[HN Gopher] Inverting the structure-property map of truss metama...
       ___________________________________________________________________
        
       Inverting the structure-property map of truss metamaterials by deep
       learning
        
       Author : bryanrasmussen
       Score  : 22 points
       Date   : 2022-01-19 12:20 UTC (2 days ago)
        
 (HTM) web link (www.pnas.org)
 (TXT) w3m dump (www.pnas.org)
        
       | Isamu wrote:
       | Definition: A metamaterial is any material engineered to have a
       | property that is not found in naturally occurring materials.
       | 
       | FTA: for applications from wave guiding to artificial bone.
        
         | kragen wrote:
         | That Wikipedia definition is not a good definition because it
         | would include things like pure hydrazine, high-speed steel, and
         | cubic boron nitride as metamaterials, which they are not.
         | 
         | A better definition is that a metamaterial is a material whose
         | properties come from a highly ordered structure rather than
         | from its composition. Negative refractive indices, negative
         | Poisson's ratio, solids with lower densities than aerogels,
         | stuff like that. Usually things like foams, fiber-matrix
         | composites, and gold nanoparticle pigments are excluded because
         | they're too disordered, and things like wood, limpet teeth,
         | nacre, and bone are often excluded because they're natural,
         | though sometimes they _are_ included.
         | 
         | In this case the particular properties they're trying to
         | achieve _are_ found in natural materials, just not the ones
         | they 're making the trusses out of.
        
           | smaddox wrote:
           | A usefull word for defining metameterials, at least in the
           | field of electromagnetic metamaterials is "mesoscopic".
           | Perhaps we need a generalization of this term to arbitrary
           | length scales.
           | 
           | Electromagnetic metameterials are composed of nanometer-scale
           | structures that alter the angstrom-scale (i.e. atom-scale)
           | behavior of pure materials. These are often described as
           | mesoscopic structures---structures that span the length
           | scales between atoms and traditional fabrication scales.
           | 
           | Some great examples from nature include Beatle carapaces and
           | the pads of gecko feet.
        
           | wefarrell wrote:
           | Would graphene nanotubes be considered a metamaterial?
        
             | smaddox wrote:
             | Isolated carbon nanotubes, probably not. But certain
             | arrangements of a large number of carbon nanotubes, almost
             | certainly. Vantablack should almost certainly be described
             | as a metameterial.
        
       | wantsanagent wrote:
       | TLDR:
       | 
       | Imagine you have a bridge with well known stress characteristics,
       | loads, etc. Now you want to run an optimizer to produce a
       | lightweight bridge made out of a material that uses repeating
       | structures throughout. This exists today but takes a long time to
       | run, you can get all sorts of interesting material compositions
       | out of such programs.
       | 
       | Instead of doing the above train a deep neural network to predict
       | from a set of known stress properties of shapes made out of
       | different types of materials to a set of the structures of those
       | materials. (Massive over simplification: should I use pyramids or
       | cubes as my basic building block for a material?)
       | 
       | After training on millions of known examples, the network is able
       | to take in a desired set of stress characteristics and produce
       | the material which will have those characteristics, very quickly.
       | 
       | They seem to be particularly interested in predicting materials
       | to make up bone replacements.
        
       | ur-whale wrote:
       | The should be a scale of 1 to 10 for how obscure a HN post title
       | is (probably automatically computed by some deep NN).
       | 
       | This one would likely take the cake, especially if you consider
       | that actually reading the article itself does not help very much.
        
         | atoav wrote:
         | Really? The paper was more or less precisely what I expected
         | and I don't even have a ML (or engineering) background.
        
       ___________________________________________________________________
       (page generated 2022-01-21 23:01 UTC)