When we need to conduct a FE simulation of guided wave propagation, small element size and time step must be used to capture the waves, especially for high-frequency cases. Millions of elements would be normal for a model with a small size, say a plate with a dimension of 200mm*200mm*1mm. Is there any shortcut to avoid this type of difficulty or software that is more suitable for this work than ABAQUS or ANSYS, .etc?
Hi Han Lu
Whether/how much you can simplify this problem depends on many different factors. You may consider the following aspects:
- Is a two-dimensional model sufficient? -> Employ plane strain assumption.
- If not, can it be treated as axisymmetric? -> Exploit rotational symmetry in FEM or even solve (semi-)analytically
- Is the frequency very low? -> Use plate elements to avoid discretizing the thickness
- Is the frequency high? -> High-oder elements can be worthwhile, together with explicit solvers (if not frequency domain)
- Is the material linear? -> Solving in the frequency domain may be more effective.
- Is the structure of (piece-wise) constant cross-section? Semi-analytical approaches (SBFEM, TLM, SAFE) can be extremely efficient.
- Can the dimensions of the model be reduced by applying absorbing boundary conditions? -> Use SBFEM, PML, absorbing regions....
I hope this gives you some ideas to start. Feel free to ask about details of any of the approaches.
have you tried the discontinuous galerkin method with Comsol Multiphysics? https://www.comsol.it/blogs/modeling-piezoelectricity-using-the-discontinuous-galerkin-method/
Personally, I suggest:
1. Using mesh-free methods (e.g., DEM, SPH) may reduce the computational cost. But it somehow requires some understanding of computational mechanics.
2. Alternatively, using GNN can help to save the computational costs, but it initially requires lots of data to train, (i.e., at least 1000 cases). Two examples are here:
Learning mesh-based simulations:
https://sites.google.com/view/meshgraphnets
Learning to simulate complex physics with graph networks:
https://sites.google.com/view/learning-to-simulate
3. you may seek a chance to assess HPC or supercomputing platform or look for a collaborator who has the resources.
Just my own thought, feel free to comment below.
- Badges
- Science topic