WCSMO-10 is over

WCSMO-10 is over, and again I realized, that the WCSMO is by far the most interesting conference for me. The whole numerical topology optmization community was there and I enjoyed it very much. I have the impression that I meanwhile know a lot of the important people (at least a little :)).


I held a talk on piezoelectric free material optimization, similar to what I talked about on the last Workshop on Direct and Inverse Problems in Piezoelectricity, but this time not starting with the music of Jean Michel Jarre :)

Free Material Optimization of Piezoelectric Material

People liked our tensor stiffness visualization, maybe I should put the Python code online ...

I also had a poster at the poster session about self-penalization. The content is from a former talk and can be also found in my thesis. My impression is, that private and direct communication/ discussion is best for the topic. The poster can also be downloaded.

This time Jamie Guest was almost omnipresent with a very large group. He appeared on 13 talks as (co-)author compared to 8 talks (co-)authored by Ole Sigmund. Well, numbers are one thing but it's impressive.

My favorite talk was by Alexander Verbart from the group of Fred van Keulen; A new approach for stress-based topology optimization: internal stress penalization. I did a little on stress optimization (e.g. piezoelectric stress optimization) but for me there are two open problems: have not grayness to confirm to the constraints and the miracle of the min vol problem which would be best for rho = rho_min. The idea of Alexander has neither stress penalization nor globalization parameters, subregions or anything else. The approach would allow gray material to confirm with the stresses but with some luck the optimizer won't find this solution. I'll have to give it a try.

The new TopOpt App from DTU

Ole Sigmund's DTU topology optimization team released their TopOpt App for iOS, Android and Web. This is pretty cool stuff. It allows to play with any variant of a compliance problem. With respect to the original TopOpt web application it is interactive and as such it is as interactive as our our iTop web application (eamc080.eam.uni-erlangen.de/iTop. IMHO, the TopOpt app suits better a tablet than a smartphone/ iPod, but on a tablet it is really interactive, and it is real fun to play with it. I let my 5 years old daughter play with the TopOpt app what would not work with iTop.

With the TopOpt app being an off-line app, I started to show almost any person (I have a beer with and who is not escaping fast enough) what this topology optimization stuff is all about on my iPod.

From the technical point of view I'm impressed by the performance. It is clear, that despite their user experience, the ARM processing power is significant slower than our iTop Xenon CPU. I would have guessed that the performance wouldn't be sufficient and I would have been wrong! I wonder about the direct solver and the Heaviside filter parameters.

So what it the difference to iTop? The TopOpt app is self-explaining while iTop deserves explanation (especially as we still did not write any documentation). But using iTop as a demonstrator tool it allows to explain what is behind

• penalization,
• regularization.
• choice of optimizer,
• the phenomenon of self-penalization,
• start designs/ local optima.

I'm pretty sure the force inverter problem will be soon integrated into the TopOpt app, while interactive material design via inverse homogenization might be a unique feature of iTop for a longer time.

Congratulations, Niels, Morten and Casper! :)

Thesis published

Finished! :)

On the 19th of December 2011 I had the defence of my thesis (Dr. Ing.).



I published my thesis online. You can access it at UB Erlangen-Nuermberg.

The thesis

Topology Optimization of Smart Piezoelectric Transducers

has three main topics:

• topology optimization of a piezoelectric loupspeaker
• topology optimization of a piezoelectric cantilevered energy harvester
• self-penalization in topology optimization

Thank's a lot to my colleagues for the great hat!! :)

I will continue as PostDoc at my current position, working for Prof. Stingl at the chair of applied mathematics from Prof. Leugering.

Interactive Self-Penalization

I added mechanism design to our iTop tool, see also my last post. This allows to play with the force inverter problem. Volume constraint bounds can be switched, regularization can be disabled by setting "reg. value" to one. It makes sense to play with "start optimization from current design". Local solutions are really a problem. Sometimes it helps to play with the physical lower bound (rho^p) and penalization values.

Clearly, the results are not necessarily more "desirable" than the volume constraint results. But IMHO this is not the point. Here not the practical relevance of the result is interesting but the strong self-penalizing effect.

BTW, I finished my PhD! :) More to write about this next week when I officially publish my thesis (online).

iTop - interactive topology optimization

Our work group developed an interactive topology optimization web client.

eamc080.eam.uni-erlangen.de/iTop

This is very similar to topopt from Ole Sigmund's group. While topopt demonstrates the possibilities of topology optimization (including mechanism design and 3D compliance), we somehow have a more educational focus.

With out interactive topology optimization (iTop) you can play with the regularization and optimizer. Furthermore we include 2D material design where e.g. auxetic material (negative Poisson's ratio) can be designed.

A good example is SNOPT as optimizer with slope constraints as regularization. Both are rather slow for compliance problems but for very challenging material design problems this might be the only possible choice.

What I often show is the non-convex property of penalized compliance. Enable "start optimization from current design" and "restart new optimization on parameter change".
For sufficient penalization, moving the force point might result in a clearly poor local solution. Now taking away the penalization (switch to 1) and activate it again (e.g. 3) the "global optimal" solution is found.

For several parameter sets optimization might fail. By this we want to demonstrate, that optimization is not a black-box problem!

The documentation and all the references are still missing, therefore we do not link to iTop officially from our chair's web page. We hope we can provide some documentation soon.

While being based on our optimization software, iTop is a demonstrator tool only. This is not necessarily the kind of problems we solve in our academic work. iTop contains just a small selection of our tools.

Enjoy and ask if you have any questions.

eamc080.eam.uni-erlangen.de/iTop

Self-Penalization in Topology Optimization

I gave a talk at EUROMECH 522 in Erlangen on the 10th of October 2011.

It appears that most people from the audience did not understand what I wanted to say - so clearly a bad presentation of mine! :(

However, discussions during the conference after the presentation helped me point out the missing items and emphasis in my presentation.

Therefore I provide here a variant of the talk with some additional slides. If you are interested in the subject I'll be very happy to discuss it with you! :)

The key point is, that I do not present anything like a method or something big. I just want to motivate to have a slightly different point of view on our daily topology optimization problems. (But to be honest, to my humble personal opinion this off the mainstream look on topology optimization might result in important insights :))

Self-Penalization in Topology Optimization

View more presentations from fwein

Thesis Submitted

I finally submitted my thesis! :) I guess the defense will be in September or October. Afterwards I will publish my thesis online for free download. The title is "Topology Optimization of Smart Piezoelectric Transducers"

It basically consists of three parts:

* Topology optimization of piezoelectric actors, with single-frequency loudspeakers in mind.
* Topology optimization of piezoelectric cantilever type energy harvesters
* Self-penalization in topology optimization

For the harvester I applied piezoelectric dynamic stress constraints, which were actually easier to handle than in elasticity as I do right now :)

The self-penalization stuff goes much further than covered in the piezoelectric self-penalization paper.

The introductory stuff became quite extensive. When the thesis is published my plan is to compile all the linear elasticity stuff to an "Introduction to ersatz material topology optimization in linear elasticity", which might serve as what the title suggests. :) The idea is to publish it as technical report and provide free access to it.

I'm happy that writing the thesis is over! My plans for the future are not settled but it looks like that I stay for some more time in academia. ;)