Skip to content

Latest commit

 

History

History

Supports

Supports using coupling

Tested with CGX 2.19 / CCX 2.19

  • Test of distributing and kinematic coupling
  • Modal analysis
  • Large rotations
File Contents
pre.fbl CGX script, pre-processing
shapes.fbd CGX script, post-processing, animation of mode shapes
anim.fbl CGX script, time history animation
post.fbl CGX script, images and surface data plots
trfix.fbl CGX script, images and surface data plots
solve.inp CCX input for frequency analysis
trans.inp CCX input for static analysis
trfix.inp CCX input for the reference solution
surface.gpl Gnuplot script for section data
test.py Python script to run the full simulation

The model contains a brick shaped beam with dimensions given by parameters in pre.fbl.

Parameters Value Description
lx 100 Length
ly 5 Thickness
lz 10 Width

Three simulations are set up:

  • Frequency analysis with one end coupled and subjected to various support conditions (as of 2.19, kinematic coupling is used).
  • Multistep static analysis with large rotation. Cantilever beam with the load distributed over the end surface by distributing coupling. Support is applied via kinematic coupling.
  • Reference solution, cantilever beam with nodal constraints at the end surface.

Pre-Processing

The script generates the elements, the surface sets and the reference nodes.

> cgx -b pre.fbl

Frequency analysis

The simulation consists of two frequency steps with different constraints of the reference node:

  1. Free (no constraints)
  2. Clamped (all dofs constrained)

For each step, 10 mode shapes are stored.

> ccx solve

Post-Processing

> cgx -b shapes.fbd

Free (unconstrained reference node).

Clamped (all dofs of the ref node constrained). This entirely clamps the support nodes without allowing for deformation.

A test with meanrot MPC as replacement for distributing coupling in 2.19 would be sensible.S

Multistep static analysis

The simulation consists of a sequence of four steps. The force to the end surface acts always downwards (-z)

  • horizontal position along x, bending about the strong axis
  • 90° rotation about z (no changes in the stress state)
  • 90° rotation about y (longitudinal axis, now bending about the weak axis)
  • rotation to align the beam with the z axis.

The clamp condition is applied as kinematic coupling and the rotations are applied as constraint to the reference node. A load of 1000 N in negative z-direction is applied via distributing coupling (translation only) to the free end surface.

The rotations are specified by the rotation vector with respect to the initial state. The intermediate states are some non-linear blend between the end states. Here, amplitude generation based on specification of the angular velocity might help.

Observations:

  • The stress state at the support is consistent with full rigid constraint of the nodes, see the reference solution further down.
  • If you switch the coupling dofs at the load application side from 1...3 to 1...6, the simulation still converges but at a much slower rate.
ccx trans
monitor.py trans

cgx -b anim.fbl

cgx -b post.fbl

The stress distribution at the support (rigid body constraint by kinematic coupling) is essentially identical to the reference solution (using direct nodal constraints)

Using the *section print command, the surface results (reactions, geometry) are written.

Reference solution

For comparison a standard clamped cantilever solution is provided.

  • left end is clamped using nodal constraints
  • right end is subjected to a distributed load (distributing coupling) of 1000 N in various directions (x, y, z)
ccx trfix
cgx -b trfix.fbl

Forces and moments at the support