Strand7:  Hints & Tips

Model checking and quality control
With the increasing use of finite element analysis techniques in more and more sophisticated structures, where the results may not be intuitive, it is important that we as FE users develop checking and quality control procedures and apply these to the analysis tasks that we are undertaking.

A well thought out set of model quality control checks should be capable of detecting most major modelling errors early in the analysis, before expensive decisions have been made or parts are starting to be manufactured. Following are some guidelines to the sorts of checks that should be carried out on all models:

Model Data
Before running any model, especially large ones, the input data should be checked and any modelling assumptions reviewed. This is best done together with another knowledgable FE user. The following are some suggestions for simple checks that can be carried out:
  • Check the units of the geometry, the material properties, loadings etc to ensure that they are consistent.

  • Check all material properties to ensure that the magnitudes of the moduli, thicknesses etc are correct.

  • Check the boundary conditions to ensure that they are representative of the physical support conditions and that they are sufficient to restrict the rigid body motion of the structure.

  • Check that adjacent elements are consistent i.e. that each edge of an element is connected to a full edge of other elements and that corner nodes of elements are only connected to corner nodes of other elements, not midside nodes. This is especially important in regions of the mesh that have been locally refined.

  • Check that the number of degrees of freedom on elements connected to one another is compatible. For example you should not connect a beam normal to a Quad4 plate as the plate cannot really accept rotation its local z axis. Of course this rule can be broken by the experienced analyst if adequate precautions are taken. Carefully check the mesh to ensure that all connected nodes are properly zipped.

  • Check that all the plate elements have their local z axis and preferably their local x-y axes pointing in the same direction. The z normal axis is particularly important. As an example consider a model of a pressure vessel. If some of the elements have their normal pointing outward and others pointing inward and we plot a contour of say the +z surface stresses, we will get a picture showing the stresses on the outer surface of part of the structure and the inner surface of the rest. If some of the elements have the incorrect orientation it is a simple matter to correct this using the flip tools in Strand7. It is also easy to check the orientations by using the "Orientation" display mode.

  • Similarly check the orientation of beam elements. This is best done by displaying the section of the beam and/or then beam axes or the Reference Node (if you are using a Beam3).
Solution Log File
Before looking at the results of a solution the log file generated by the solver should be examined in its entirety (e.g. SLG file in linear static runs). The sorts of things that we need to look for are as follows:
  • Check that the singularities (suppressed drilling freedoms) listed are consistent with the types of elements in the model and the boundary conditions applied.

  • Check the summation of the forces produced by the solver, both on the elements and in the total load vector. Does this equal the applied loads? Note that there may be some minor difference between the applied loads and the summation reported in the total load vector if some loads are applied to nodes that have fixed boundary conditions; load applied to these nodes is not included in the summation.

  • Check for warning messages. Ensure that you understand all warning messages and the implications of these on the results. Typical warnings are for warped elements, elements with acute internal angles etc. To gain a better understanding of the warnings, see the extra information on the solver warnings in the Strand7 Online Help Files.
Results
It is good practice to question or even distrust the results from any FE model until you are sure that they make sense. Checking results is as much experience as a set of rigid rules but the following simple guidelines should give you an idea of the areas in which to start:
  • Check that the sum of the reactions equals the applied loads.

  • Check the displacements at the supports or symmetry conditions to ensure that the boundary conditions have worked correctly (i.e. that the displacements are zero in the correct directions).

  • At a few locations where the stresses can be calculated using simple equations, do a 'back of the envelope calculation' to verify the output stresses. Again using the example of a pressure vessel, the stresses can be calculated approximately from the classic PR/2t type calculations for hoop and axial stress. Ensure that the stresses are within a few percent of these. Obviously these simple equations cannot calculate the stresses at the ends of the vessel where the stress state is more complex due to bending etc, but this is the reason we use finite elements.

  • Similarly check the magnitude of the displacements at some point where the results are easily checked by a simple calculation. Just because the stresses are correct does not necessarily guarantee that the deflections are also correct. If the modulus is out by a factor of 1000, say, then the deflections would be wrong but in a linear static analysis the stresses would be correct.

  • Check the deflected shape of the structure and ensure that this is consistent with the stiffness and geometry of the structure, the loads and boundary conditions.

It would be considered good practice, and in fact may be required under some quality systems, to formalise these checks and create a check list with provision for indicating that a particular model has been checked.
If you are new to finite element analysis or experienced but conducting a new type of analysis with which you are not familiar, then there are some additional checks that you should be doing. These involve the running of simple benchmark tests. The idea is to make several small models that are representative, in a very simplistic way, of the real structure you are interested in. The answers for these models should be readily calculated from elementary mechanics or should be published experimental test data. The idea here is to hone your modelling methods, to ensure that you know what the required input data means, that your results are correct and that you know how to interpret them. Even experienced FE analysts employ simple models of this type if they are not absolutely confident with the techniques to be used in a particular application.