A cool title, isn’t it? Hello ANSYS blog readers! This is my first time in this blog as a guest blogger. You will notice a brief resume of mine together my photo as the author of this post, but let me introduce myself so that you can understand why I am here writing about mesh morphing to the ANSYS audience.
I am a Professor at University of Rome, with good experience in fluid structure interaction (FSI) and Fluent customization using UDF programming. Five years ago, driven by a Formula 1 Top Team, I developed a powerful mesh morphing tool crafted by tough specifications. Managing any kind of mesh, precise, fast and parallel! Nothing at that time was able to do this kind of job. We tried to go with (RBFs) Radial Basis Functions mesh morphing, one of the most promising techniques. And we made it.
In 2009, I entered the ANSYS Partnership Program and launched RBF Morph into the market as Fluent’s add-on so that a tool able to make happy Formula 1 CFD designers (and their 300 million cells meshes!) could be available to all ANSYS Fluent users. With our motto “Welcome to the world of fast morphing!”, the exciting RBF Morph adventure had its inception.
OK, I hope now you are curious enough and have asked yourself “What the heck is RBF Morph and what does it do?”
Lets starts with an example. Take a look at the picture to the right. You are looking at a carotid bifurcation. There are many possible anatomical variations of this bifurcation — should we have a geometry of each and every possible carotid shape? No! Researchers use RBF-Morph to study different clinically relevant anatomical modifications without the need of a geometry of each and every single one. The mesh a previous shape to a new one instead of remeshing every possible shape. Therefore, morphing the computational grid to the different carotid bifurcation shapes dramatically speeds up the simulation process.
RBF Morph is a software add-on for ANSYS Fluent (with a GUI a TUI, a user guide and whatever you would expect) that allows you to define shape parameters directly inside Fluent by interacting with mesh through the RBF Morph GUI and TUI. You can define as many shape parameters as you want and store them as a small piece of information (just a couple of files per parameter, a way smaller than mesh). Such an information can be then used to transform your Fluent model into a shape parametric one. Parametric? How?
Make it simple using an example. Suppose that you are working on an external CFD study of a car and you are trying to reduce the drag changing the roof angle and the boat tailing angle. After tweaking your set-up for a couple of hours, by means of the RBF Morph GUI (it typically takes one hour to set and refine a single shape modifier), you store your RBF solutions: “roof-angle”, “tail-angle”. Both of them are based on 1° set-up value. Now you want to update your mesh such to gain 1.5° for the roof angle and 2.5° for the boat tailing angle. Just use a single TUI command:
(rbf-morph ‘((roof-angle 1.5) (tail-angle 2.5)))
That’s all! No remeshing, no need to reload your mesh, just parametric mesh morphing with your Fluent loaded in memory and the mesh distributed across partitions. And of course, yes, you could update the parameters yourself but you can rely on optimization solution such as DesignXplorer to automatically explore variations of your design and optimize it.
The following equation express the idea: (ok I received a recommendation “Marco please, not equations in the blog!”, but this should not bother you)
1 Fluent model + 1 (optional) Fluent solution + 2 RBF Morph file pairs = infinite meshes
Typically, the morphing time is way lower than the time to load a mesh in memory, so you can explore as many shapes as you want, it is just a matter of CFD time. Moreover, using consistent meshes (because with morphing you will just update nodal positions), the convergence time is generally lower with respect to baseline because you begin with a hot restart.
Another great advantage of mesh morphing is that you avoid remeshing noise. In fact, if you generate a new mesh to represent a small variation, you will introduce the effect of shape modification and the (undesired) effect due to new mesh that comes with a different mesh convergence error that could be similar to the effect of shape parameter itself.
To give you the big picture (and close my post that is now quite long!) let me complete the example using the 50:50:50 procedure, an exciting project ran in 2012 by ANSYS, supported by Intel and Volvo. The 50:50:50 challenge was an optimization with no compromises with respect to accuracy (50 million cells CFD model), completeness (50 shape configurations investigated) and speed (50 hours for the whole process). A key for the automation of such workflow was RBF Morph to make the shape parametric with respect to: roof angle, green house angle, boat tail angle and front spoiler angle. In the 50 hours of the workflow the quota consumed by RBF Morph was about 50 minutes, i.e. one minute for each mesh update.
Consider the great benefit that you receive when the meshing of a 50 million cells model can be replaced by just one minute of mesh morphing. A negligible time if you think that to store the mesh and the solution takes about ten minutes.
These types of benefits are typical of mesh morphing methods. But not all mesh morphing approaches allow you to get the same performances. Radial Basis Functions mesh morphing combine the flexibility of popular meshless methods based on FFD (Free Form Deformation, i.e. deforming boxes) with a high accuracy. Therefore the accurate RBF-Morph morphing technology can be used for applications and simulations that FFD methods cannot address. For example, applications and simulations that have local constraints, STL target, adjoint sensitivity sculpting, fluid structure interaction, ice accretion and many many others.
Thanks for reading, I hope to have future chances as guest blogger to go further in depth on this exciting topic, but in the meantime, here are some great resources I would like to share with you.
Editors Note: RBF Morph was awarded the “Most Advanced Approach Using Integrated and Combined Simulation Methods” at the European Automotive Simulation Conference (EASC 2009). Recently awarded for the “Best Use of HPC” at ANSYS Automotive Simulation World Congress (ASWC 2013).