SPH -- particles deletion / rapid increase of kinetic energy of the particles

[blazejpop]

Hello everyone!
I'm performing an SPH analysis of the airbag impact. It's equipped with an exhaust vent. The simulation goes well until some point in time when some rapid increase in kinetic energy of the fluid can be observed (it is unjustified and maybe it's a kind of an artifact or the characteristics of the SPH method in Abaqus). I've tried to play with the definition of air material (used this blog post for reference: [URL unfurl="true"]https://info.simuleon.com/blog/how-to-use-cel-to-model-air-pressure[/url]) but it didn't help.
Now here comes my question:
Is it possible to delete the SPH particles depending on some of their kinematic characteristics (like the velocity or displacement magnitude) so the energy wouldn't rise so much? Or maybe someone has some ideas for other options to deal with this issue?

Below you can find some pictures showing the problem:
1. Initial state

2. Velocity field a frame before the spike

3. Velocity field a frame after the spike

4. Displacement field before the spike

5. Displacement field after the spike

 

[FEA way]

There are multiple criteria (including user subroutine-based one) but for finite element conversion to SPH particles.

Have you considered using other methods to simulate this phenomenon? Fluid cavity feature is usually used to simulate airbags. There’s even an example similar to your case in Abaqus documentation ("Side curtain airbag impactor test").

 

[blazejpop]

Hello "FEA way"! Thank you for the interest!
Yes, I have used the 'fluid cavity' functionality. I wanted to use this more sophisticated approach, as now I'm interested in the behavior of the structure with specific shape of the exhaust vent (and possibly some obstacles to the flow which could choke it a bit (something like the discharge coefficient)).
I know about the conditions for transforming solid elements into SPH particles. I know also about the possibility to delete solid elements due to some defined damage conditions.
I know that the other possibility is to use CEL, however I wanted to try SPH as it is described as kind of a midway between UPM (fluid cavities) and CEL.

 

[FEA way]

Have you tried increasing the number of particles ? Have you noticed any specific/unusual behavior when taking a closer look at the time history animation around the point of rapid energy increase ? Can you attach the kinetic energy plot ?

 

[blazejpop]

I'm in the process.
The problem is that after this spike in the energy occurs the simulation starts to consume very big amounts of RAM and it just stops after some time (I didn't checked how much exactly but for sure more than 30 GB, while at the beginning it says it will use about 200 MB). Without any error or something. It just calculates some more integration steps and stops.
About the number of particles:
I started the simulation with converting C3D8R elements into SPH particles with 2 particles on the isoparametric direction. Now I changed it to 1 and I think it performs a bit better, however the problem still persists.
The kinetic energy plot was something like the broken line -- first with some small slope and then suddenly with the higher one. For now I cannot attach it because the simulation is on and for now it just gives me a dot placed in zero time.
I've investigated the moments around the problematic time point (frames attached in my original post), however I didn't notice anything "strange". Air material inside the airbag behaves "normal" even after this problematic time point (as I wrote it manages to calculate some more time steps).

I will try some combinations with number of particles and maybe the velocity of impacting body and will get you noticed.

 

[FEA way]

Generally speaking, various jumps or sudden peaks are very common in Abaqus/Explicit results. Usually, they can be ignored or filtered out if the values don’t change too much and the fluctuations show some pattern. However, such isolated energy changes should be investigated. Maybe plots of other variables (especially different types of energies) will give you some insight into what happens at this point of the analysis.