I'll be at the 7th International Workshop on Direct and Inverse Problems in Piezoelectricit, October 4th - 7th, in Oer-Erkenschwick, Germany. It is a very small town but Moondog lived there for some years. I was at the 5th and 6th Piezo Workshop. It is a small event in a very comfortable atmosphere with very easy exchange to other people. I learned a lot about piezoelectric material there.
Following is the EUROMECH522 "Recent Trends in Optimisation for Computational Solid Mechanics" October 10th-13th in Erlangen. It is an colloquium with invited participants only (as far as I understand). I hope they accept my abstract. I want to talk about self-penalization in topology optimization. I have some new results/observations in addition to the paper On the effect of self-penalization of piezoelectric composites in topology optimization. The dynamic stuff is still an open issue and I have not explored some ideas I had. In case I'll get accepted to EUROMECH522, I'm quite curious about it as I haven not been on such an colloquium before. I hope many Danish topology optimizers come, they are the reference.
For private reasons I won't come to WCSMO-9 :( That is sad as WCSMO-8 was the most important conference for me up to now.
Saturday, 26 February 2011
Thursday, 4 November 2010
Stress Constraints
I started to work with stress constraints. It's pretty interesting and up to now more stuff seems unclear to me that being clarified.
I read a bunch of papers, but need to read the original Duysinx and Bendsoe; Topology optimization of continuum structures with local stress constraints; 1998 paper again (maybe there are the answers to my questions hidden :)). Our results are in good agreement with the standard solutions, so our approach seems to work. However, the results of Le; Stress-based topology optimization for continua; 2010 are really extraordinary good, unmatched by any other publication I read and far better that our results.
I still wonder, why
min vol
s.th. stress constraint
works. At least for qp-regularization (Bruggi; On an alternative approach to stress constraints relaxation in topology optimization; 2008) rho_min should be the global optimum. But the results are all similar to the compliance solution. I cannot imagine, that the regularization just hides this global optimum as we use a second order optimizer ... well, I hope I'll find out.
The next thing is, that beside Le, all (of what I read up to now) published results show some grayness embedded in full material. What happens if we have to map to manufacturable black and white topologies? I have not found any discussion of this grayness yet.
I read a bunch of papers, but need to read the original Duysinx and Bendsoe; Topology optimization of continuum structures with local stress constraints; 1998 paper again (maybe there are the answers to my questions hidden :)). Our results are in good agreement with the standard solutions, so our approach seems to work. However, the results of Le; Stress-based topology optimization for continua; 2010 are really extraordinary good, unmatched by any other publication I read and far better that our results.
I still wonder, why
min vol
s.th. stress constraint
works. At least for qp-regularization (Bruggi; On an alternative approach to stress constraints relaxation in topology optimization; 2008) rho_min should be the global optimum. But the results are all similar to the compliance solution. I cannot imagine, that the regularization just hides this global optimum as we use a second order optimizer ... well, I hope I'll find out.
The next thing is, that beside Le, all (of what I read up to now) published results show some grayness embedded in full material. What happens if we have to map to manufacturable black and white topologies? I have not found any discussion of this grayness yet.
Wednesday, 13 October 2010
Paper on Piezoelectric Self-Penalization Published
Our second paper is published (online):
F. Wein, M. Kaltenbacher, B. Kaltenbacher, G. Leugering, E. Baensch and F. Schury; On the effect of self-penalization of piezoelectric composites in topology optimization; Structural and Multidisciplinary Optimization; DOI: 10.1007/s00158-010-0570-2
springerlink
It covers the effect of self-penalization of many piezoelectric topology optimization problems. That means that no penalization, volume constraint and regularization is applied to gain often mesh independent black and white solutions. It's not a method we describe but a phenomenon, so to report when it does not work is part of the job.
I guess we implemented almost all relevant piezoelectric objective functions and to compare them based on the same model is IMHO also of interest. The objective functions are mean transduction, displacement, sound power, electric potential, electric energy, energy conversion and electric power.
We give an explanation based on the piezoelectric model: stiffness, electrostatic and piezoelectric coupling. More density is bad with respect to stiffness but good for the other fields, opposite for less density. That means that there is balance of the effects and when this balance is outside the feasible design set (0,1], there is no greyness.
Self-penalization has been, to my knowledge, first reported by Ole Sigmund and Jakob S. Jensen 2003 with (elastic) wave guiding. It also appears in other, especially non-elastic problems but has never been investigated. Meanwhile I almost belief that probably most non-compliance problems show some form of self-penalization. I hope I can find a general explanation as this might eventually lead to new method in topology optimization. :) I have at least one idea on how to proceed ... Let's see when I find the time and if I have success.
P.S.: It seems that the term "self-penalization" has not been used before in the context of topology optimization (I actually don't want to know what google finds for self-penalization in other contexts :) ) but I learned the expression from Ole Sigmund, so he deserves the credits.
F. Wein, M. Kaltenbacher, B. Kaltenbacher, G. Leugering, E. Baensch and F. Schury; On the effect of self-penalization of piezoelectric composites in topology optimization; Structural and Multidisciplinary Optimization; DOI: 10.1007/s00158-010-0570-2
springerlink
It covers the effect of self-penalization of many piezoelectric topology optimization problems. That means that no penalization, volume constraint and regularization is applied to gain often mesh independent black and white solutions. It's not a method we describe but a phenomenon, so to report when it does not work is part of the job.
I guess we implemented almost all relevant piezoelectric objective functions and to compare them based on the same model is IMHO also of interest. The objective functions are mean transduction, displacement, sound power, electric potential, electric energy, energy conversion and electric power.
We give an explanation based on the piezoelectric model: stiffness, electrostatic and piezoelectric coupling. More density is bad with respect to stiffness but good for the other fields, opposite for less density. That means that there is balance of the effects and when this balance is outside the feasible design set (0,1], there is no greyness.
Self-penalization has been, to my knowledge, first reported by Ole Sigmund and Jakob S. Jensen 2003 with (elastic) wave guiding. It also appears in other, especially non-elastic problems but has never been investigated. Meanwhile I almost belief that probably most non-compliance problems show some form of self-penalization. I hope I can find a general explanation as this might eventually lead to new method in topology optimization. :) I have at least one idea on how to proceed ... Let's see when I find the time and if I have success.
P.S.: It seems that the term "self-penalization" has not been used before in the context of topology optimization (I actually don't want to know what google finds for self-penalization in other contexts :) ) but I learned the expression from Ole Sigmund, so he deserves the credits.
Wednesday, 6 October 2010
Mean Transduction
In 1999, Silva et al. introduced in
Silva, Nishiwaki, Kikuchi; Design of piezocomposite materials and piezoelectric transducers using topology optimization—Part II; Archives of Computational Methods in Engineering; 1999
the objective function mean transduction. It describes the piezoelectric coupling, and it is quite intuitive to maximize that coupling. The mean transduction was applied quite often by Silva. To my knowledge there are no recent publications, where it is used, but I might be wrong.
I don't use the function but as I started 4 years ago with it, it is still in my mind and somehow I still try to actually understand it properly.
I make currently good progress in writing my thesis and just work on a section about mean transduction. I finally found the link between mean transduction and classical adjoint based sensitivy analysis - which clearly helps in understanding the nature of the mean transduction. Just, I don't know if this is of interest, as I don't know, if the mean transduction is used any more. For sensor applications, maximizing the electric power as done by Cory Rupp is IMHO more practical - and for actor applications the physics might be necessary, e.g. acoustic formulations to prevent acoustic short circuits as I have shown.
This piezo stuff is really heavy (optimization) stuff :)
Silva, Nishiwaki, Kikuchi; Design of piezocomposite materials and piezoelectric transducers using topology optimization—Part II; Archives of Computational Methods in Engineering; 1999
the objective function mean transduction. It describes the piezoelectric coupling, and it is quite intuitive to maximize that coupling. The mean transduction was applied quite often by Silva. To my knowledge there are no recent publications, where it is used, but I might be wrong.
I don't use the function but as I started 4 years ago with it, it is still in my mind and somehow I still try to actually understand it properly.
I make currently good progress in writing my thesis and just work on a section about mean transduction. I finally found the link between mean transduction and classical adjoint based sensitivy analysis - which clearly helps in understanding the nature of the mean transduction. Just, I don't know if this is of interest, as I don't know, if the mean transduction is used any more. For sensor applications, maximizing the electric power as done by Cory Rupp is IMHO more practical - and for actor applications the physics might be necessary, e.g. acoustic formulations to prevent acoustic short circuits as I have shown.
This piezo stuff is really heavy (optimization) stuff :)
Sunday, 30 May 2010
Applet visualizing square plate vibrations
I found this very cool applet from Paul Falstad. There are many more technical applets on his homepage.
The applet allows to interactively play with the structural resonance modes of an elastic plate. I also like the transient simulation when the plate is "pinged" by the mouse.
Paul made cool applets for several physical problems, worth to have look :)
The applet allows to interactively play with the structural resonance modes of an elastic plate. I also like the transient simulation when the plate is "pinged" by the mouse.
 |
| Von Pizeo Blog |
Paul made cool applets for several physical problems, worth to have look :)
Sunday, 23 May 2010
Acoustic near field topology optimization
Here I share my talk I gave on ECCM 2010. It presents acoustic near field optimization. Last year, at WCSMO-8 I compared acoustic topology optimization where the acoustic field was assumed to be far field/ plane wave type/ constant acoustic impedance at the structure itself (so no acoustic is calculated) and at the boundary of by acoustic domain. Now I did the step further and optimize the acoustic power directly w/o any assumption. Thanks to Michael Stingl who helped with some very useful comments! It turned out, that the same objective function had already been used in Jensen and Sigmund; Topology optimization of photonic crystal structures: a high-bandwidth low-loss T-junction waveguide; 2005. It was disappointing that it had (for other physics) already been done - but it would have been really bad if I had not found it and claimed it as a own result.
For my numerical examples the optimal results where quite close to the acoustic far field optimization. This was quite surprising but before I could for lower frequencies not guarantee any physical validity. So it was definitely worth it.
For my numerical examples the optimal results where quite close to the acoustic far field optimization. This was quite surprising but before I could for lower frequencies not guarantee any physical validity. So it was definitely worth it.
ECCM 2010 is over
I just came home from ECCM 2010 (European Conference on Computational Mechanics) in Paris. I was invited by Ole Sigmund to the multiphysics topology optimization minisymposium (I'm proud of it, therefore I mention it all the time :)). I gave a talk on Acoustic near field topology optimization of a piezoelectric loudspeaker, see my other blog post about it. I also contributed to the talk my collegue Fabian Schury gave about Design of isotropic lightweight material structures by inverse homogenization and topology optimization. What he actually showed is, that an old regularization method, the slope constraints (Petersson and Sigmund; Slope constrained topology optimization; 1998), is actually quite useful when used with appropriate optimizers. The standard MMA implementations simply cannot handle the many thousand constraint problem (we have up to 400.000!) but e.g. SnOpt can do it and we are working on other optimizers within our group.
Ole Sigmund gave a semi-plenary talk with a variation of the robust optimization presentend last year at WCSMO-8 but now including heaviside type density filters. It seems that this finally really solves the problem of the resolution of hinges in mechanism design (e.g. force inverter). The details of the hinges are a long known problem - yet often ignored. The optimizer abuses the numerics of low order linear elasticity and bends elements ("bricks") only by their momentum free nodal or edge connection. I actually wonder if this could be resolved by stress constraints (I known they have their own problems)? I would say that this special area of length-scale control is one of the few areas where significant contribution to the SIMP method can still be done! It would be very cool if our group (the one of Michael Stingl) could come up with a new approach :)
Again the talks from the Danish guys where advanced and interesting (DTU *and* Aalborg), including a talk on transient optimization to reduce the speed of light with several thousand time steps of "adjoint history" by Rene Matzen. More related to our own work was Optimization of the pressure coupling coefficient in periodic poroelastic materials by Casper Andreasen.
At least at ECCM I noticed no further strong topology optimization group. Apart from the minisymposium I heard not that much interesting talks. A lot turned out in using genetic algorithms or other stochastic methods and one cannot really learn much from it. It's a "This is our interesting problem, we put a black box over it, show some colorful graphs labeled A...H because we do science and now we come to the conclusion." Nice for the guy who solved his existing problem but no new idea I can take home. I wish they would clearly identify this genetic stuff talks - especially as there where about 40-50 talks in parallel!! and I often heard just the wrong one.
Paris is a nice place, I like all the open places where people gather and street artists give performances. We had a really nice little hotel, Hotel Helipolis at a good location and a very nice vibrations for 77 Euro a single room.
I'm looking forward to WCSMO-9 2011 in Japan!
Ole Sigmund gave a semi-plenary talk with a variation of the robust optimization presentend last year at WCSMO-8 but now including heaviside type density filters. It seems that this finally really solves the problem of the resolution of hinges in mechanism design (e.g. force inverter). The details of the hinges are a long known problem - yet often ignored. The optimizer abuses the numerics of low order linear elasticity and bends elements ("bricks") only by their momentum free nodal or edge connection. I actually wonder if this could be resolved by stress constraints (I known they have their own problems)? I would say that this special area of length-scale control is one of the few areas where significant contribution to the SIMP method can still be done! It would be very cool if our group (the one of Michael Stingl) could come up with a new approach :)
Again the talks from the Danish guys where advanced and interesting (DTU *and* Aalborg), including a talk on transient optimization to reduce the speed of light with several thousand time steps of "adjoint history" by Rene Matzen. More related to our own work was Optimization of the pressure coupling coefficient in periodic poroelastic materials by Casper Andreasen.
At least at ECCM I noticed no further strong topology optimization group. Apart from the minisymposium I heard not that much interesting talks. A lot turned out in using genetic algorithms or other stochastic methods and one cannot really learn much from it. It's a "This is our interesting problem, we put a black box over it, show some colorful graphs labeled A...H because we do science and now we come to the conclusion." Nice for the guy who solved his existing problem but no new idea I can take home. I wish they would clearly identify this genetic stuff talks - especially as there where about 40-50 talks in parallel!! and I often heard just the wrong one.
Paris is a nice place, I like all the open places where people gather and street artists give performances. We had a really nice little hotel, Hotel Helipolis at a good location and a very nice vibrations for 77 Euro a single room.
I'm looking forward to WCSMO-9 2011 in Japan!
Subscribe to:
Comments (Atom)