Nonlinear material behaviour in topology optimization
Hi everyone,
I’m currently doing some work regarding topology optimization in the civil engineering industry. One of the focuses of this work is to try to combine topology optimization with nonlinear material behaviour. I already worked with Abaqus regarding nonlinear static analysis using the concrete damaged plasticity constitutive model and with topology optimization assuming linear material behaviour.
I was searching for this topic online and I found the Altair software, which seems to be advanced regarding topology optimization.
I’m wondering if is possible to perform topology optimization processes considering the nonlinear behavior of concrete material, using altair products such as optistruct.
If not, even if you know or have some idea/suggestion about how this could be performed, (for example combining this software with some external programming), would serve as a great help in my work.
Thanks in advance,
Best regards,
gustacsilva
Answers
-
Hi @gustacsilva
Right now, OS supports nonlinear contacts combined with optimization. NL materials are not yet supported, so far.
Can you give some more details on the use case that you have in mind?
The main challenge that i see in combining NL problems with Optimization it is the run time that increases a lot.
Specially speaking on topology optimization, where you can easily have thousands of design variables, and need to compute sensitivities for each of them, if becomes really though.
When you combine external tools, such as HyperStudy with OptiStruct, the main problem go to sensitivities, as even worse, the external solver doesn't have access to sensitivities and need to rely on some global search or finite difference scheme for sensitivities.
Maybe knowing more on what you want to do, would help.
1 -
Of course.
So, my goal is to perform topology optimization processes considering the nonlinear behavior of fiber-reinforced concrete. In terms of design variables, for now, as is already difficult because of the material and geometric nonlinearity, I'm just using the most common ones, strain energy and volume. Where the goal is to reduce the strain energy for a specific volume. Maybe, later on, I would try to complicate a little more and change or add more design variables but for now, I'm keeping more simple.
The structure, for now, would be a simply supported beam and the load would be applied as pressure or displacement (depending on what is better of convergence).
Many parameters can be changed to achieve convergence, such as types of support, mesh, constitutive law, step parameters, and many others. But, for this case, despite the sensitivities problem, do you think that it can be achieved?
Thank you for the feedback!
0 -
Adriano A. Koga_21884 said:
Hi @gustacsilva
Right now, OS supports nonlinear contacts combined with optimization. NL materials are not yet supported, so far.
Can you give some more details on the use case that you have in mind?
The main challenge that i see in combining NL problems with Optimization it is the run time that increases a lot.
Specially speaking on topology optimization, where you can easily have thousands of design variables, and need to compute sensitivities for each of them, if becomes really though.
When you combine external tools, such as HyperStudy with OptiStruct, the main problem go to sensitivities, as even worse, the external solver doesn't have access to sensitivities and need to rely on some global search or finite difference scheme for sensitivities.
Maybe knowing more on what you want to do, would help.
Hi @Adriano A. Koga. Do you have any suggestions regarding what I replied to you?
Thank you!
0 -
gustacsilva said:
Of course.
So, my goal is to perform topology optimization processes considering the nonlinear behavior of fiber-reinforced concrete. In terms of design variables, for now, as is already difficult because of the material and geometric nonlinearity, I'm just using the most common ones, strain energy and volume. Where the goal is to reduce the strain energy for a specific volume. Maybe, later on, I would try to complicate a little more and change or add more design variables but for now, I'm keeping more simple.
The structure, for now, would be a simply supported beam and the load would be applied as pressure or displacement (depending on what is better of convergence).
Many parameters can be changed to achieve convergence, such as types of support, mesh, constitutive law, step parameters, and many others. But, for this case, despite the sensitivities problem, do you think that it can be achieved?
Thank you for the feedback!
Hi,
i would say that if you're looking for some parametric or shape studies, HyperStudy coupled with OptiStruct would be something to test, assuming that you don't have too many DVs.
Now if you go to Topology Optimization, today I don't see much to do in combination with nonlinearities. As NL takes some time to run, I would check if a 2-phase process would fit: 1st optimize for linear, and then check for NL.
0 -
Adriano A. Koga_21884 said:
Hi,
i would say that if you're looking for some parametric or shape studies, HyperStudy coupled with OptiStruct would be something to test, assuming that you don't have too many DVs.
Now if you go to Topology Optimization, today I don't see much to do in combination with nonlinearities. As NL takes some time to run, I would check if a 2-phase process would fit: 1st optimize for linear, and then check for NL.
Yes, for now, I'm trying with only 2 DVs. But shape optimization is used to decrease stress concentration in specific areas of the geometry right? So usually is used after a topology optimization study right?
So today you still have not seen many studies regarding nonlinearities in topology optimization or do you just think that is not worth doing? Sorry, I didn't quite understand what did you meant by "today I don't see much to do in combination with nonlinearities".
Yes, I have already conducted linear topology optimization studies using Abaqus, for the same type of structures (geometry, load and supports conditions), that I mentioned before, and the results were good. Now as I said, I'm trying to perform topology optimization considering the nonlinear behavior of concrete, by using the concrete damaged plasticity model. I already have conducted some processes. Some of them converge others didn't (depending on the optimization and step parameters, as well on the material model). But the ones that did converge, the results are far from being satisfactory...
So, I'm really wondering if is worth trying to implement the same thing in other software, in this case, OptiStruct (which seems to be also very interesting, from what I read). If you want, I can send you the CAE models for you to take a close look.
Maybe combining the software with some external coding, is something that also could work, what do you think?
Thank you!
0 -
gustacsilva said:
Yes, for now, I'm trying with only 2 DVs. But shape optimization is used to decrease stress concentration in specific areas of the geometry right? So usually is used after a topology optimization study right?
So today you still have not seen many studies regarding nonlinearities in topology optimization or do you just think that is not worth doing? Sorry, I didn't quite understand what did you meant by "today I don't see much to do in combination with nonlinearities".
Yes, I have already conducted linear topology optimization studies using Abaqus, for the same type of structures (geometry, load and supports conditions), that I mentioned before, and the results were good. Now as I said, I'm trying to perform topology optimization considering the nonlinear behavior of concrete, by using the concrete damaged plasticity model. I already have conducted some processes. Some of them converge others didn't (depending on the optimization and step parameters, as well on the material model). But the ones that did converge, the results are far from being satisfactory...
So, I'm really wondering if is worth trying to implement the same thing in other software, in this case, OptiStruct (which seems to be also very interesting, from what I read). If you want, I can send you the CAE models for you to take a close look.
Maybe combining the software with some external coding, is something that also could work, what do you think?
Thank you!
i meant OptiStruct still doesn't support combining NL material or NLGEOM with Topology Optimization. There might be something in the roadmap, but not for now.
I believe using external codes could be very time consuming as you don't have sensitivities embed directly on the solver, and you would need to use some approach such as finite differences.
0 -
Adriano A. Koga_21884 said:
i meant OptiStruct still doesn't support combining NL material or NLGEOM with Topology Optimization. There might be something in the roadmap, but not for now.
I believe using external codes could be very time consuming as you don't have sensitivities embed directly on the solver, and you would need to use some approach such as finite differences.
Hello @Adriano A. Koga
Ok, thank you for the clarification. Just out of curiosity, is there an estimation for when that could be available?
Ok. But if it were something implemented on the program scripting interface, that communicates directly with the program, would still have the problems that you aforementioned? Do you know if Altair programs have a python scripting interface or something similar?
Thanks again!
0 -
gustacsilva said:
Hello @Adriano A. Koga
Ok, thank you for the clarification. Just out of curiosity, is there an estimation for when that could be available?
Ok. But if it were something implemented on the program scripting interface, that communicates directly with the program, would still have the problems that you aforementioned? Do you know if Altair programs have a python scripting interface or something similar?
Thanks again!
i don't know what are the plans for this to be there.
But, concerning some customization, are you aware of OptiStruct DRESP3 environment?
It essentially allows you to add custom responses to your model, by using fortran, C, excel or Compose (Altair programming language).
This expands the range fo your responses. With this, you could create your own responses, based on some basic responses or variables. It will use finite differences for sensitivities, OR you could provide sensitivities in your code.
0 -
Adriano A. Koga_21884 said:
i don't know what are the plans for this to be there.
But, concerning some customization, are you aware of OptiStruct DRESP3 environment?
It essentially allows you to add custom responses to your model, by using fortran, C, excel or Compose (Altair programming language).
This expands the range fo your responses. With this, you could create your own responses, based on some basic responses or variables. It will use finite differences for sensitivities, OR you could provide sensitivities in your code.
Hi @Adriano A. Koga,
Ok, thank you!
No, I didn't know the OptiStruct DRESP3 environment, but it seems very interesting! But do you think that could help me to perform nonlinear material topology optimization? By creating nonlinear responses? Is that even possible?
Thanks!
0 -
gustacsilva said:
Hi @Adriano A. Koga,
Ok, thank you!
No, I didn't know the OptiStruct DRESP3 environment, but it seems very interesting! But do you think that could help me to perform nonlinear material topology optimization? By creating nonlinear responses? Is that even possible?
Thanks!
Hi @gustacsilva
Maybe it would be possible to create your responses, BUT i believe it wouldn`t be efficient due to sensitivities (unless you`re able to create an analytical expression for sensitivities).
0 -
Adriano A. Koga_21884 said:
Hi @gustacsilva
Maybe it would be possible to create your responses, BUT i believe it wouldn`t be efficient due to sensitivities (unless you`re able to create an analytical expression for sensitivities).
Ok. Thank you again for the feedback!
Best regards,
Gustavo Silva
0
Altair Employee