SPH, SMP, same input produces different results

[Pavel Dybskiy]

Hi,

I'm learning how to simulate metal cutting using SPH. At some point I ran into a situation when identical input k-files produce different results. 

How such discrepancy can be explained? So far I knew that different solvers and/or different CPU number, and/or different memory allocation may produce different results. Otherwise, if there are  identical input, hardware, software, then one should expect identical results, unless solver algorithms use randomization somewhere.  

Both results were obtained using the same computer, same SMP-D solver,  same CPU=4, same memory allocation.

Figures below show that second cutting goes longer before SPH part explodes. 

 

Best Regards,

pavel 

[l...@schwer.net]

In general, I agree with you that running the same input twice as you have indicated should result in the same response.

 

However, when the simulation involves an incipient instability, e.g. SPH exploding, then there is an exception to your expectation.

 

Instabilities are, well unstable and, subject to any system noise, e.g. numerical roundoff. For example look at the 5000 step times for the two runs:

                        5000  1.6818E-2 and 1.6824E-2

A small difference, but enough to indicated even at this early stage in the runs something is different.

 

Perhaps you would have better luck with this SPH simulation if you DEACTIVATED the particles that become part of the chip?

 

                --len

 

Question: Why did you run the same input twice?

 

From: ls-d...@googlegroups.com <ls-d...@googlegroups.com> On Behalf Of Pavel Dybskiy
Sent: Monday, September 5, 2022 11:37 AM
To: LS-DYNA2 <ls-d...@googlegroups.com>
Subject: [LS-DYNA2] SPH, SMP, same input produces different results

 

Hi,

I'm learning how to simulate metal cutting using SPH. At some point I ran into a situation when identical input k-files produce different results. 

How such discrepancy can be explained? So far I knew that different solvers and/or different CPU number, and/or different memory allocation may produce different results. Otherwise, if there are  identical input, hardware, software, then one should expect identical results, unless solver algorithms use randomization somewhere.  

Both results were obtained using the same computer, same SMP-D solver,  same CPU=4, same memory allocation.

Figures below show that second cutting goes longer before SPH part explodes. 

 

 

 

Best Regards,

pavel 

 

 

--
You received this message because you are subscribed to the Google Groups "LS-DYNA2" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ls-dyna2+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ls-dyna2/badf8da7-8233-415a-a56c-48f6f8652d30n%40googlegroups.com.

[James M. Kennedy]

Dear Pavel,

 

Perhaps these presentations may be of some interest:

 

Yreux, E., "MLS-Based SPH in LS-DYNA for Increased Accuracy and Tensile Instability", 15th International LS-DYNA Users Conference, Dearborn, Michigan, June, 2018.

 

https://www.dynalook.com/15th-international-ls-dyna-conference/sph/mls-based-sph-in-ls-dyna-r-for-increased-accuracy-and-tensile-stability

 

https://www.youtube.com/watch?v=Qo9tmlDfiWQ

 

https://www.youtube.com/watch?v=pwdoZfi9OBA

 

Duckworth, H., Sharp, D.J., and Ghajari, M., “Smoothed Particle Hydrodynamic Modelling of the Cerebrospinal Fluid for Brian Mechanics: Accuracy and Stability”, International Journal for Numerical Methods in Biomedical Engineering, Vol. 37, 17 pages, January, 2021.

 

https://onlinelibrary.wiley.com/doi/pdf/10.1002/cnm.3440

 

Sincerely,

James M. Kennedy

KBS2 Inc.

September 5, 2022

 

From: ls-d...@googlegroups.com [mailto:ls-d...@googlegroups.com] On Behalf Of Pavel Dybskiy
Sent: Monday, September 05, 2022 1:37 PM
To: LS-DYNA2 <ls-d...@googlegroups.com>
Subject: [LS-DYNA2] SPH, SMP, same input produces different results

 

Hi,

--

[Pavel Dybskiy]

I did run it twice to get excited with what one should expect when dealing with ls-dyna :) 

Someone said, implicit simulation is a skill, explicit is an art, or and vice versa.

Indeed I got stuck with what I thought would be a quick two-day validation run using published data on a "simple and straightforward" metal cutting model.  

1) 2020 “A comparison of smoothed particle hydrodynamics (SPH) and coupled SPH-FEM methods for modeling machining” by Nishant Ojal et al

2) 2013 "On the SPH Orthogonal Cutting Simulation of A2024-T351 Alloy" by Martin Madaj et al

I'm trying and trying various combinations of parameters in Section_SPH, Contact_XXX, and Control_SPH.

So far only partial success. I'm in between two states, 1) either cutting into SPH part starts more or less ok and then it blows early, or 2) dyna throws a run-time error. 

 

Basically, default settings don't give me the results I see in publications, and I don't have guiding principles on which parameters I should change next.

Research is not always fun :)

Sincerely,

paul

--

Best

-p.d.