Why are open source FEA tools so far behind commercial FEA tools? 8

[ParabolicTet]

I began using I-DEAS back in 1997 . Nearly a quarter century later we still do not have a free open source FEA package that is comparable to that 1997 I-DEAS. Back in 1997, I-DEAS had a huge list of cool feature, like non-linear analysis, contact elements, powerful geometry and meshing tools, built in direct and iterative solvers etc.. And it offered all this via a unified user-friendly interface. No need to export decks , edit them and visualize in some other tool ( as we have to do now with open source pacakges ).

So how come now nothing out there can come close to that ? FEA is supposed to be one of the oldest and mature disciplines in computer history.
I know people are going to reply because FEA is complex and no one is going to work for free. That makes sense . However, other disciplines like AI, data science, genomics, chemistry etc. are loaded with sophisticated open source packages. So why are those folks working for "free"?

 

[FEA way]

It’s not only a problem with FEA, pretty much the same applies to CAD - the only software worth attention in this field is FreeCAD which has quite advanced capabilities but is missing many convenient functions, suffers from significant issues (infamous topology naming problem) and has very outdated GUI.

When it comes to FEA, the best compromise between computational capabilities and user-friendliness is offered by CalculiX. But you will still need a pre- and postprocessor if you don’t want to work with keywords (even though they are very convenient).

 

[IRstuff]

There's still a big difference; in the case of AI, multiple for-profit companies have released their fully-funded developments into the public domain, Google, Facebook, etc. But, the market for AI is EASILY 100x that of FEA, and there are still a lot of radically different AI approaches, so some of the motivation is to get everyone on board Tensor Calculus, or some such, vs. a competitor's approach.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm

 

[Coreform_Greg]

Here's your "FEA is complex" response

There are two major things to think about when it comes to FEA: capability and user-workflow. I would argue that the capability is (largely) available in the open-source domain. It's the user-workflow that's hard. For instance, there are some groups who need to work on large complex midsurfaced assemblies that are integrated within a sophisticated PLM system that provides material models etc. There are others who work with voxelized MRI data, some who receive facetted models (e.g. STL from a surface scan). Some groups need to link multiple simulations together to model an entire part's manufacture from bar-stock through forging, cutting, welding, assembly, and use; and other groups that just need to compute a simple eigenvalue result. Some groups prefer to model changes in boundary conditions within separate "steps", while others prefer monolithic steps with wide variations in boundary conditions. Some groups want to run on a single computer, others need to run on a 100-node high-performance computer while others on a super-computer. Some groups have simulations so massive the only way to visualize full-body data is via in situ visualization (see Catalyst), while others eschew visualization altogether in favor of "probe" data. Some groups prefer to use "general contact", while others require specific "surface-to-surface" pairs. Some groups want to model a weld via "element activation" whilst others use might use activation of tied-contact. Some customers are fine using quadratic tetrahedra, while others need to devote months to building a high-quality hex mesh. Some people want/need to write their own PDE to solve, while others just want to solve standard elastoplasticity. Some people think we should call them "displacement" or "temperature" boundary conditions, while others would call these "Dirichlet" boundary conditions. Some people want fracture solved via element deletion, others via J-integral crack opening, and others via phase-field fracture.

These things (and more) reflect different user wants and imply workflows and I think it's been hard for the open-source community to "decide" on a specific workflow and UX. Some of the best cases of good "full-package" FEA softwares include Calculix (replicate Abaqus workflow) and Salome Meca (meant for the open-source Code Aster code designed for nuclear reactor efforts in France). Notice that both of these are pretty focused in their scope / workflow.

But I do think the open-source community is exceptionally rich & mature in FEA workflow building-blocks. Consider the following partial lists:

FEM Preprocessing: CodeAster, CalculiX, ONELAB, Computational Model Builder, FreeCAD
FEM Meshing/Discretization: TetGen, Gmsh, fTetWild, Sigma|MeshKit, SAMRAI, libCEED
FEM Solvers/Frameworks: FEniCS, MOOSE, MFEM, ElmerFEM, Sparselizard, OOFEM, FEBio, deal.ii, FreeFEM, CalculiX, Goma, Nalu, GetFEM++, Hermes, NASTRAN, pyNASTRAN, Albany, JuliaFEM, Firedrake, libMesh, LANL FEHM, Truchas, Tusas, ExaConstit, libROM
FEM Equation Solvers: PETSc, Trilinos, XBraid, SUNDIALS, Hypre
FEM optimization: DAKOTA, OpenMDAO, hiop, Ceres Solver
FEM Post: ParaView/VTK, GLvis, VisIt
And even: "the open source computer aided engineering Linux distribution CAElinux."

And this list doesn't even include things like threeJS, OpenGL, etc.

 

[ESPcomposites]

You might want to have a look at

http://www.mystran.com.

MYSTRAN is similar to NASTRAN (the decks are almost equivalent), but has a modern code base (the NASTRAN source is very old and quite difficult to work with). We are doing to some great things with it. It will be incorporated into FreeCAD at some point. The main advantage is that the deck is "classical" (similar to NASTRAN, ANSYS, ABAQUS). Code_Aster is a great solver, but the deck can be quite confusing to those used to a classical approach. MYSTRAN has a full suite of "true" classical elements. CalculiX is an amazing program, but all elements are expanded internally into solid, which can lead to some challenges. As far as I know, MYSTRAN is the only solver that uses a classical input deck style with classical elements and has a modern code base (and its quite easy to use if you already know NASTRAN).

Brian

http://www.espcomposites.com

 

[Mustaine3]

I think the main issue is the critical mass of people that would participate in such a development.
You need people that are familiar with FEA, continuum mechanics and software development. There are probably not too many who can do this. When we assume that only 0.1% of those people are willing to spend their free time working on such a project, then this might be avery small group. They are then probably distributed across many projects and this results in very few people for each project. With that it is not possible to keep up with commercial developments.

 

[JoshPlumSE]

Who is going to do the work for Open Source applications? What's the motivation?

When there is a profit motive, there is a real incentive to make continuous advancements to gain a competitive edge and encourage your users to buy an upgrade.

 

[ESPcomposites]

Some potential users:

- Students who don't want to use limited codes
- Small companies (general)
- Small/large companies that want unlimited number of licenses for large runs
- (Similar to above) Small/large companies that want to use FEM in combination with in-house tools (hybrid solutions). This way all users can have access to the home grown tools without worrying about license issues (which could be hundreds or more users).
- Anyone who wants to customize the code
- Analysts who want to perform home/independent work. Some of them spent years learning analysis/FEA but don't have the funds to purchase expensive software.

Brian

http://www.espcomposites.com

 

[Slagathor]

JoshPlumSE,

What is difference vs the CFD world? Because OpenFOAM seems to have broken through and is now considered a viable platform vs commercial codes. Is this because CFD is more computationally intensive, and an open source package that has no "per core" licensing restrictions has a lot of advantages, and this drives adoption and development. Also, the users of CFD tend to be more sophisticated, and may not have as much need to cleanly packaged front ends that do the work for you....