Searching for similar shapes
Hydro Aluminium is a major extruder of aluminium profiles world wide. They produce tens of thousands of different shapes, introducing many new shapes every day. All the different shapes are available in 2D cross section CAD drawings, meaning that they have a vast number of historical and current drawings available in their databases. If they could be able to search these databases for earlier products that are similar to for instance a new product or a new customer request, this could have many uses.
|
Pattern recognition techniques were used to find similarities between cross-sections of earlier produced aluminium profiles and new ones. Based on the similarities, process parameters for the new aluminium profile could be estimated. Here are shown some examples of hollow profile shapes. |
SINTEF has on behalf of Hydro Aluminium developed a sysem that searches a database of CAD drawing for the ones that most resembles the shape of a new profile presented by the user.
The system uses 20 – 30 different shape attributes pre-computed and stored for all the shapes in the database. The same attributes are computed for the new shape, and the similarities to the shapes in the database are defined by a distance measure in the shape attributes. The search itself takes less than a second and results in a presentation of for instance the 20 most similar shapes. The system is fast and easy to use, and is now in regular use in Hydro plants.
This project involved the development of geometry analysis methods and a set of shape attributes that are specially designed for the application. Also variable selection methods, distance measures, pattern recognition and empirical modelling were challenges in the project.
The project was supported by the Norwegian Research Council (NFR) through the PROSMAT programme.
If you are interested in more information please contact
Tom Kavli
.
