Improve Simulation Accuracy / Best Way

[ParticleSkull]

Hey guys, what is the best way to improve a simulation's accuracy? 

I've managed to do it decreasing the Time Step on the AutoDopNetwork's Simulation tab. Altough it is linked to some parameters which I don't really understand.
What I did was divide everything by 20.

Would it have any other, maybe most commonly used, way to achieve it?

here's an snapshot of what i did:
https://www.dropbox.com/s/xxh30sbycicc6vl/sim-step.jpg?dl=0

Thx

[Ahmed Saady]

Well , for the time step i did not test it before but what i read in houdini help it is : 

Determines if the current solver timestep will be used to apply this node.

If set, the current timestep size will be multiplied by the scale and used for the time increment for this operation. Otherwise, the time scale will specify an absolute fictitious time to integrate by.

By disabling the link between the actual real time and the microsolver time, you can perform operations in a separate, fictitious, time. 

 

but if we are talking about making your sim much accurate i think you could add more sup steps so the frame will calculate twice or 4 time or more dependence on what you are looking for .

 

In destruction adding sup steps could give a huge change and maybe change your destruction look ,, but not all the time.

so you have to test and compare .

[rayman]

Setting TimeStep directly may not be the best way to change the accuracy as it works in seconds. Value of 1 means that DOP network will be evaluated every second. On the other hand this is related to the scene FPS and it may starts to feel a bit confusing trying to set the proper value.

That's why there is an expression set for you by default and it is related to the Sub Steps Parameter (evaluation per frame).

[ParticleSkull]

So the best approach would be increasing the Sub Steps?

Thx guys

[ParticleSkull]

I'm actually working on a FEM R&D and, no matter how much i increase the Sub Steps value, the donut always go through the seccond box.

I realise that the problem happens 'cos the box is too thin, although i would like to understand how to solv it like this, for thin object.

 

I've tryied increasing the Sub Step value to 100 but then the torus kind of exploded and it went through the second boxs.

Here's the file in anyone like to have a look:
https://www.dropbox.com/s/swa5ttsa6awr58l/FEM_SimpDonut.rar?dl=0

[lukeiamyourfather]

Yes, increasing substeps will make the solvers calculate changes in the simulation more frequently in scene time. For fast moving objects or highly detailed simulations it might take many substeps to resolve. Sometimes 100+ depending on the simulation but typically you can get away with using fewer. Resolution is the other big factor. Simulations for liquids, fire, smoke, and other natural phenomena become more believable with higher resolutions. As the resolution increases usually so do substeps.

[rayman]

Keep in mind that some solvers in DOP network has its own substep parameters which evaluates within 1 DOPnet substep.

[ParticleSkull]

Thx Luke! It's good to know that i'm not going to crazy with 100+ substepts.

Pavel, just did some tests increasing the Min Substep Rate inside the FiniteElementSolver and it did change the results. 

Thx guys.