Accuracy Issues with Single-Pass Adaptive Convergence...

Most of the time you will get better and faster solutions with the SPA convergence than with MPA. This is especially true if you are interested in the maximum stress in the model. That is because the error estimation process is stress based.

Models run using SPA convergence are sensitive to the same issues as MPA solutions. These include:

> High Stress Gradients
> Element Shape Quality
> Re-entrant Corners
Point or Edge Loads and Constraints


If you suspect a high stress gradient problem or a poorly shaped element problem you will need to use the method of mesh refinement talked about and re-run the solution to check for any change in the result. Usually one or two iterations are all that will be required.



There is a unique problem using SPA with models that have re-entrant corners, point loads, edge loads or any constraints. That problem is that the SPA algorithm recognizes these as possible singularities and excludes any elements from the error estimation that touch these singularities. In some models this can be extremely significant. Most effected will be the overall stiffness of the structure. You can detect the problem first by looking at the summary report. If the number of equations between pass 1 and pass 2 does not increase by much (approx. < 20%) then there may be a significant number of elements being ignored during the error estimation. Another clue is to look at the P-level plots. If most of the model is at P-level 3 then there may also be a significant number of elements being excluded. The solution to this problem is to use MPA, add fillets or change the load and constraint method.



The SPA algorithm is driven by the maximum stress in the model. Generally it does a great job calculating these maximum stresses, however, if there are only very few areas of high stress with large stress gradients, other areas of the model may not provide as accurate of results. It has been found that stresses and deflections in lower stress areas of the model may be as much as 30% off. You may recognize these potential problems if your stress plots have only one or two very localized high stress gradient areas and everything else is at a much lower stress level. The solution to this problem is to run it using MPA.


<TSD Engineering HOME>

Copyright 2003, TSD Engineering