Jul 24, 2018, Luma Bendini

Q&A | Automated adaptive mesh optimization/refinement

'Would you like to have a cloud-based service for automated adaptive mesh optimization/refinement of your CFD-mesh?''

That was the question our CEO and Co-founder, Sebastian Desand, asked his LinkedIn followers on July 5th. It was great to see the interest from CFD/CAE-engineers in the subject and many questions were raised. We asked the helpful hand of our Co-Founder Prof. Johan Hoffman and here are the answers:

Combustion and Modelling Engineer: Can it refine hex or poly dominant meshes? Can structured boundary layer regions also be refined?

Johan Hoffman and Sebastian Desand:Yes, this should work in general, please contact us to set up a meeting to discuss the details of your specific

Aerospace Engineer: What about Scalability?

Johan Hoffman and Sebastian Desand: It’s highly scalable. In theory up to 50.000 cores depending on the simulation. In most cases we use hundreds or a few thousand cores on supercomputer resources. This means that we can deliver a refined mesh within a few hours (depending of course on the mesh size/ level of refinement).

CFD Engineer: I have used the adaptive mesh for multiphase flow modelling. Often the VoF algorithm need a big refinement of the mesh around the interface between the non miscibible phases, but being this a guess of the problem the mesh has to be refined in a big subdomain. Thus the mesh adaptation/refinement helps to better detect the interface/s of the fluid by means of a local automatic refinement which follow a value of the volume fraction. This is really helpful to reduce the computational demand and the accuracy of the interface detection.

Johan Hoffman and Sebastian Desand: Yes, this is great way to use adaptive mesh refinement, to detect the interface in a multiphase problem. Adding a posteriori error estimation to this, you could also specify regions (or quantities) of interest away from the interface region where you need high accuracy. The adaptive algorithm can then automatically optimize the mesh also based on this information.

Finite element methods Development Specialist: Can the solution be applied to solid meshes? I presume you're refining the mesh based on the error estimators, in that case, let us say I want your adaptivity solution to refine the mesh to an accuracy of 5% stress error. Can the adaptivity solver maintain a 5% error throughout my structure or there will be regions of smaller (than 5%) error around? Can the error be different (based on end user preference) over different regions / locations in the structure?

Johan Hoffman and Sebastian Desand:You can define regions with different associated error tolerances, for example in stress. But for a solid mesh, the error estimate will depend on the specific constitutive laws of the solid. Please contact us to discuss your particular case of interest.

CFD, Physics and Math Modeller: The first thing I would like to understand is how dependent the optimised mesh is in terms of error reduction on the solver and solver numerics (i.e finite volume solver with different divergence schemes)?

Johan Hoffman and Sebastian Desand: In principle it is independent of the solver numerics, since the mesh refinement algorithm is based on the underlying matematical model (Navier-Stokes equations).

Here is the link to the post (as long as LinkedIn is maintaining it online): CLICK. And make sure to contact Sebastian (Sebastian at IngridCloud dot com) if you need any customized solution for your CFD-mesh!