Learning Forum

[Andres Mena]

Hello all,

I am running a simulation using LS-Dyna with *LOAD_NODE_SET and constraining some DOF at distinct ends with *BOUNDARY_SPC_SET. Following some recommendations given on previous posts, I am using a ramp to define the curve and then use a constant value (see graph below). To check the quasi-static condition, I am using glstat and matsum to verify that the kinetic energy is below 5% of the internal energy; this is true for glstat and for matsum only one part had high kinetic energy.

My question comes from checking some variables (see graph below for v-m stress of just 2 elements, whole part rigid-body motion) for which I observe this weird oscillating behavior that prevents me to obtain confident results at the end of the simulation; naturally I am expecting a line that shows results converging to a value (similar to my loading curve). 

Has anyone experience a similar issue like this? Is this caused by the ramp application curve... maybe it's too fast? Or maybe it's a lack of boundary conditions? Any idea would be greatly appreciated.

Also a follow up question. I noted there's an option for rigid-body acceleration in the history variable option: I have nonzero (still relatively small considering my time units) values for some parts. Is zero acceleration also a condition for quasi-static behavior or is the internal vs kinetic energy ratio sufficient?

 

 

 

[Alex R.]

Hello Andreas,

In any explicit simulation some degree of dynamic effects will be present, even if the ratio of kinetic energy to internal energy falls below 5%. There are couple of options that you could try: 

1. Include damping in your simulation.
2. Check the material models, if all of them are elastic and there is no damping, the system will oscillate more than a physical one. Please check if there are materials with viscoelastic properties in the model. 
   
3. Incorporate a dynamic relaxation phase in your simulation. Please check: https://ftp.lstc.com/anonymous/outgoing/jday/bolt_preload3.pdf
4. If you are not interested in dynamic events at all, the implicit solver can be used to solve the model.

Regarding your follow up question. It is hard to define what is a quasi-static condition for an explicit model. The 5% threshold is also just a rough estimate, not a specific threshold value. It  might not be applicable to some models. The question that should be asked if the rigid body would have a 0 acceleration will the result change, by how much? If the answer is that it will change too much then yes, the 0 acceleration should be pursued. 

Please let me know,

Thank you,
Alex

[Andres Mena]

Hello Alex, 

Thank you for getting back to me. Indeed, I am using only elastic material models so I will try these points in my model to see if I can decrease the dynamic effects to some extent. 

Regarding damping, I suppose *DAMPING_Global would suffice, or do you recommend another keyword? For the dynamic relaxation, I checked the link you provided and just to see if I got it correctly, I would apply my curve after this dynamic relaxation which requires another curve, right?

I understand your comment and it makes absolute sense. My thought was that this acceleration is related to the dynamic effects I am experiencing, but I believe adding damping should tackle both problems at the same time. I am trying to simulate a static case (isometric grasping, basically applying a constant load on one boundary) and I didn't know of the implicit solver when I started, so I will investigate if decreasing the acceleration changes the results. Maybe switching to the implicit could solve all of my problems in the near future. Thanks again for your help!

Thanks, 

Andres

 

[peteroznewman]

Hello Andres,

I agree with all the points Alex made and believe your best path forward in this case is the implicit solver.

In a different case where dynamics was required, I would add one more method for reducing oscillation in the solution: use a smooth displacement function. If you take the derivative of a linear ramp followed by a constant displacement, you get two step changes in velocity. Those two steps introduce a lot of oscillation. If you replace the linear ramp with a smooth function such as a half-period of a cosine function, the two step changes in velocity are replaced with a velocity that starts at zero, increases to a maximum and decreases to zero in a half-period of a sine function. This introduces much less oscillation in the dynamic solution.

Regards,
Peter

[Andres Mena]

Hello Peter,

Thank you for you reply! This new curve suggestion sounds excellent and I will try it. For my specific case, I believe introducing a smoother transition between the two curves could also help to stop these oscillations a little (like the figure)? I am aware that the step changes in velocity would still exist, but I would like to know how you handle whenever you need to use different loading curves in one simulation.

 

Best, 

Andres

[peteroznewman]

Andres, there are two sharp transitions that you should smooth out.  The first is at t=0 because the displacement before then is 0 then it suddenly becomes a ramp.  That is why I said a half period of a cosine function, the slope at t=0 is zero and the slope is zero again at the half period time.

[Alex R.]

Hello Andres, 

Please consider using *DEFINE_CURVE_SMOOTH in LS-DYNA. 

 

Thank you,

Alex