Micro and Macro relation in CPM pararmeters
In the name of God
Hello All
I am working on calibration of CPM model elastic parameters [1]. When calibrating E and nu, there is not enough information about selected E for inter particle interactions and obtained E from calibration process. Indeed i tried to create cylinder with determined E and nu, then i loaded it and computed the macro parameters (E, nu) following the procedure explained in [1]; however there was a great difference between the micro and macro E and nu?
Any suggestion? Can we set a geometry which lead to macro parameters the same as micro ones?
I have checked [2] but still the question is unresolved.
[1]: https:/
[2]:Jan Stránský, Milan Jirásek and Václav Šmilauer "MACROSCOPIC ELASTIC PROPERTIES OF PARTICLE MODELS" ; MS’10 Prague
Proceedings of the International Conference on Modelling and Simulation 2010
22 – 25 June 2010, Prague, Czech Republic
Question information
- Language:
- English Edit question
- Status:
- Solved
- For:
- Yade Edit question
- Assignee:
- No assignee Edit question
- Solved by:
- Jan Stránský
- Solved:
- Last query:
- Last reply:
Revision history for this message
|
#1 |
HI Mohsen,
could you please provide a MWE (how you evaluate "determined" and
"computed" E and nu ).
Next time plese be more specific what "great difference" is (10%? 1000%?)
Can we set a geometry which lead to macro parameters the same as micro ones?
do you mean that you set CpmMat(
modulus would be 25 GPa? For sure there exists suxh geometry, but it would
be very special case.. If I did not get it correctly, please correct me.
Thanks
Jan
PS: both your references [1] and [2] refers to the same method..
2016-10-23 15:03 GMT+02:00 mohsen <email address hidden>:
> New question #403385 on Yade:
> https:/
>
> In the name of God
>
> Hello All
>
> I am working on calibration of CPM model elastic parameters [1]. When
> calibrating E and nu, there is not enough information about selected E for
> inter particle interactions and obtained E from calibration process. Indeed
> i tried to create cylinder with determined E and nu, then i loaded it and
> computed the macro parameters (E, nu) following the procedure explained in
> [1]; however there was a great difference between the micro and macro E and
> nu?
> Any suggestion? Can we set a geometry which lead to macro parameters the
> same as micro ones?
> I have checked [2] but still the question is unresolved.
>
>
>
> [1]: https:/
> [2]:Jan Stránský, Milan Jirásek and Václav Šmilauer "MACROSCOPIC ELASTIC
> PROPERTIES OF PARTICLE MODELS" ; MS’10 Prague
> Proceedings of the International Conference on Modelling and Simulation
> 2010
> 22 – 25 June 2010, Prague, Czech Republic
>
> --
> You received this question notification because your team yade-users is
> an answer contact for Yade.
>
> _______
> Mailing list: https:/
> Post to : <email address hidden>
> Unsubscribe : https:/
> More help : https:/
>
Revision history for this message
|
#2 |
Hi Jan
>could you please provide a MWE (how you evaluate "determined" and
>"computed" E and nu ).
E and nu are the parameters which CPM material needs. Hence the phrases 'inter particle E and nu' or 'mecro ' are clearer! Surrey for wrong phrase!
Computed E and nu are calculated after axial loading (z axis) up to special strain (1e-2) as follows:
E=slope of sigma-epsilon graph
nu= - 0.5(epsx+
>Next time plese be more specific what "great difference" is (10%? 1000%?)
Yes depending on geometry it can be in the order of 1000% or even more!
Can we set a geometry which lead to macro parameters the same as micro ones?
I could not set it till now. It may be because of high desired nu which is near 0.35. Setting inter-particle nu as 0.35 causes large values of E which seems correct; however it is not may case
>do you mean that you set CpmMat(
>modulus would be 25 GPa?
Yes
>For sure there exists suxh geometry, but it would be very special case.
For cylinder with height to diameter ratio of two, I have not found such geometry till now. Also it may be interesting for you to check [3]. They could not achieve a set of inter particle parameters (E, nu, phi) that matches the macro ones. I think it is also the case for [2].
What do you think am i right?
Can we say: It is very hard to find a geometry satisfying this hypothesis: 'micro and macro parameters are the same'. If it is true then:
'Is there any reference that can interpret the difference between micro and macro parameters theoretically '? Is it a usual case in DEM modeling?
Anyway, my question finds still no answer: In references [1] and [2] have they compared values of micro and macro E and nu? I could not find any details about geometry in [2].
[3] : Wang X L, Li J C. Simulation of triaxial response of granular materials by modified DEM. Sci China-Phys Mech Astron, 2014, 57: 22972308, doi:
10.1007/
Revision history for this message
|
#3 |
Jan
There is another point that i forgot: why in CPM material the ratio of ks/kn is called Poisson ration? That is clear by increasing ks/kt (or inter particle nu), macro nu decreases as the lateral deformations are affected by shear relative movements. Hence such a physical definition (ks/kn) and terminology (nominating it as nu) are not consistent.
Revision history for this message
|
#4 |
Hi,
Short answer:
There is indeed no simple relationship between ks/kn and the material parameter called Poisson, but a dependency does exist (with both the solid phase Poisson coeff. and bulk scale Poisson) hence the conventional name.
Longer answer:
Since material parameters and interaction parameters in Yade are independent things (different functors will lead to different dependencies) it is not realistic to expect a physical consistency of the naming. In the end the parameter "Poisson" is just a value, it doesn't have to be ks/kn - only some particular functors may really use it to define ks/kn.
Besides, I can't follow you on the question "Can we set a geometry which lead to macro parameters the same as micro ones?"
If a particular micro-macro relationship (e.g. equality, in your case) exists the it should hold independently of the geometry of the specimen, else it is a physical aberration. So geometry is obviously not the right button to push to solve your problem.
Bruno
Revision history for this message
|
#5 |
@ Bruno
Hi Bruno
>Besides, I can't follow you on the question "Can we set a geometry which lead to macro parameters the same as micro ones?"
The geometry means: dimensions of cylinder or sample, radius of spheres,
or even interaction radius if we can take it into account.
>If a particular micro-macro relationship (e.g. equality, in your case) exists the it should hold independently of the >geometry of the specimen, else it is a physical aberration. So geometry is obviously not the right button to push to solve >your problem.
You are completely true.
I think you are agree with me that radius of particles are essentially effective on the results; this is also discussed by [3]; however about geometry practically may be following could be useful:
defining a parameter like 'number of particles per unit volume' and keeping it constant with different geometries; all other parameters such as number of interactions, CPM material parameters and ... should be the same.
Does it work? any better suggestion?
Revision history for this message
|
#6 |
Actually no, the micro-macro relations can always be put in a way such that they do NOT depend on particle sizes.
In Yade, at least, it is the case (and it is obviously a good feature as you may found out). If you agree that cylinder size/aspect ratio should not play a role either, then interaction radius if any is the only remaining geometrical parameter, but it is in the end a contact parameter more than a geometrical parameter.
What should bee kept the same in every geometry is porosity, coordination number, and fabric (an)isotropy. I think that's it.
Bruno
Revision history for this message
|
#7 |
Hi Mohsen,
please really provide a MWE :-) it would be much easier to determine the
source of problem..
First some explanation why the equivalence of micro and macro parameters is
almost never the case.
DEM is discrete, so what is really computed and important on contact is
f=k*u
f ... force [N]
u ... penetration depth [m]
k ... contat stiffness [N/m]
For easier definition of material models (and to make the elasticity
particle size independent, see Bruno's answer), it is possible to define a
contact law as stress=E*strain
For this, you need to convert the contact to **fictitious** truss,
assigning its length L and crosssection area A.
L can be easily defined as distance of particles' centers.
A is much more tricky, it is basically arbitrary value. It is covnenient to
make it proportional to particle size square. It is radius*radius in
FrictPhys models, but pi*radius*radius in CPM. So by changing this, you
change the explicit micro-macro relationship, preserving the model
consistency.
Another parameter is interaction ratio (aabbEnlargeFactor and
interactionDete
basically overall this more trusses makes higher stiffness.
Computed E and nu are calculated after axial loading (z axis) up to special
> strain (1e-2) as follows:
> E=slope of sigma-epsilon graph
> nu= - 0.5(epsx+
1e-2 is relatively high, there might be already some inelastic processes
(with the MWE I could tell if yes or no)
I could not set it till now. It may be because of high desired nu which is
> near 0.35. Setting inter-particle nu as 0.35 causes large values of E which
> seems correct; however it is not may case
nu=0.35 is relatively high and difficult to achieve with cohesive particle
models, this indeed might be the reason.
For cylinder with height to diameter ratio of two, I have not found such
geometry till now. Also it may be interesting for you to check [3]. They
could not achieve a set of inter particle parameters (E, nu, phi) that
matches the macro ones. I think it is also the case for [2].
What do you think am i right?
> Can we say: It is very hard to find a geometry satisfying this hypothesis:
> 'micro and macro parameters are the same'. If it is true then:
yes, it is hard, but there is no reason for it (see above).
'Is there any reference that can interpret the difference between micro
> and macro parameters theoretically '? Is it a usual case in DEM
> modeling?
It is a usual case in DEM modeling. The difference is intrinsic feature of
DEM models.
Anyway, my question finds still no answer: In references [1] and [2]
> have they compared values of micro and macro E and nu? I could not find
> any details about geometry in [2].
yes, macro E, nu and micro (young and poisson) parameters were compared.
Cubic periodic cell was used, bud the elastic parameters should be
independent on geometry.
There is another point that i forgot: why in CPM material the ratio of
> ks/kn is called Poisson ration? That is clear by increasing ks/kt (or
> inter particle nu), macro nu decreases as the lateral deformations are
> affected by shear relative movements. Hence such a physical definition
> (ks/kn) and terminology (nominating it as nu) are not consistent.
it is not only in CPM, but through almost all material models. I agree that
the name is not the best chosen, but on the other hand, it is the only
parameter influencing macroscopic Poisson's ratio :-)
You can observe (and theoretically prove) that changing poisson changes
both Poisson's ratio and Young's module, but changing young parameter does
change only Young's modulus, but not Poisson's ratio.
cheers
Jan
2016-10-25 16:27 GMT+02:00 Bruno Chareyre <
<email address hidden>>:
> Question #403385 on Yade changed:
> https:/
>
> Bruno Chareyre proposed the following answer:
> Actually no, the micro-macro relations can always be put in a way such
> that they do NOT depend on particle sizes.
> In Yade, at least, it is the case (and it is obviously a good feature as
> you may found out). If you agree that cylinder size/aspect ratio should not
> play a role either, then interaction radius if any is the only remaining
> geometrical parameter, but it is in the end a contact parameter more than a
> geometrical parameter.
>
> What should bee kept the same in every geometry is porosity,
> coordination number, and fabric (an)isotropy. I think that's it.
>
> Bruno
>
> --
> You received this question notification because your team yade-users is
> an answer contact for Yade.
>
> _______
> Mailing list: https:/
> Post to : <email address hidden>
> Unsubscribe : https:/
> More help : https:/
>
Revision history for this message
|
#8 |
@ Bruno
Thanks! I should investigate your suggestion. It looks great.
@ Jan
>please really provide a MWE :-) it would be much easier to determine the
>source of problem..
So according to #1 I mentioned the way i computed E and nu. But i think i did not understand the meaning of MWE. Can you tell me what it is?
> So by changing this, you change the explicit micro-macro relationship,
If micro-macro relationship changes, i think the model consistency would change too.
>1e-2 is relatively high, there might be already some inelastic processes
(with the MWE I could tell if yes or no)
NeverDamage is active! also I decreased it to 1e-3
>Next time plese be more specific what "great difference" is (10%? 1000%?)
Jan I did not consider Coordination number in the simulations. Hence the previous results can not be compared.
I want to conclude:
1- there is no rule that "micro and macro parameters should be the same".
2- macro parameters must be independent of geometry (However I am simulating a cylinder trying to keep CoordinationNumber, porosity and fabric constant and the macro parameters vary about 20%).
3- For calibration by selecting the appropriate contact law and setting the geometry constant:
3-1- The porosity should be the same as reality (first parameter suggested by Bruno @6)
3-2- The micro parameters should change in such a way that desired macro ones obtained (I do not have any idea about how to set coordination number and fabric isotropy). In the first try it is preferable to select micro parameters the same as macro.
3-3- It is better to validate the obtained results (micro parameters and the three suggested by Bruno) by changing the geometry; However the parameters should be kept constant.
Regards
Mohsen
Revision history for this message
|
#9 |
Hi Mohsen,
by MWE (Minimal Working Example [4]) I meant a script simulating the
cylinder and computing resulting parameters (e.g. computing strain parallel
to cylinder axis might be a bit tricky). I f you can provide such script,
still it would be good, thanks :-)
> So by changing this, you change the explicit micro-macro
> relationship,
> If micro-macro relationship changes, i think the model consistency would
> change too.
ok, it depends on the definition of consistency :-) you can define the same
material model only changing the definition of "interaction cross section
area" r*r or pi*r*r. The micro-macro parameters relation would not be the
same, but only multiplied by a constant factor..
NeverDamage is active! also I decreased it to 1e-3
even if neverDamage=True, there might be some plasticity (which is not
deactivated by neverDamage)..
1- there is no rule that "micro and macro parameters should be the same".
exactly
> 2- macro parameters must be independent of geometry (However I am
> simulating a cylinder trying to keep CoordinationNumber, porosity and
> fabric constant and the macro parameters vary about 20%).
**ELASTIC** macro parameters must be independent on geometry... actually it
does depend on geometry, because every time the interaction network is
random. Also the number of particles should be large enough to decrease
these random effects and also boundary effects (boundary behaves
differently from bulk, by increasing number of particles, the boundary
become less and less significant). But the variation should not be "large".
3- For calibration by selecting the appropriate contact law and setting the
> geometry constant:
> 3-1- The porosity should be the same as reality (first parameter suggested
> by Bruno @6)
If you simulate physical grains, then yes. In some application (typically
concrete simulated with CPM) particles are just artificial discretization
of in reality not porous material, and then this criterion does not make
sense, just you should keep the simulation porosity constant for different
simulations (e.g. randomDensePack with constant particle radius results in
porosity very close to 38%).
3-2- The micro parameters should change in such a way that desired macro
> ones obtained (I do not have any idea about how to set coordination number
> and fabric isotropy). In the first try it is preferable to select micro
> parameters the same as macro.
coordination number is controlled by aabbEnlargeFactor and
interactionDete
should be constant, too
fabric isotropy should be "autimatic" by default.
The "first try" of micro parameters is a tricky part. For normal models, a
good idea is to use default values or proposed in some example script. By
changing microparameters, you observe change of macroparameters and you can
use some optimization technique to obtain the desired values.
You can also use dimensional analysis to make the process with less
variables. E.g. for elastic parameters, the dimensional analysis shows that
macro Poisson's ratio only depends on micro poisson and that macro Young's
modulus is linearly proportional to micro young.
3-3- It is better to validate the obtained results (micro parameters and
> the three suggested by Bruno) by changing the geometry; However the
> parameters should be kept constant.
It is a nice approach. For example (any other tests are possible :-)
Calibrate parameters on cylinder and then use it on 3-point bending if you
have experimental or theoretical data.
cheers
Jan
[4] https:/
2016-10-26 14:58 GMT+02:00 mohsen <email address hidden>:
> Question #403385 on Yade changed:
> https:/
>
> mohsen posted a new comment:
> @ Bruno
>
> Thanks! I should investigate your suggestion. It looks great.
>
> @ Jan
> >please really provide a MWE :-) it would be much easier to determine the
> >source of problem..
>
> So according to #1 I mentioned the way i computed E and nu. But i think
> i did not understand the meaning of MWE. Can you tell me what it is?
>
> > So by changing this, you change the explicit micro-macro
> relationship,
> If micro-macro relationship changes, i think the model consistency would
> change too.
>
> >1e-2 is relatively high, there might be already some inelastic processes
> (with the MWE I could tell if yes or no)
> NeverDamage is active! also I decreased it to 1e-3
>
> >Next time plese be more specific what "great difference" is (10%? 1000%?)
> Jan I did not consider Coordination number in the simulations. Hence the
> previous results can not be compared.
>
> I want to conclude:
> 1- there is no rule that "micro and macro parameters should be the same".
> 2- macro parameters must be independent of geometry (However I am
> simulating a cylinder trying to keep CoordinationNumber, porosity and
> fabric constant and the macro parameters vary about 20%).
> 3- For calibration by selecting the appropriate contact law and setting
> the geometry constant:
> 3-1- The porosity should be the same as reality (first parameter suggested
> by Bruno @6)
> 3-2- The micro parameters should change in such a way that desired macro
> ones obtained (I do not have any idea about how to set coordination number
> and fabric isotropy). In the first try it is preferable to select micro
> parameters the same as macro.
> 3-3- It is better to validate the obtained results (micro parameters and
> the three suggested by Bruno) by changing the geometry; However the
> parameters should be kept constant.
>
> Regards
>
> Mohsen
>
> --
> You received this question notification because your team yade-users is
> an answer contact for Yade.
>
> _______
> Mailing list: https:/
> Post to : <email address hidden>
> Unsubscribe : https:/
> More help : https:/
>
Revision history for this message
|
#10 |
Dear Jan
I will Provide a MWE soonly. However i think by applying your recommendation, there would be no change by geometry.
Revision history for this message
|
#11 |
Thanks Jan Stránský, that solved my question.