Copying to libmesh-devel in case anyone else wants to interject. On Wed, 28 May 2014, Robert Weidlich wrote:
> My aim is to solve standard solid mechanic tasks with contact, so I > usually need a material and some kind of boundary conditions. In my > current implementation I have one subclass of FEMSystem and own class > hierachies for materials (NeoHooke, Ogden, etc) and boundary conditions > (displacement, traction, standard contact, mortar contact, etc). > Material and boundary conditions are instanciated and called in my > FEMSystem subclass. > > I saw that you factored out DifferentiablePhysics and FEMPhysics some > time ago. Now I am wondering if I can use this to simplify my code. It > is quite obvious that I can just subcalss FEMPhysics to implement e.g. > NeoHooke material and attach that class to the FEMSystem. But as I can > only attach one FEMPhysics class I wonder what is the best place for > e.g. contact code? The original goal with the refactoring was to add a "SumPhysics" subclass which would be attached to the FEMSystem and which would in turn allow an arbitrary number of other DiffPhysics subclasses to be attached to it. I got sucked into other things before I could finish that, but I'd be happy to write the code later this week if I could talk you into doing the testing. > The other thing which is unclear to me are the QoI. Can it be used for > projecting the stresses from quadrature points to nodes in > postprocessing? Currently I am doing this in a seperate system. At the moment, all the library code using the QoIs assumes that your final result is a small-enough-to-serialize set of scalars. You could use the QoI assembly as a place to hook in other postprocessing, but if you don't have those scalars then you don't have any need for the library-level tricks with them (sensitivities, goal-oriented adaptivity) and so there's probably no benefit to refactoring your code to use them. But postprocessing isn't the best place to project stresses either, since that's outside the nonlinear solver loop and so limits you to pseudo-transient solves, right? I wonder if I should add a default-noop FEMPhysics::preassembly()/postassembly() method? --- Roy ------------------------------------------------------------------------------ Time is money. Stop wasting it! Get your web API in 5 minutes. www.restlet.com/download http://p.sf.net/sfu/restlet _______________________________________________ Libmesh-devel mailing list Libmesh-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-devel
