Learning Forum

[ekayar01]

We received the segmentation error at various timesteps for the same model just increasing the speed linearly as shown below. I tried to solve implementing various solver parameters like double-precision, medium priority hardware sources and so on. It is interesting that the error occurs inconsistently. Therefore, it is not easy to diagnosis from my point of view. All the models run on a cluster using single-precision Dyna solver with ncpu=48 & 16GB as shown below.

Can you please let me know about your comments to find out the source of the problem? Partially the slurm file shared below.

-----------------------------------------------------------------------

30kph

     node  number 362930 deleted at time  4.5900E-03

forrtl: severe (174): SIGSEGV, segmentation fault occurred

 

60kph

      node  number 387621 deleted at time  3.7800E-03

forrtl: severe (174): SIGSEGV, segmentation fault occurred

 

90kph

     solid element 330148  failed at time  2.9691E-03

forrtl: severe (174): SIGSEGV, segmentation fault occurred

 

120kph

     shell element 584554  failed at time  1.1339E-02

forrtl: severe (174): SIGSEGV, segmentation fault occurred

------------------------------------------------------------------------------

#SBATCH --ntasks=48
#SBATCH --partition=k2-hipri
#SBATCH --nodes=1
#SBATCH --mem-per-cpu=16G

[Alex Rycman]

Hello,

Segmentation errors are not only related to hardware issue. Before looking at the hardware and simulation setup option, it would be good to check if your model behaves as expected. There are some node deletion and shell element failure just before segmentation error. That is a good place to start and find out why those nodes and elements are being deleted. You can visualize it by increasing the frequency of d3plot (*DATABASE_BINARY_D3PLOT). Another option is to look at global energy (GLSTAT) and energy in each part (MATSUM), and look for spies, negative values and unexpected behavriours. 

It is advised to use the latest release of LS-DYNA. You can download LS-DYNA here:

https://ftp.lstc.com/user/download_page.html

username: user

password: computer

 

Let me know how it goes. 
Alex 

[ekayar01]

Hi Alex,

Thank you for your reply. The model and the keywords were checked before the solutions. Enhanced UD lay-up defined with a damage model (*MAT55). The elements reach the damage limit that is why elements are deleted. The energy results as well as energy ratio are monitored for all the cases. The table below is expressing the percentage of the solution achieved for both single and double precision until the solution is terminated due to segmentation fault. It means that the normal termination almost achieved for 30kph and 90kph cases using double-precision where the rest of them were failed. Can you please explain the difference between the percentage achieved?

Regards,

[Alex Rycman]

Hello,

The fact that the double precision model terminates later suggests that the model is quite non-linear, and instability occurs. Please define in your input *CONTROL_SOLUTION, ISNAN=1 that might give you more insight into what is going on. 

Another issue might occur if erosion occurs in a model and a part is split between two different CPUs. If you think that there is a certain part that creates problem you can use *CONTROL_MPP_DECOMPOSITION_ARRANGE_PARTS. This keyword will force solver to isolate that part in a single processor. 

Please let me know how it goes.
Alex. 

 

[ekayar01]

Hi Alex,

*CONTROL_SOLUTION, ISNAN=1

*CONTROL_MPP_DECOMPOSITION_ARRANGE_PARTS (Type:11) keywords are added to the models.

You may see that there is no big difference as shown in the table below. At the left hand side the previous cases (120-90-60-30kph impact velocity cases respectively) and the right hand side the keywords added. All the cases performed using double precision.

Regards, 

[ekayar01]

Hi Alex,

Do you have any other advices? I performed by modifying and adding the keywords you mentioned but I can not reach a full solution as showed in the thread.

Regards,

[Alex Rycman]

Hello,

Thanks you for the update.

Did the termination error has change? Is is still forrtl: severe (174): SIGSEGV, segmentation fault occurred, or maybe you can see NaN errors or number of nodes has out-of-rage forces? 

I will suggest to simplify the model as much as possible so those simulations are going to run to compleation. Once the model terminates normally, then you can add complexity. That way you can determine what is causing a segmentation error and look at this keyword more closely.

If you are a commercial customer, you can create a support case and we will have a look:

customer.ansys.com

Note that Ansys employees are not allowed to upload and download files on the Ansys Forum.

Alex. 

 

[ekayar01]

Hello,

All the cases except 30kph & double-precision is terminated with forrtl: severe (174): SIGSEGV, segmentation fault occurred message. So, the FE model (30kph & double-precision) can run this case and terminate normally, but as I mentioned before when it comes to increasing the speed for the same case this message terminates the solution.

Can you please express maybe you can see NaN errors or number of nodes has out-of-rage forces this further and your solution?

Can not I ask for a support as an academic customer?

Regards,

Eray

[Alex Rycman]

Hello, 

*CONTOL_SOLUTION ISNAN =1 -> The keyword invokes algorithm that looks for  NaN or out-of-range (large values e+20) in quantities like velocity, force, added mass etc. Sometimes segmentation faults occur because the solver is dealing with numbers like that. Then that’s the information what could be wrong in the model. Here is an article that you can take a look https://lsdyna.ansys.com/not-a-number-nan-1/

Here you can also find other tips how to battle a segmentation error https://lsdyna.ansys.com/sigsegv-d61/

Without looking at the model I can only advise to simplify the model, get it to the stage where it is running and then start adding complexity. Please use the latest LS-DYNA release R14.1.

Regarding using Ansys customer support, you will have to contact a personal at your university/company that manages Ansys licensing. They will be able to give you more details if you are eligible for customer support.   

Best regards,
Alex