2D flux field visualization

[termi-official]

using FerriteViz
import GLMakie

include("docs/src/ferrite-examples/heat-equation.jl");
dh,u=manufactured_heat_problem(Triangle, Lagrange{2,RefTetrahedron,2}(), 21)
(dh_grad, u_grad) = FerriteViz.interpolate_gradient_field(dh, u, :u);

FerriteViz.surface(dh_grad, u_grad)

Stepped up the game and checked analytically the gradient field. It converges. :)

[koehlerson]

nice, I think it would be nice to support vector valued fields in 2D for this.

Can you elaborate on https://github.com/Ferrite-FEM/FerriteViz.jl/blob/master/src/utils.jl#L292-L297 ?

[termi-official]

The problem mentioned in https://github.com/Ferrite-FEM/FerriteViz.jl/blob/master/src/utils.jl#L292-L297 is that for volumetric entities we need an interpolation on entities with dimension 2 (for the most common problems). Now, L2 spaces have support on the local portion of a face, but not on the actual shared face. However, the current implementation just extract the faces from the provided interpolation to build the face triangles - which is an empty set for L2 spaces.

[termi-official]

Apart from test coverage for interpolate_gradient_field this one should be fine.

Edit: 3D is future work when clip planes are implemented.

[koehlerson]

Edit: 3D is future work when clip planes are implemented.

How does the docs example with 3D plasticity work then? :D

[termi-official]

Derp.

[koehlerson]

take the incompressible elasticity, there we can visualize both, stresses and strains, or alternatively heat example with the heat flux (so negative gradient with some constant) to show off how to get from gradient to flux

[koehlerson]

few suggestions

[termi-official]

Okay, I think now only the next Ferrite release is missing (because of the L2 fix you pushed @koehlerson ).

[koehlerson]

Okay, I think now only the next Ferrite release is missing (because of the L2 fix you pushed @koehlerson ).

builds locally with linear ansatz for the displacements 👍

[koehlerson]

we can also think about having Ferrite#master as docs dependency

[termi-official]

I would vote for sticking with the safe route, manually upgrading at each release to capture regressions.

[termi-official]

Gradient field converges for the heat problem towards the analytical gradient in 2d

julia> maximum(u_grad) - π/2
0.0072785284134289086

and 3d

julia> maximum(u_grad) - π/2
0.01160556577741545

[koehlerson]

I think we could add a Union in the field argument such that either one gradient or multiple ones are computed. We should also provide the option to add a deformationfield to the dh_grad, the latter will be annoying for setting up the total DoF vector for dh_grad

[termi-official]

Before we start introducing more and more hotfixes, I think we should focus on reorganizing the code.

[termi-official]

I will skip the arrows plot in favor of #55 and #54 . Should be good to go (minus getting Ferrite master into test CI).

[koehlerson]

Otherwise LGTM

[termi-official]

Correctness of shear component computation has been roughly tested via some crappy simple shear

include("docs/src/ferrite-examples/incompressible-elasticity.jl")

function create_cook_grid(nx, ny)
    grid = generate_grid(Triangle, (nx, ny));
    # facesets for boundary conditions
    addfaceset!(grid, "traction", x -> x[1] ≈ 2.0);
    return grid;
end;

function create_bc(dh)
    dbc = ConstraintHandler(dh)
    add!(dbc, Dirichlet(:u, getfaceset(dh.grid, "left"), (x,t) -> zero(Tensors.Vec{2}), [1,2]))
    add!(dbc, Dirichlet(:u, getfaceset(dh.grid, "right"), (x,t) -> Tensors.Vec{2}((0.0, 0.1)), [1,2]))
    close!(dbc)
    t = 0.0
    update!(dbc, t)
    return dbc
end;

u_quadratic,dh_quadratic = solve(quadratic, linear, mp);

(dh_quadratic_grad, u_quadratic_grad) = FerriteViz.interpolate_gradient_field(dh_quadratic, u_quadratic, :u)
plotter_quadratic = FerriteViz.MakiePlotter(dh_quadratic_grad, u_quadratic_grad)
cmap = :jet

f = GLMakie.Figure()
axs = [GLMakie.Axis(f[1, 1], title="u_{x}"),GLMakie.Axis(f[1, 2], title="u_{y}"),
       GLMakie.Axis(f[3, 1], title="u_{x,y}"),GLMakie.Axis(f[3, 2], title="u_{y,x}")]
p1 = FerriteViz.solutionplot!(axs[1], dh_quadratic, u_quadratic, process=u->u[1], colormap=cmap)
p2 = FerriteViz.solutionplot!(axs[2], dh_quadratic, u_quadratic, process=u->u[2], colormap=cmap)
f[2,1] = GLMakie.Colorbar(f[1,1], p1, vertical=false)
f[2,2] = GLMakie.Colorbar(f[1,2], p2, vertical=false)

p4 = FerriteViz.solutionplot!(axs[3], plotter_quadratic, process=∇u->∇u[2], colormap=cmap, colorrange = (-1e-2, 1e-2))
p5 = FerriteViz.solutionplot!(axs[4], plotter_quadratic, process=∇u->∇u[3], colormap=cmap, colorrange = (-0e-2, 6e-2))
f[4,1] = GLMakie.Colorbar(f[3,1], p4, vertical=false)
f[4,2] = GLMakie.Colorbar(f[3,2], p5, vertical=false)

[termi-official]

CI failed because it turns out that Tensors.jl gradients are the outermost indices, but access to a Vec (case: scalar problem) with two indices does not work. :) Should be fine now.