have major implications if real-world tests of his design match the simulations.
Genetic algorithms are quite common in optimization - they take a lot longer than gradient-based methods, but are less likely to get hung up on local minima rather than finding global ones. I don't know why they're making such a big deal of it.
And no, you couldn't do this with Polywell. To do optimization you need a physics model that adequately represents the system and is computationally inexpensive, and to my knowledge no such model of the Polywell is possible with the current level of understanding.
And no, you couldn't do this with Polywell. To do optimization you need a physics model that adequately represents the system and is computationally inexpensive, and to my knowledge no such model of the Polywell is possible with the current level of understanding.
thanks for the answers. Cant you just imput the basic laws of physics and let the a quantum computer run all the simulations?93143 wrote:Genetic algorithms are quite common in optimization - they take a lot longer than gradient-based methods, but are less likely to get hung up on local minima rather than finding global ones. I don't know why they're making such a big deal of it.
And no, you couldn't do this with Polywell. To do optimization you need a physics model that adequately represents the system and is computationally inexpensive, and to my knowledge no such model of the Polywell is possible with the current level of understanding.
(if we had working quantum computers, of course )
-
- Posts: 149
- Joined: Fri Mar 28, 2008 3:51 pm
and a fully functional set of physical laws. We've got some good first order approximations at the present.
I remember one case GA's were used was coming up with a better antennae and the computer popped out what looked to be a 5 years old squiggly line. But it was a very explicit squiggly line.
I remember one case GA's were used was coming up with a better antennae and the computer popped out what looked to be a 5 years old squiggly line. But it was a very explicit squiggly line.
The pursuit of knowledge is in the best of interest of all mankind.
Ah, but how well did that squiggle work as an antenna? That would have been the deciding factor.Professor Science wrote:I remember one case GA's were used was coming up with a better antennae and the computer popped out what looked to be a 5 years old squiggly line. But it was a very explicit squiggly line.
I remembered something similar from nasa, a quick google and I recovered it:Professor Science wrote:and a fully functional set of physical laws. We've got some good first order approximations at the present.
I remember one case GA's were used was coming up with a better antennae and the computer popped out what looked to be a 5 years old squiggly line. But it was a very explicit squiggly line.
http://www.nasa.gov/centers/ames/news/r ... _55AR.html
Is it the same you are talking about?
Edited:
Here is a better article from another website from google cache:
http://webcache.googleusercontent.com/s ... =firefox-a
Agreed, on both counts. A second-generation PW might benefit from such an approach, which I've also seen used elsewhere.93143 wrote:Genetic algorithms are quite common in optimization - they take a lot longer than gradient-based methods, but are less likely to get hung up on local minima rather than finding global ones. I don't know why they're making such a big deal of it.
And no, you couldn't do this with Polywell. To do optimization you need a physics model that adequately represents the system and is computationally inexpensive, and to my knowledge no such model of the Polywell is possible with the current level of understanding.
They're making a big deal of it because reporters are morons.
n*kBolt*Te = B**2/(2*mu0) and B^.25 loss scaling? Or not so much? Hopefully we'll know soon...
-
- Posts: 149
- Joined: Fri Mar 28, 2008 3:51 pm