Issues to solvers, multi core, stress parameter
I have problems with using the arc-length method in a non-linear analysis in Z88 Aurora.
I'm a relatively newbie to Z88 and FEA in general.
I am modelling axisymmetric concentric corrugated suspensions similar to be found in speaker drivers.
I need to evaluate the reaction force vs displacement curve.
This is an axisymmetric non-linear mechanical problem, which was migrated from CalculiX because
the spooles solver couldn't solve it (for small thickness, in comparison to the overall dimensions).
The cause seems to be snap-through buckling at specific load times.
I have converted the geometry for Z88 Aurora, and z88nl was able to solve it with element type 8, PARDISO
and the Newton-Raphson method.
In most cases (where the thickness is small e.g. 0.25mm) a large number of increments was needed, but
the NR method managed to solve it.
What bothers me is that I can't get the arc-length method to work (which is designed for these problems AFAIK)
The method either doesn't reach the 100% load time, or it gets into some type of loop and eventually runs out of
I have tried to adjust the solver parameters (NLAERH, BGLAENG, AUTOGAUSS) and there are combinations where
the solver reaches 100% of course, but in many cases it doesn't (if I adjust NLAERH, the arc length will be
readjusted, and it still doesn't reach 100%).
I need an automated solver what always reaches the 100% load, adjusting it is not allowable during the
evaluation, but I consider the loop problem more serious, the solver circulates between certain load times and runs
out of memory.
First, is there any law that tells me how to set up NLAERH and BGLAENG properly? To be sure that the
solver always reaches 100%.
Next, is the circulating phoneme a bug or a drawback of the method?
How can I avoid it?
I am attaching the Z88 Aurora files, for a coarse and for a finer geometry.
I was able to solve the coarse geometry both with the NR and the arc-length method, but the
finer geometry can't be solved by the arc-length method.
I've attached a few screenshots as well.
Any help would be appreciated
Z88 Aurora version 3 (Apr 1 2015)
Ubuntu 15.10, wily
Intel(R) Core(TM)2 Duo CPU E4600 @ 2.40GHz 4Gb RAM
Advanced Micro Devices, Inc. [AMD/ATI] Turks XT [Radeon HD 6670/7670]
thanks for your message. I will look into your project and will give you feedback as soon as possible.
I've noticed that the project used a material that doesn't exist in the database by default.
I have set it to polypropylen and uploaded it again.
The coarse and the fine geometries aren't the same, displacement is 32mm for both cases.
- (656.66 KiB) 122-mal heruntergeladen
I tested the coarse project and could reproduce the behaviour you described. As you alread thought, the circulating/looping/toggling effect is a drawback of the method. However the increasing memory requirement may be a flaw of the implementation. We will double-check this issue.
I could also reproduce the behaviour of not applying 100% of the load for the fine project. I have to make some additional tests and will provide feedback as soon as possible.
after reviewing your issue i came to the conclusion that unfortunatelly there is no law that tells you how to set up NLAERH and BGLAENG properly, sometimes it is still trial and error to get to a well working configuration.
If it is possible for your calculation, you should stick with the Newton-Raphson method.
thank you very much for your help Daniel