When I use your 2D SG building tool for the microstructure that I want, I get very accurate results comparing with RVE method and Chamis (taken for literature), yet what I need is a 3D, but the SwiftComp build-in only provides limited shape. Therefore, the microstructure (let’s say 3D RVE) that I want to create, would require to be done manually due to uncertainties I want to use. But, when I do so, the result are off by 50% especially in E1. I was able to get the accurate results by doing a slight modification in my geometry, which was making the fiber slightly smaller (shorter) than the matrix body, making it completely surrounded by matrix.
Could you please advise what you think about the above? As ideally, my use will require to have the fiber visual in the geometry, representing a periodic SG.
Kindly have a look at the attached photo...
Please let me know if you require more details or explanation to the above issue…
Yes, I'm working with unidirectional FRP. I need to do models as attached. I know this is no longer counts as FRP due to being Aperiodic. But at further steps of the study I will introduce uncertainties then regain the periodicity of the SG/RVE.
Is it related with fiber diameter change or orientation change? If just orientation change, this can be handled by coordinate transformation. Of course, nothing wrong with handling using 3D SG. Main issue could be the mesh. Some of my students will give you a better answer.
---- Emailed forum response from wenbinyu@purdue.edu
Could you please check your *.echo file in your work directory (at the end of this file)? I think there should be some nodes at the boundary surfaces without periodic nodes on the corresponding boundary surfaces. If this happens, a constraint weaker than periodic boundary condition(PBCs) but stronger than static uniform boundary conditions(SUBCs) will be used instead, which makes the result be off from what you expected( usually results from using PBCs). The discrepancy in the predicted effective properties will be decreased if you increase the RVE size (with more inclusions in the RVE).
To avoid it, you can try to make a mesh be more accurately periodic, or you can provide the periodic node pairs in your *.sc file directly(which cannot be generated automatically using the current GUI).
Thank you for your reply. It’s actually related to both, change in diameter and orientation. The difference in the results in both of the 3D SGs that I showed in the first photo is significant, although the size of the fibre is basically the same (Vf ratio).
.ech showing this:
Nodal coordinates of node without corresponding periodic node in y2 direction
--------------------------------------------
5.3341400E-01 1.0000000E+00 2.1457200E-01
.
.
.
So I sent it up as Aperiodic in y2, but still not getting the results that I expect. (not getting the above in the ech file)...
So the 50% off is between the 3D SG with full fiber and 3D SG with fiber embedded completely inside the matrix, both computed by SwiftComp? It is very possible that there are very different. Can you compare each to the RVE analysis you done to see the difference.
What are the boundary conditions you used for the abaqus results? SwiftComp is capable of handling various mesh mapping at the boundaries but the results should be better than other methods using the same mesh.
Hello Prof Winben, I’m using periodic boundary conditions similar to what is applied using digimat.
The results from SwiftComp using the SG 2D generation tool make perfect sense, and I’m a fan of how easy and fast the interface is. Even with the trick of embedding the fiber inside the ‘RVE’ result are fine. But of course, it would be better to present some work done with fibers where they should be, and I can’t take it further as at this stage I have a tool and data to verify my results.
I hope something would come up to tackle this issue, I’ll give it more tries and see if I can come up with something…
I understand that for RVE with fiber touch the surface you have a mesh which is not periodic. SwiftComp can deal with such mesh without problem, and the results should always be better than other methods. The results could be very different from the case with fibers embedded inside RVE. So now, let me ask you what is exactly your question or unexpected results?
I am curious, since you don't have corresponding mesh on the boundaries, how do you apply the periodic boundary conditions in Digimat?
I understand that for RVE with fiber touch the surface you have a mesh which is not periodic. SwiftComp can deal with such mesh without problem, and the results should always be better than other methods. The results could be very different from the case with fibers embedded inside RVE. So now, let me ask you what is exactly your question or unexpected results?
My first run was supposed to be a periodic 3D SG, however, I didn’t bother to make exact meshes on both from and back side as swiftcomp can handle such small mesh variation (where further I my study I may introduce aperiodic SGs), but the result were far less than the RVE method, Chamis equations and data in this paper (A unified periodical boundary conditions for representative). But when I made the SG with an embedded fibre, result were matching. The results are attached.
The problem is a composite with the following properties (as in the above paper):
Fibre: E=379300, v=0.1, Vf=0.47
Matrix: E=68300, v=0.3
I am curious, since you don't have corresponding mesh on the boundaries, how do you apply the periodic boundary conditions in Digimat?
I’m applying my BCs as in the above paper for E1, where its very easy to do so as the side of the RVE are with the exact mesh. As for how am I doing it using Digimat, I created a model in Digimat, where digimat can apply periodic BCs. then exported it as .py file, where in Abaqus I only change the part (and associated sets) and keep all constraint equations s imported from Digimat. But, later on in my study I would apply it using only Abaqus.
I hope my answers give you an idea of the problem that I’m facing. It would be graet if someone can run the above model (with fibres contacting the surfaces as they should) and give us a feedback of the results that they are getting ( I hope I'm doing something wrong)…
What I meant by someone try to do the same model, is by modeling a 3D part using Abaqus and applying homogenization, not by creating the part with the build-in SG 2D builder tool. again, these are my results from swiftcomp:
Can you also provide the model you created for the aperiodic case? And also can you provide the paper of Chamis.
---- Emailed forum response from wenbinyu@purdue.edu
I did the homogenization using your model 3D3.sat and material properties of fiber and matrix. In this model, the fiber intersect with the boundary surface of the unit cell. Here are the results I have obtained:
Sadik Omairey @ on — Edited @ on
Hello everyone,
When I use your 2D SG building tool for the microstructure that I want, I get very accurate results comparing with RVE method and Chamis (taken for literature), yet what I need is a 3D, but the SwiftComp build-in only provides limited shape. Therefore, the microstructure (let’s say 3D RVE) that I want to create, would require to be done manually due to uncertainties I want to use. But, when I do so, the result are off by 50% especially in E1. I was able to get the accurate results by doing a slight modification in my geometry, which was making the fiber slightly smaller (shorter) than the matrix body, making it completely surrounded by matrix.
Could you please advise what you think about the above? As ideally, my use will require to have the fiber visual in the geometry, representing a periodic SG.
Kindly have a look at the attached photo...
Please let me know if you require more details or explanation to the above issue…
Best regards
Wenbin Yu @ on
First, are you working on unidirectionally fiber reinforced composites? If yes, then I could not understand why you need 3D SG.
Sadik Omairey @ on — Edited @ on
Hello Prof Wenbin,
Yes, I'm working with unidirectional FRP. I need to do models as attached. I know this is no longer counts as FRP due to being Aperiodic. But at further steps of the study I will introduce uncertainties then regain the periodicity of the SG/RVE.
Wenbin Yu @ on — Edited @ on
Bo Peng @ on
Hi,
Could you please check your *.echo file in your work directory (at the end of this file)? I think there should be some nodes at the boundary surfaces without periodic nodes on the corresponding boundary surfaces. If this happens, a constraint weaker than periodic boundary condition(PBCs) but stronger than static uniform boundary conditions(SUBCs) will be used instead, which makes the result be off from what you expected( usually results from using PBCs). The discrepancy in the predicted effective properties will be decreased if you increase the RVE size (with more inclusions in the RVE).
To avoid it, you can try to make a mesh be more accurately periodic, or you can provide the periodic node pairs in your *.sc file directly(which cannot be generated automatically using the current GUI).
Best regards,
Bo
Sadik Omairey @ on
Thank you for your reply. It’s actually related to both, change in diameter and orientation. The difference in the results in both of the 3D SGs that I showed in the first photo is significant, although the size of the fibre is basically the same (Vf ratio).
.ech showing this:
Nodal coordinates of node without corresponding periodic node in y2 direction
--------------------------------------------
5.3341400E-01 1.0000000E+00 2.1457200E-01
.
.
.
So I sent it up as Aperiodic in y2, but still not getting the results that I expect. (not getting the above in the ech file)...
Bo Peng @ on
May I ask which two results are you comparing please?
Sadik Omairey @ on
These are the results of the SG of the full fiber
E1 = 1.4194360E+05
E2 = 1.0270554E+05
E3 = 1.4195264E+05
G12 = 4.6863766E+04
G13 = 4.5805618E+04
G23 = 4.6832869E+04
nu12= 2.0174314E-01
and these are for the SG with fibers not touching the SG boundries
E1 = 1.4477740E+05
E2 = 2.1519646E+05
E3 = 1.4478109E+05
G12 = 5.4923056E+04
G13 = 4.6025626E+04
G23 = 5.4940977E+04
nu12= 1.3072595E-01
E2 is kinda of a main intrest.
Wenbin Yu @ on
So the 50% off is between the 3D SG with full fiber and 3D SG with fiber embedded completely inside the matrix, both computed by SwiftComp? It is very possible that there are very different. Can you compare each to the RVE analysis you done to see the difference.
Sadik Omairey @ on
Sorry for the later reply, Prof Wenbin, the results from abaqus as below:
E11 for a normal RVE = 215.124 Gpa
E11 for a covered fibre within the RVE = 215.050 Gpa
Basically the same results...
Generally speaking, is SwiftComp capable to handle various mesh mapping at the boundaries?
Many thanks
Wenbin Yu @ on
What are the boundary conditions you used for the abaqus results? SwiftComp is capable of handling various mesh mapping at the boundaries but the results should be better than other methods using the same mesh.
Sadik Omairey @ on
Hello Prof Winben, I’m using periodic boundary conditions similar to what is applied using digimat.
The results from SwiftComp using the SG 2D generation tool make perfect sense, and I’m a fan of how easy and fast the interface is. Even with the trick of embedding the fiber inside the ‘RVE’ result are fine. But of course, it would be better to present some work done with fibers where they should be, and I can’t take it further as at this stage I have a tool and data to verify my results.
I hope something would come up to tackle this issue, I’ll give it more tries and see if I can come up with something…
Best regards
Sadik
Wenbin Yu @ on
Let us get to the heart of the problem.
I understand that for RVE with fiber touch the surface you have a mesh which is not periodic. SwiftComp can deal with such mesh without problem, and the results should always be better than other methods. The results could be very different from the case with fibers embedded inside RVE. So now, let me ask you what is exactly your question or unexpected results?
I am curious, since you don't have corresponding mesh on the boundaries, how do you apply the periodic boundary conditions in Digimat?
Sadik Omairey @ on
Hello Professor
Below are my answers to your comments
I understand that for RVE with fiber touch the surface you have a mesh which is not periodic. SwiftComp can deal with such mesh without problem, and the results should always be better than other methods. The results could be very different from the case with fibers embedded inside RVE. So now, let me ask you what is exactly your question or unexpected results?
My first run was supposed to be a periodic 3D SG, however, I didn’t bother to make exact meshes on both from and back side as swiftcomp can handle such small mesh variation (where further I my study I may introduce aperiodic SGs), but the result were far less than the RVE method, Chamis equations and data in this paper (A unified periodical boundary conditions for representative). But when I made the SG with an embedded fibre, result were matching. The results are attached.
The problem is a composite with the following properties (as in the above paper):
Fibre: E=379300, v=0.1, Vf=0.47
Matrix: E=68300, v=0.3
I am curious, since you don't have corresponding mesh on the boundaries, how do you apply the periodic boundary conditions in Digimat?
I’m applying my BCs as in the above paper for E1, where its very easy to do so as the side of the RVE are with the exact mesh. As for how am I doing it using Digimat, I created a model in Digimat, where digimat can apply periodic BCs. then exported it as .py file, where in Abaqus I only change the part (and associated sets) and keep all constraint equations s imported from Digimat. But, later on in my study I would apply it using only Abaqus.
I hope my answers give you an idea of the problem that I’m facing. It would be graet if someone can run the above model (with fibres contacting the surfaces as they should) and give us a feedback of the results that they are getting ( I hope I'm doing something wrong)…
Best regards
Sadik Omairey @ on
What I meant by someone try to do the same model, is by modeling a 3D part using Abaqus and applying homogenization, not by creating the part with the build-in SG 2D builder tool. again, these are my results from swiftcomp:
The Engineering Constants (Approximated as Orthotropic)
----------------------------------------------------------
E1 = 1.4178170E+05
E2 = 1.0683406E+05
E3 = 1.4177778E+05
G12 = 4.6731682E+04
G13 = 4.5889837E+04
G23 = 4.6701956E+04
nu12= 1.8321713E-01
nu13= 2.3770799E-01
nu23= 1.3780276E-01
Attached you will find the part…
Regards
model
8 KBClick to download
Wenbin Yu @ on — Edited @ on
Sadik Omairey @ on
Hello Prof. Webin,
Sorry for the late reply, I was not able to logon, and th issue was fixed now.
Attached yu will find the paper and the model.
Best regards
3D3.sat
9 KBClick to download
Sadik Omairey @ on
Paper
1-s2.0-S0020768303000246-main.pdf
300 KBClick to download
Wenbin Yu @ on — Edited @ on
Bo Peng @ on
Hi Sadik,
I did the homogenization using your model 3D3.sat and material properties of fiber and matrix. In this model, the fiber intersect with the boundary surface of the unit cell. Here are the results I have obtained:
E1 = 1.4340502E+05
E2 = 2.1330599E+05
E3 = 1.4337793E+05
G12 = 5.3961830E+04
G13 = 4.5769287E+04
G23 = 5.3949870E+04
nu12= 1.3152204E-01
nu13= 2.5285844E-01
nu23= 1.9567735E-01
The fiber direction is the Y as shown in the attached picture. Therefore the modulus in the fiber direction is 2.1330599E+05 .
The abaqus input file is attached.