Debugging LoopStructural#

Interpolation#

Surfaces are not displaying#

  • check to see what the scalar field looks linked

  • look at the values of the nodes of the implicit function supports feature.interpolator.c if all of the nodes are 0 the interpolator has not solved.

  • Does the scalar field show different colours? view.add_scalar_field(feature)

  • If the scalar field shows the right geometry, are the surface values correct?

Surface geometry is erroneous#

  • Try using a different solver - sometimes pyamg does not work as it tries to pick a pattern in the interpolation matrix indicating the mesh geometry.

  • If the solution doesn’t fit your data points try adding more elements

  • Try using only orientation constraints

  • Try using only value constraints

  • Change gradient norm constraints to gradient direction constraints (rename nx,ny,nz to gx,gy,gz)

  • Try removing some of the data points

  • Try changing to using interface constraints instead of value, to do this rename column val to code:interface but only use FDI.

Alternatively use surfe with method=interface. Using either of these methods the value of the scalar field will change. - The tolerance of the conjugate gradient solver is scaled to the size of the model assuming that the scalar field is interpolated to have a magnitude of the gradient norm to be close to 1. The tolerance is saved in feature.builder.build_arguments[‘tol’] but can be overwritten using the argument in code:create_and_add tol=val. Smaller values mean the solution will be a closer fit to the data + regularisation. - Try changing the relative weighting of the data points and regularisation.

  • regularisation - changes the weighting of the constant gradient or the finite difference second derivative. 1.0 is equivalent to 0.1 constant gradient.

  • npw=1.0 changes the weighting of the gradient norm constraints

  • gpw=1.0 changes the weighting of gradient direction constraints

  • cpw=1.0 changes the weighting of the control points.

  • A good strategy for debugging a model is to lower the weights of the data points until a over smoothed solution is found. Then increase the weights until the model starts to overfit.

Interpolator hasn’t solved#

  • check that there are sufficient data points to constrain a surface. There needs to be at least two unique value constraints or a single value constraint and a gradient norm constraint.

  • Try using the default solver, solver="cg"

Interpolator is taking a long time#

  • Check how many elements you are using, 1e5 should take under 5 minutes on a modern laptop

  • How many faults are associated with the feature? len(feature.faults), you can remove the faults by setting this to an empty list

Folds#

When using the create_and_add_folded_foliation or create_and_add_folded_fold_frame LoopStructural the quality of the results can be dependent on the correct parameterisation.

S-Plot doesn’t fit the observations#

  • Is the fold axis correct? If you expect a constant fold axis check that the calculated fold axis is consistent with your interpretation.

  • Is the wavelength obvious in the s-variogram? If it is, check the guessed wavelength is reasonable Check the s

  • Try specifying the wavelength using limb_wl or axis_wl

  • If the data points in the S-Plot do not look correct, check the data points in the model by plotting the axial foliation and the data. The fold geometry

should be visible. - If the fold is non-cylindrical, try building a cylindrical fold by removing some data - Check the polarity of the datapoints, the fold modelling requires the polarity of the vectors is constant. Check uising vector plots view.add_data(folded_foliation,vectors=True)

Faults#

No displacement#

  • Check that the displacement is large enough that it should be visible. E.g. compare displacement to model bounding box

  • Has the fault frame interpolated correctly? Coordinates 0 and 1 should both have values.

  • Increase the resolution of the visualisation mesh view.nelements=1e6

  • Is the fault parallel to the feature being faulted?

  • Has the fault been added to the feature being faulted?

  • Is the fault displacement vector correct? Add the vector field to the visualisation

On this page
Show Source