We are working with a 3D-printed TPMS lattice structure (fig b) that was tested under compression (we have 4 scans at different loading). I am currently segmenting the sample so that we can compare the deformation of individual unit cells (fig a) before and after loading.
For the unloaded sample, I can separate the unit cells (fig a) using a regular grid pattern because the geometry is still uniform. However, after compression, the unit cells are deformed and no longer align well with the original grid, so I am having difficulty separating each individual unit cell consistently.
Has anyone worked on a similar problem? Is there a recommended workflow for tracking or separating individual unit cells in a defor
I have not worked with this type of repeating structure before, but if the deformations are less than one repeating unit, I would expect that non-linear registration (using ANTs or Elastix) would probably work well, extracting a deformation transformation between the undeformed and deformed versions. The process MIGHT work well even with larger deformations, but I could also see the process falling into bad local minima which spoil the optimization which drives the registration. If it works, you can define a set of points on your undeformed structure, and then use the deformation transformation on a copy of the points to see where each goes. You can then follow the unit cells by following the points associated with each cell.
If the deformations are large and/or nonlinear registration does not work for you, you might be able to consistently get the same points by applying a 3D gaussian with a sigma based on your undeformed lattice spacing, smoothing your images and finding local maxima. If the lattice isn’t ever fully crushed, I bet you could make this work. You’d have to figure out a reliable way to identify pair up the corresponding points, but maybe this will seem easier than finding the whole image matching deformation.
The easiest approach probably depends on the range of deformed results you see in your experiments. If you’re fully crushing these sometimes, that’s a much harder problem than if they just squish a little bit.
