An examination of engineering parts in large CAD-databases in order to create adjacency matrices and build clusters

This paper presents different approaches to extract information from large databases in order to process the gained information and find appropriate clusters. Starting from an unsorted collection of varying engineering parts in CAD format, the whole database will be automatically transformed to the non-proprietary IGES-format. Aided by Python-programmed macros controlling diverse measuring functions, every part is analyzed and every surface the part is consisting of is considered separately. For the further approach the following three different matrices are introduced. The first one is a simple adjacency matrix that indicates the connections of all surfaces among themselves. The second matrix is called distance matrix. Beside the spaces between the surfaces the part is built of, also some additional information, like the surface sizes or the centroids, is included. The last matrix is labeled as a thinned matrix. While all small surfaces are removed, it contains only the remaining large ones in order to use that kind of information to find clusters. In that last step the algorithm creates a number of classes automatically and assigns every engineering part from the initial database to its appropriate cluster. In addition to the whole methodology, a further program is presented that visualizes the matrices in a three-dimensional graph structure. Beside a comprehensive description of all programs showing simple examples, also an encompassing performance test on an actual database is conducted.