A Novel Input Representation for ANN-Based Features Recognition

Feature technology is considered a focal research subject in computer-aided design and manufacturing (CAD/CAM). There have been two major approaches: feature recognition and design by features. Feature recognition makes direct use of a geometric model and generates application-specific feature models using various recognition rule sets regarding the application. In contrast, design by features specifies a design model using a set of design features defined in the modelling system. Design by features reduces remarkably the amount of work of recognising features, but does not eliminate the need for feature recognition [2]. Various feature recognition methods have been proposed such as graph matching, volume decomposition, etc. However, there are still problems with feature recognition hindering its practical applications, such as low recognising speed, inability to learn, etc. Artificial Neural Network-based (ANN-based) feature recognition techniques have become attractive because they can eliminate some drawbacks of conventional feature recognition. The main problems with it include the limited range of features recognised and additional heuristic rules needed. This paper presents a new input representation for ANN-based feature recognition aiming at tackling these problems.