Time to solve a problem

[dw1982]

Is there a way to determine how much time is required to solve a contact problem?

Abaqus solved a similar problem with half the number of elements in 30 minutes but this analysis has been running for more than 3 hours now.

According to the Data file, the memory to minimize I/O is 3141 but I could only preallocate 1500 MB.

Does the wallclock time (202 sec) relate to the actual analysis time?

Thanks

 

[rstupplebeen]

Previous experience is about all you have for nonlinear analysis. View the results with CAE to see if there is an issue.

Having enough memory definitely helps. HD speed is much slower. Try meshing more stingy and then before using the results do a mesh sensitivity study.

Wall clock is similar to analysis time except that you need to factor in the number of CPUs you have. I usually look at the LOG file instead.

I hope this helps.

Rob Stupplebeen

 

[loki3000]

what is a mesh sensitivity analysis? where do i find it? (i don't have search in help for now unfortunately)

 

[rstupplebeen]

It is a general FEA concept. Basically mesh with different mash sizes to determine where the results level off for your particular study and desired accuracy. I hope this helps.

Rob Stupplebeen

 

[loki3000]

is this adaptive meshing or you can calculate it manually? (ansys has something like that based on the element energy or the diff. between the node avg. and node stress).

 

[TERIO]

It's not adaptive meshing. Basically, you would do something like this:

1. Create geometry
2. Generate mesh
3. Solve
4. Generate a new mesh with smaller elements (e.g., half the size of step 2)
5. Solve with new mesh
6. Compare results of step 3 and 5 to see the impact of the mesh on the solution
7. Repeat until you get roughly the same answer with both meshes

 

[ss770640]

dw1982

the time it takes any FEA program to solve a problem is linked to

(a) the increment step size
(b) the logging rate

the larger these two are, the shorter the analysis time. But rule of thumb is logging should be at half that of increment. This will avoid spikes in yoru output data.