# Is Model order reduction good for complex models with a lot of contact points in them?

we know that MOR ( or model order reduction ) is used to reduce the size of heavy FEM models. but if there are so many faces, edges of the part that are in contact with another part in the model or within itself, is it usefull or wise to use MOR ?

We used to go for sections ie finding geometric similarity to reduce the model or to have the same number of cells more concentrated into a section for more detail. Also we would concentrate cells where we thought the interesting bits were likely to happen. Another method was to run a very coarse model to locate areas of interest...

Relying totally on a "button" to reduce the model appropriately needs to be done with care or blind confidence...

• Is the whole thing ,tracing the nodes or elements role and location and zeroing their value in the matrix ? Commented Aug 28, 2019 at 13:10
• @FabioSpaghetti your comment has me lost... Commented Aug 28, 2019 at 13:18
• I was asking how is the procedure of matrix modification Commented Aug 28, 2019 at 13:21
• We used to design our computational field (mesh, grid) as I described above. Is that what you mean by matrix, since it is a term not used in your question. Commented Aug 28, 2019 at 13:23
• @FabioSpaghetti Asking about the numerical methods for MOR in general is far too broad. See something like amazon.co.uk/Reduction-Techniques-Applications-Element-Analysis/… if you really want to know. Commented Aug 28, 2019 at 20:43

If parts of the model are fixed in contact with each other, that does not have any particular influence on MOR.

On the other hand if the contact is not fixed (e.g. sliding surfaces in dynamics simulations) or there are nonlinear effects at the interface between parts of the model you need to be careful that the reduced model has the same physical behaviour as the unreduced model.

Even in that situation, you can often make large reductions in the parts of the model that are linear. For example we routinely reduce FE models with about 300,000 degrees of freedom down to about 3,000 degrees of freedom and then do nonlinear analysis on the reduced model, and the validity of the reduced model results has been verified by experiments on the real world structures.

• thank you very much, so you mean in the linear parts I reduce the middle and interpolate between the remnants ? Commented Aug 29, 2019 at 6:24