simtk.org

The meeting schedule of Cleveland Clinic team for the remainder of year 2006 is listed below. Please feel free to join us by calling the meeting room at (216) 445 0593.

Oct 4, 11:00 AM - 12:00 PM (EDT)
Oct 18, 11:00 AM - 12:00 PM (EDT)
Nov 1, 11:00 AM - 12:00 PM (EDT)
Nov 15, 11:00 AM - 12:00 PM (EDT)
Nov 29, 11:00 AM - 12:00 PM (EDT)
Dec 13, 11:00 AM - 12:00 PM (EDT)
Dec 27, 11:00 AM - 12:00 PM (EDT)

Notes of meeting 10/18/06, 11 AM
CCF, Room ND2-43

Present: Ahmet, Al, Xuemei, Ton

1. Ahmet has created a SimTk forum "Organizational". This is where we will post meeting notes (such as those you are reading now). Other SimTk forums could be started for scientific discussions of work that involves Stanford or Utah.

2. Ahmet and Xuemei have installed the MoinMoin (Wiki) server at CCF which will be our platform for discussing algorithms and models, for the work done at CCF. Collaborators can participate here with VPN access but we will do collaborative discussions as much as possible on the SimTk.org forum. We will see how that goes.

3. The server also hosts the Trac system for development of software and models within the CCF team. Similar functionality exists on SimTk, and the SimTk system can be used for collaborative work.

4. MoinMoin and Trac will both have an e-mail notification system which can send an e-mail whenever a document has changed.

5. Trac will have a directory structure where we can store software (multibody modeling, FE modeling, surrogate modeling) and applications/models. Ahmet will set up that structure. An SVN client must be used to access this.

6. We have a 2D FE model of the foot which is suitable for coupling to a 2D gait model. The foot model will be driven by position and orientation of the calcaneus, compute the corresponding reaction loads, which will then be applied to the multibody model. This will be our first application which allows us to prototype the entire framework for multidomain coupling. Ton will start putting the musculoskeletal model and software on Trac, Xuemei will maintain these, Ahmet will do the same with the 2D foot model, and Al will work on putting it all together into the application.

7. We identified a fundamental issue that we had not thought about. The original idea was that the FE model would compute reaction loads as a function of generalized coordinates of bones: Q = f(q). For the 2D foot model, the input would be position (x,y) and orientation of the calcaneus (theta), and the reaction loads would be Fx, Fy, and moment coming from the ground. However, because of translational invariance, x should have no effect on Fx. Rather, because of friction between foot and ground, Fx will depend on the history of x. Since the surrogate modeling algorithm will not run FE simulations in a temporal sequence, this is a problem. We should therefore use dx/dt rather than x as an input to the FE model. Ahmet will see if this is possible. With the coulomb friction model, Fx will be very sensitive to dx/dt when dx/dt is near zero but this should be OK. The surrogate modeling algorithm will detect the sensitivity and automatically sample the FE model very densely around zero velocity. Which is where it will be most of the time because the foot will stick to the ground.

8. Postdoc recruitment. Jiang Yao is coming for an interview tomorrow (10/19/06). She will be giving a presentation and have individual meetings. She will be available in January.

Notes of meeting 11/01/06, 11 AM
CCF, Room ND2-43

Present: Ahmet, Al, Xuemei, Pavana, Ton

1. Ahmet presented an overview of all SimTk features (www.simtk.org). We are currently using the forum and documentation features. Documentation will be material of general interest, for instance, currently there is a link to the Software Carpentry article. There was a question what the SimTk policy is for posting PDFs of journal publications which are copyrighted. The forum is used for communication with collaborators outside CCF and should be monitored by all team members. There was a question whether forums are searchable. We looked at all other SimTk features to make sure we are aware of those.

2. Ahmet presented an overview of our internal Trac and MoinMoin servers (http://thrace.lerner.ccf.org). Trac is for software development. MoinMoin for scientific content. Both have a Wiki but the Trac is not as good for organized content development. Ideally if SimTk had a Wiki with MoinMoin capabilities, we would use it and this would eliminate the firewall problem for outside collaborators. Once we can demonstrate how we use MoinMoin, we may try to convince SimTk to install the same system.

3. Ahmet demonstrated the milestone and ticket features of Trac. For example he has created a milestone for creating the software repository, with two tickets within that milestone. Ton will complete the initial design of the repository so we can start collaborating on the SDBIO library and 2D gait/foot model.

4. Ahmet demonstrated the MoinMoin system. There are templates and categories to help maintain organization of the content. A bottom-up way to generate content is to click on "CreateEntry", and then choose a category and corresponding template. For instance: project, article, abstract. Top-down organization can then be added at a later stage simply by typing the name of such new pages within a higher level document (e.g. the home page of the project). We are planning to use the "article" document type to co-author our first article and see how that goes. Ahmet will install LaTeX for rendering of equations.

5. Ahmet made a MoinMoin page HistoryDependenceFEA (it is posted as PDF on SimTk under Documentation). He solved the 2D foot model five times for the same kinematic state (x,y,theta), starting from different initial states. This showed a strong history dependence, which is a problem for surrogate modeling where reaction loads must be a function of kinematic state only. The history dependence is physically correct when there is dry friction: it makes a large difference whether we push the foot down and then rotate, or first rotate (while in the air) and then push down. There are two possible ways to resolve this issue. The first is to define state as (xdot,y,theta) and use a viscous approximation to dry friction. The second option is to define state as (xprevious,yprevious,thetaprevious,x,y,theta), i.e. a doubling of dimension to account for history. One FE simulation will generate multiple 6D states, because all intermediate 3D states can also be considered as initial 3D states, all leading to the same final 3D state. This may work and would require no changes to the FE modeling. We will try this second option first. I will update the MoinMoin page accordingly (and probably post another PDF on SimTk...)

6. Organizational. Tomorrow (11/2/06) we will interview Jason Halloran for the postdoc position. We also received a very strong application from a French/Swiss computer scientist who works in human animation. He may be very suitable for musculoskeletal and optimization. The plan is to hire two postdocs, one with a FE focus and one with a movement focus.

7. Ton and Al have communicated about ODE solvers. Ton will summarize on MoinMoin. Al has previously developed efficient solvers for stiff systems that may avoid the high cost of Jacobian calculation.

8. We will have a conference call with the entire team (CCF, NASA, Stanford, Utah) on Nov 9th. Instructions are on the SimTk forum. Ahmet will investigate whether we can do some sort of webcast so we can share a view of our computer screen. Otherwise we will just prepare documents ahead of time and share them via SimTk.