Search by shape: new CAD-independent search engines automatically code and classify shapes for quick retrieval

Automotive Design & Production, March, 2007 by Lawrence S. Gould

Group technology (GT) is based on order arising from the structure of part families. By identifying parts uniquely and grouping similar parts together, retrieving one part can lead to finding similar parts. Great idea. Useful in concept. And yet conventional GT search methodologies--namely, alphanumeric text-based search--fail primarily because of their manual roots. Someone has to manually determine part attributes, manually code those attributes, manually tie those codes together to make up a part's identification number, and then manually enter those codes and ID numbers into a computer. Add to this, says Abir Qamhiyah, president of iSEEK Corporation (Ames, IA; www. iseekcorp.com), "Different users, different groups, and different applications will have different interests and different classification criteria." Moreover, she says, alphanumeric-based GT has "limited value because file names are often part numbers or cryptic references to the assembly in which the part belongs."

Shape search technologies follow the same GT philosophy. The only difference is in the implementation. Shape search uses 3D part geometry as the basis for the search, it is fully automated, and predefined clusters of shapes are not required. GT finally works the way it should. People find parts, similar parts, and even redundant parts fast.

How to search

CADSeek from iSEEK and Geolus Search from UGS (Plano, TX; www.ugs.com), and other shape search products do for parts what Google does with text: They find and display parts similar to the initial inquiry. That initial inquiry can come in various ways. There's the conventional textual approach: Users simply enter all or a portion of a part name or number, using wildcard characters if necessary. Or part metadata can be searched for specific part or shape attributes. Another approach is to load an existing file of the target part, or just sketch an image of the target part in the computer-aided design (CAD) program, and then have the search engine find similar parts. Suitable matches are displayed in a web browser for further inspection. As necessary, users can rotate these displays to better visualize the shapes. Click on the desired shape and the search engine will go find similar shapes or open that shape file in CAD. (Users can tell the engine how many results to return.) Users can also scan through (fly over) a 3D representation of the company's CAD library. This representation shows families, or "clusters," of similar shapes and parts. By clicking on a cluster, users can drill deeper into the specific part classification. Last, users can page through a digital parts catalog. The catalog shows 3D thumbnail images of parts, as well as part characteristics as text (metadata). Again, clicking on the desired image opens it up in CAD.

[ILLUSTRATION OMITTED]

[ILLUSTRATION OMITTED]

Behind the scenes

Shape search engines classify shapes in different ways. Explains Qamhiyah, shape in 3D space can be described by its voxels, the volume it occupies, or by the surface that surrounds it. (A "slightly more sophisticated," but simplistic, voxel approach involves "thinning" the volume into a graph model, explains Qamhiyah.) The voxel methodology has problems in scaling, changes in orientation, storage requirements, and such. The surface-based approach, which is how CADSeek works, requires less storage and is faster to process. Geolus Search uses a hybrid approach. It looks at volumes (voxels), surfaces, edges, overall size, and other attributes of shape "to figure out exactly what's going to match," says Chris Kelley, vice president partner and platform marketing for UGS.

Implementation begins with source geometries in a tessellated file format (JT and STL for CADSeek; both those and VRML for Geolus Search). CADSeek takes the surface information and creates a point cloud for each part, which is fed into the search engine algorithm. (CADSeek can even take point clouds from CMM devices. Again, it only needs the skin of a part.) The CADSeek algorithm uses statistical analysis to classify the shapes, create a hierarchy of shapes, and cluster those shapes appropriately. The shape code for each shape is "nothing but a tag," explains Don Flugrad, senior vice president and COO of iSEEK. "Once you retrieve it, you can then pull out all other related information, including production information." Production and other metadata comes from the CAD design files. At installation, the search engine scans the CAD files and captures field and field contents automatically. New fields are added as they come up in the metadata either during initial installation or later as CAD files are created or modified.

In general, the shape search products consist of three basic pieces. Various server-side components (including connectors, indexers, and the database) "perform all the heavy lifting," explains Kelley. These components convert the source geometries into mathematical representations of the shape. The second piece are the plug-ins that let users initiate a search from virtually any major product lifecycle management and CAD system. The third part is the web browser on the client side (typically, Internet Explorer, Netscape, and Firefox) to search and view results.