Component Identification and Clustering

Challenge

A proper partitioning of elements often depends on the point of view. A mechanical designer may divide in terms of vehicle or assembly modules. A calculation engineer will divide into groups with identical properties (material, thickness etc.). In many applications a partitioning of connected elements of the same type or size is a more proper choice. An application for this kind of partitioning is e.g. a causal chain analysis.

Solution

The standard solution is to manually define element groups in the finite element pre- or post-processor. This is time consuming and must be updated at every model change.

commodo gives several methods to find a useful partitioning. One can collect shared nodes and split the mesh at some element type (beams, rigid body elements etc.) or collect elements of the same type with shared nodes. This solution does not need any predefined groups and can even be used to re-identify components or arbitrary selections in other models even if section names or the mesh has changed.

Sometimes a more detailed partitioning is needed or one needs sets of similar size. For this purpose the commodo has some method for spatial clustering of elements since again, there is not one solution for all applications. k-medoids or k-means may be useful to get clusters with similar numbers of elements or similar spherical size. Spectral clustering methods are again useful if the resulting clusters must be connected.