Salut, mes amies! Please note: My blog posts are lagging about three weeks behind current events. I need to do this faster! Well, ready to find out about more adventures in the City of Light? Ok.
So yeah, that's work! Enough about the nerdy stuff, let's talk about the delicious stuff! That also happens often, when someone has a "pot". A "pot" is what the French in my group call a small gathering with food. Basically MIT has a lot of those. Whenever someone in our group has something to celebrate (like a birth of a child, a marriage, etc), one sends out an email to everyone and gathers them together over a small (or large) feast. This is also so for when someone enters the group or leaves it. The first "pot" I attended was actually like a small party, with wine, fruits, cheese, delicious chocolate, and other yummy things. The boss of the entire group read a toast for the worker who was leaving EDF, people smiled, laughed, had a good time.
Let's start with work; it takes up most of my week, after all. It is progressing well, I think, although I feel like it could be going better. I am basically trying to run simulations with this code, and find how it can be improved to decrease CPU time. I am simulating damage on concrete. There are two ways to do this, in general. One is to try to take the continuous model and derive an analytical model with an energy minimization method (let's face it, it is always about energy, because it makes things easier!). This model is very theoretical and very complicated when the sample is more than just 1D. That is why we are using Method #2, that is, finite elements analysis (FEM). This means we just break apart the continuous pieces, triangles or tetrahedrons (a mesh) which are connected by nodes. One can apply a load on these nodes and get values of stress, strain, damage, etc, using the program my company designed. In the old days one had to do draw out each finite element and carry out these calculations manually, but now the program just does it for you. But we want to go even further than just simple stress calculations and actually analyze how the damage is occurring, to perhaps even be able to predict where the damage will occur and under what load. So, we want to refine the region where the damage is; break the elements into smaller and smaller pieces. This can be done by hand after damage is observed, or this can also be done automatically. This is what I am doing. I was given a piece of code which I have to work with to apply to different models of concrete, and eventually graphite. This code does automatic remeshing on these models, which is very efficient and useful in industrial applications, such as testing the resulting damage profile of concrete block used for a nuclear power plant container. It is not very easy to run, however, because it is super finicky! It uses a numerical method similar to Newton, cutting up the elements and displacements based on the projected damage field in the previous iterations. There are two main problems:
1) Convergence
2) Run time
Convergence is usually the main problem. If somehow the criteria for refinement is made too large and the steps are too small, the program will keep cutting away. Or if the steps are too big and you overstep the maximum cut, there is another error. Run time is also a problem. It takes about 3 minutes or less to apply a loading to a sample without remeshing, and like half an hour or more to do three refinements using the automatic mesh.
So right now I tried to fix the lack of convergence when the damage region is large, by applying more damage after several refinements have already happened. This trick still did not work. I have a few ideas though, so I'll try tomorrow and see if it works. Wish me luck!
Oh, and on the topic of work, on Friday I went to LMT Cachan, a research site of EDF, with my boss and coworkers (grad students) to see the hexapod. This is what the haxapod looks like:
Me, the group director, and my supervisor with hexapod |
No, that is not Wall-E! That is a hexapod that applies a very* precise load on a sample. Here they are doing the study with a specific sample of concrete, which is very well known to me by now, as I had done a few preliminary simulations on it out of curiosity. Theoretically it should have a very unique crack path. However, when they did the experiment the concrete mostly followed one of the crack paths and then suddenly broke in two. Oops. So my boss actually thought that was really cool and brought back one of the broken samples to his office. It is a good idea, I think; a nice reminder that the computer simulations we are doing actually do relate to real life! But the asterisk next to "very" is there for a reason; even though the hexapod is super cool and moves its eight arms to apply the correct displacement, there is still lots of error. (Well, on a micron scale, but still error that matters when your displacement is also a few microns). Basically it is hard for them to move the sample after it has been already placed, and also difficult to calibrate the sensors that actually measure the displacement -- for this reason. Oh, these sensors are actually pretty cool. They are super good at measuring lateral movement, they are basically little transformers where the tiny displacement induces a magnetic field, and this sends an electric signal to the receiver. Quite cool. All these sensors and also cameras mounted to the aluminum are connected to three computers, on which you can observe its variation with time in real time. The displacement itself is also programmed with code.
Andreea, the grad student working here, is explaining how data is collected |
The "pot" |
Other pots can be less global, involving some desserts. A grad student who just finished his internship brought in some delicious cheese eclairs that he made himself and other pastries, and we just had them in our small coffee lounge. Since it was the Fourth of July recently, some of my coworkers are saying that I owe them gâteau, especially if it's carrot cake ("American specialty"). Apparently there are not a lot of carrot cakes in France! However, I am pretty satisfied with the cake options in my cafeteria. :)
As part of my adaptation to the work space, I had a few impromptu ping pong matches. Sometimes I even faced my coworkers from China (you know what this means in ping pong?) and yet stood my ground pretty well. Both sides were surprised. Next to the cafeteria at EDF there is also a music room where you can hear the trumpet or piano while you play ping pong, as well as a video room where one can rent DVD's.
Weekends, once again, are relaxing and full of discovery. That weekend I tried to go to an open-air market, but instead found a four-day International Arts Festival, which exhibited things like funky stained glass, highly adorned carpets, Mexican paintings, antique clocks, goldwork and antique harpsichords. These harpsichords were really cool; the seller even played several baroque pieces by Couperin for me, and showed me the mechanism behind the keys, because I seemed to be very interested in how it works. Despite my limited French in this area, I learned that the harpsichord works almost exactly like a guitar, except it is horizontal and there are keys.
1700's harpsichord. Photo (c) Danica Chang |
Trying on a real Russian headdress! |
An actress from the Moscow Medieval Theater... in 30 degrees C! Photo (c) Danica Chang |
Content with seeing the tower from the Trocadero, I thought I was going to head to see this freaking Arc de Triomphe which I kept not getting to. I thought. Somehow I ended up strolling into a Greek-looking building that promised some theater plays, or something. Guess where I found myself? Of course, another museum. Me and museums, it's almost like magnetism!
Cité de l'architecture et du patrimoine! |
This museum is un bijoux! |
Then I actually headed to meet a friend to go to the MIT Paris Club cocktail party at an alumnae's really beautiful apartment by the Seine. On my way there, I stumbled a really cute summer market by the Seine.
The cocktail party was very fun, and the view from the apartment was so beautiful! It was really interesting to share experiences with other MIT people, and also talk with some local students about their expectations and thoughts on entering MIT master's programs in the fall.
The next day, on Sunday, we headed to explore a new market on Rue Raspail, but it was organic and waay above our budget, so we went to see the one on Boulevard de Grenelle, where I was the day before, and got some more of those yummy strawberries. Seriously, if there was one thing worth going to the market for, it's these amazing French strawberries! Anyway, from this market we took the metro to Bastille, from which we easily found the most amazing "street" in the world -- Promenade plantée, a blossoming street above street level, from which you can almost see the roofs of buildings, and is completely covered with flowers and plants, even with bamboo! A good perspective of this Promenade (the first elevated park in the world!) can be found online:
It was seriously amazing to walk through this street of flowers, and then find yourself at a huge "sunbathing station" under a bridge.
It was seriously amazing to walk through this street of flowers, and then find yourself at a huge "sunbathing station" under a bridge.
Sunbathing is popular! Photo (c) Danica Chang |
After, the Promenade continues until the huge forest/park, Bois de Vincennes, which I was sure became my "new favorite place in Paris". Why? This will come in a new blog entry about the amazing "green places" in Paris! :) Because it is one of the things I so love about this city. Sneak peak? Here's a photo of me in one of the little grotto's of the lake at the Bois, where we also sailed a boat!
Photo (c) Danica Chang |
you know that just because professional Chinese ping pong players work super hard and play really well doesn't mean regular Chinese people are supposed to be good at ping pong, right?
ReplyDeleteWell, I don't know. I just posed a question. :)
DeleteWoot FE methods! Good luck in fixing the convergence issues. =)
ReplyDeleteQuestion about the run time issue - are the matrices involved dense or are they ever sparse? I'm trying to remember the material equations and I know some reduced order models yield sparse matrices ...
They are sparse, since the method we are using takes the K stiffness matrik from Ku=f and diagonalizes it into LLt, LDLt, or LU depending on the type of matrix K is, producing all sparce matrices. :)
DeleteAnd yes, FEM is super cool, so happy you appreciate it too! :D
This comment has been removed by the author.
Deleteare they being stored in a sparse format?
Delete