Ruochun, That is much clearer, thank you! I will relax the Young's modulus, and try to run some tests to have a correct range of time steps for my application.
Thanks! On Saturday, March 16, 2024 at 3:11:29 AM UTC-4 Ruochun Zhang wrote: > Hi Yves, > > The particle size and the physics model come into play in determining the > step size, too. There might also be other factors. So using velocity and > stiffness alone is not enough to decide. However, if you already know the > proper step size for a set of properties, then you can perhaps empirically > decide how the step size should change if you change the properties. For > example, an increase in the max velocity should linearly decrease the > required step size. That's how I typically do it. > > For the wall properties, I don't see it being governed by different rules > other than the particles', should you need to relax the physics. The > underlying contact model is the same. > > Thank you, > Ruochun > > On Saturday, March 16, 2024 at 2:12:35 AM UTC+8 [email protected] > wrote: > >> Thank you, Ruochun, for your answer. >> >> That makes sense. Is there a common criterion (equation) for determining >> the max velocity/Young's modulus vs. the time step? >> In addition, what are common wall properties that people would use? In my >> field, the ball properties are extensively discussed, but the wall >> properties are never. >> >> Thank you! >> >> >> >> On Friday, March 15, 2024 at 2:01:04 PM UTC-4 Ruochun Zhang wrote: >> >>> Hi Yves, >>> >>> This is expected. The step sizes 1e-4, 5e-5, and 3e-5 you used gave >>> unphysical responses since they were not fine enough to capture the >>> collision. Past those, it becomes better. I'd like to note that the test >>> scenario is challenging to begin with, since the ball impacts with a high >>> velocity and the wall stiffness is very high (2.5e11 Pa), making the >>> contact difficult to resolve. I am not surprised at all that it requires a >>> step size as small as 1e-5s. You should try smaller step sizes too, and >>> they should converge to a specific bouncing pattern, and this is perhaps >>> the way to find an appropriate step size to use (the largest one that gives >>> this "converged" bouncing). >>> >>> In reality, the vast majority of DEM simulations use artificially >>> reduced stiffness to relax the physics, with empirically decided step >>> sizes. The rule is trying to ensure that the contact events are resolved >>> with no less than 4 time steps (but more is even better), but it's not >>> always necessary to do separate tests to find that out. Reasonable bulk >>> metrics like total energy or max velocity are sometimes enough of an >>> indicator of good simulations, plus many simulations have main physics >>> driven by prolonged contacts like grinding rather than "hard" collisions, >>> making resolving "worst" contacts like the collisions that happen only in >>> parts of the simulation not a big concern. >>> >>> You probably know that DEM-Engine has the method *UpdateStepSize *to >>> allow for on-fly step size change, should the physics change significantly >>> at different parts of your simulation which mandates an update on the step >>> size. However, that is manual and requires that you have an idea of the >>> appropriate step size. As for automatic step size adaption, it is a bigger >>> engineering problem, and the way I plan to implement that involves a >>> complex prerequisite subsystem too (I won't spoil it). It will not be there >>> too soon, because the testing needed to show it is a meaningful mechanism >>> sounds like a full paper to me. >>> >>> Thank you, >>> Ruochun >>> >>> On Saturday, March 16, 2024 at 1:00:42 AM UTC+8 [email protected] >>> wrote: >>> >>>> Hello, >>>> >>>> I wanted to verify that my time step was right in my "ball drop" >>>> simulations. in DEM-Engine. >>>> One minimal example I created is attached. I just drop one ball in a 5 >>>> meter-high box from z=2m with a null velocity. Then, I record and plot the >>>> velocity and elevation of the ball over time. >>>> >>>> Here are some examples attached, with dt=1e-4, 1e-5, 2e-5, 3e-5, and >>>> 5e-5 seconds. >>>> >>>> What I found is that the behavior of the ball is completely different >>>> between different time steps: the bounces can make the ball go higher than >>>> its initial position, which breaks the energy conservation. Even stranger, >>>> setting a very small time step also leads to that issue. >>>> >>>> Here, I do not know which one should be taken as true, dt=1e-5s seems >>>> good, but if I double the time step there are still bounces, just lower. >>>> Here this is for a very simple example, but that tells me that there is >>>> too >>>> much variability in the results based on the time step. >>>> >>>> Therefore, I was wondering: is there an issue with my test? If not, how >>>> do we know the time step we are setting is hitting the sweet spot: not too >>>> low, not too large? Is there a criterion we can use? And finally, can >>>> DEM-Engine adapt its time step automatically to prevent the user to play >>>> with that value when it might produce unrealistic behaviors? >>>> >>>> Thanks! >>>> >>>> -- You received this message because you are subscribed to the Google Groups "ProjectChrono" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/projectchrono/19c269e7-5dc4-4200-955a-81d0a4e1de02n%40googlegroups.com.
