- Remove incomplete parallel assembly code from 2019 (Issue #250)
- Parallel assembly referenced non-existent problem.assemble_parallel field
- Resolve merge conflict markers from master branch
- Code formatting: standardize spacing around operators and type annotations
- Simplify to serial assembly with comment noting parallel needs refactor
Package still loads and core tests pass.
Sometimes user may have the wrong kind of elements in element set when
assembling 3d continuum problem. This could happen for example in
situations, where mesher is giving also segment elements. They may have
some use in certain situations, but currently we don't support them.
When assembly is failing for this reasons, we give a meaningful error
message:
[ Info: It looks that you are trying to assemble elements of type Seg3
to 3d continuum problem. However, they are not supported yet. To filter
out elements from a element set, try `filter(element->!isa(element,
Element{Seg3}), elements)`
ERROR: LoadError: Tried to assemble unsupported elements of type Seg3 to
3d continuum problem.
This commit closes issue #211.
Improve performance in elasticity by moving out the computation of X from the integration point loop.
Before:
```
12.576967 seconds (37.65 M allocations: 1.852 GiB, 11.65% gc time)
```
After:
```
11.251927 seconds (35.27 M allocations: 1.692 GiB, 11.27% gc time)
```
2 million allocations less.
Store element connectivity as "Mixed" format to Xdmf file. It's a long
list where first number is element code and then there is connectivity
data (starting from 0). Xdmf format does not support several
Topology-elements in Xml file format, so before it was not possible to
store e.g. both Tri3 and Quad4 elements in same problem. After this
commit this should now be possible. Paraview Xdmf3 reader is needed to
use.
Use FEMBeam.jl to solve beam problems. Added an example, where
natural frequencies of frequencies of 3d frame structure are
calculated. Some minor modifications to Modal analysis is done to make
Xdmf writing of 6 dof nodes work.
Development of auto-differentiated mortar contact mechanics in 2D is moved to own separate package, MortarContact2DAD. Other changes are similar to what is done with MortarContact2D: elements are added to problems using `add_slave_elements!` and `add_master_elements!` instead of `add_elements!`, to make interface more explicit. Also, problem name is `Contact2DAD`, so the dimension is now explicitly stated in problem name. (Also have `Mortar2DAD`, compare to the `Mortar2D` and `Contact2D` of `MortarContact2D.jl`.)
Moved plane contact related stuff to own separate package
`MortarContact2D.jl`, where the development continues.
The following changes to test files are done:
1) Problem name for plane mortar coupling is `Mortar2D` (was `Mortar`
before), and later on 3d coupling will be `Mortar`. So the dimension
of coupling operator is explicitly given in a problem name.
2) Before elements to coupling was defined using
```julia
update!(problem.elements, "master elements", master_elements)
add_elements!(problem, [slave_elements; master_elements])
```
Now, explicitly give master and slave elements as
```julia
add_slave_elements!(problem, slave_elements)
add_master_elements!(problem, master_elements)
```
Keep on mind that Lagrange multipliers are in slave side.
Heat transfer analysis is moved to its own package where the development continues. Two small modifications are needed for test files:
- Instead of `problem.properties.formulation`, we have two separate problems, `PlaneHeat` for two-dimensional problems and `Heat` for three-dimensional problems.
- Unnecessary prefixing of field names is changed. For example, now we simply have only "thermal conductivity" and not prefixed "temperature thermal conductivity".
Core functionality is moved to base package called FEMBase.jl. The
aim is that when developing new elements, solvers, materials and so on,
user only imports FEMBase.jl and uses the functionality there.
JuliaFEM.jl is a sort of "metapackage" collecting together all the
packages and features can be programmed in smaller packages focusing
only on one thing. This structure makes it attractive to contribute
smaller amount of code e.g. in the form of thesis. Moreover, FEMBase.jl
is under 2000 lines of code, which will be very clearly documented thus
everyone can understand the basic concepts behing JuliaFEM easily.
Mass matrix can be analytically solved if Tet10 metric is constant, i.e.
the midnodes are in the midpoint of corner nodes. This should increase
assembling speed of mass matrix.
Slicing is very expensive operation on sparse matrices. Avoid slicing
by calculating matrix product `K_red = C1*K*C1`, where C1 is diagonal
matrix. This should be much cheaper.
Now style is to assemble all same kind/dimension elements at same time
in one function call, so it's possible to allocate all necessary
matrices only one time.
Jukka Aho