A feature-based reverse engineering system using artificial neural networks
Reverse Engineering (RE) is the process of reconstructing CAD models from scanned data of a physical part acquired using 3D scanners. RE has attracted a great deal of research interest over the last decade. However, a review of the literature reveals that most research work have focused on creation of free form surfaces from point cloud data. Representing geometry in terms of surface patches is adequate to represent positional information, but can not capture any of the higher level structure of the part. Reconstructing solid models is of importance since the resulting solid models can be directly imported into commercial solid modellers for various manufacturing activities such as process planning, integral property computation, assembly analysis, and other applications. This research discusses the novel methodology of extracting geometric features directly from a data set of 3D scanned points, which utilises the concepts of artificial neural networks (ANNs). In order to design and develop a generic feature-based RE system for prismatic parts, the following five main tasks were investigated. (1) point data processing algorithms; (2) edge detection strategies; (3) a feature recogniser using ANNs; (4) a feature extraction module; (5) a CAD model exchanger into other CAD/CAM systems via IGES. A key feature of this research is the incorporation of ANN in feature recognition. The use of ANN approach has enabled the development of a flexible feature-based RE methodology that can be trained to deal with new features. ANNs require parallel input patterns. In this research, four geometric attributes extracted from a point set are input to the ANN module for feature recognition: chain codes, convex/concave, circular/rectangular and open/closed attribute. Recognising each feature requires the determination of these attributes. New and robust algorithms are developed for determining these attributes for each of the features. This feature-based approach currently focuses on solving the feature recognition problem based on 2.5D shapes such as block pocket, step, slot, hole, and boss, which are common and crucial in mechanical engineering products. This approach is validated using a set of industrial components. The test results show that the strategy for recognising features is reliable.