Absolute or Relative. Which is best?

Last week I told you how to check your global mesh refinement by performing a number of simulations with a denser mesh and looking for convergence of your results. Have you given it a go yet?

This week we held a 2 day simulation training session covering the topics of motion and non-linear studies for our resellers and one of the surprising things for a lot of people was the sensitivity of the results to various parameters. The motion studies were highly sensitive to the contact settings and a non-linear buckling analysis we ran was very dependent on the mesh size chosen.

When discussing the implications of this it became clear to everyone that building a simulation, solving it and immediately believing the result can be a dangerous pastime. We call this ‘one-shot’ approach an Absolute Analysis, giving one absolute answer.

But what would you say if I asked how certain you are of your input loads? Could they be, say 5-10% out? Also, have you modelled all the boundary conditions accurately or used an approximation, perhaps fixing a face when in reality it is bolted to a plate? Have you chosen the true material data or an approximation from the database simply because it is there?

To mitigate these cumulative errors it is better to perform a Relative Analysis, a series of simulations where you change each parameter to see how it impacts the result.

If you are not sure that your loading is exactly 500N, try one at 450N and another at 550N. Is the direction of the loading really normal to the face of application? Perhaps you could check the result by running a couple of simulations with the load offset at 5 degrees each way to see how the induced moment affects your design. Using a bolt connection and a no penetration contact will give you a different result to a fixed restraint so try it to see just how different the results are. And finally perform a few studies with slightly varying material properties to see what impact this has.

The power of this approach is similar to the convergence check, you are making sure that your model is robust, i.e. not sensitive to variation, and if it is, you can begin to determine just how sensitive. This approach is also one way of building a real understanding of your product so you can appreciate how it behaves in the non-ideal world we actually live in.