Re: continuous looping in fkern

["David Meeker" <dcm3c@xxxxxxxxxxxxx>]

--- In femm@xxxx, "shazanaz2000" <smartmicro@xxxx> wrote:
> hi,
> 
> I'm doing some small fea analysis. my pc is a celeron500 with 64mb 
> memory. My problem is not more than 100mm x 100mm planar. mesh size 
> varies from 0.1mm in some blocks to 5mm in air. total node count is 
> around 18k to 30k (depending on mesh sizes).
> 
> The problem I face is:
> sometimes..., fkern loops for an inordinately long time without 
> upgrading the newton iteration counter while continuously looping 
> in "matrix construction" and "conjugate solver". If I stop fkern and 
> adjust my file (I move any one or two node points or change the mesh 
> size -slightly- at any one place) the next time fkern completes. this 
> I've noticed in not one problem but in many. Am I doing something 
> wrong? is my computer slow? I've even seen the NI count decreasing!!
> Is that supposed to happen?

There can sometimes be convergence problems in nonlinear problems if
you are using a material with a strong "initial permeability" region.
The most expedient way of fixing things is to zap out the first few
points of your B-H curve, eliminating the initial permeability region.
I'm looking at ways of trying to get the solver to recognize and
remedy this when it happens automatically. At the very least, I'll put
a user-specifiably relaxation factor for the Newton iteration in the
next significant revision of the program so that materials with a
significant initial permeability can be forced to converge.

Dave.