Extraction of part feature information from CAD Model
Different feature recognition (FR) techniques have been used in the past to extract features from
CAD model. FR techniques have been classified into six basic groups: topological, heuristic,
symbolic, volumetric and process-centric and hybrid. Further, each group is divided into several
subgroup. The taxonomies of feature recognition techniques are provided in Figure M8.2.1.
Figure M8.2.1: Taxonomies of feature recognition techniques (Shah et al. (2001))
Feature recognition techniques
Graph Based Methods
Graph based approach use boundary model to create attributed face adjacency graph (AAG) as
shown in Figure M8.2.2. In this case, AAG faces are represented as nodes, the edges as arcs, and
concave/convex attributes are assigned to the arc edges corresponding to the angle between
adjacent faces. In order to determine required features in AAG, face nodes with all incident arcs
with a convex attribute are deleted. The remaining graph components are then matched with
feature AAGs to recognize the features. This approach works properly for non-intersecting
depression type features.
Figure M8.2.2: Attributed face adjacency graph (AAG) (Shah et al. (2001))
Hint Based Methods
In this approach, the face pattern is used as hints to the existence of feature. In order to define a
hint, a specified pattern is defined in the boundary of a part by taking geometric and topological
relations. Feature existence rule provides the clue for feature hints. These feature existence rule
specify how much of the boundary of a feature must remain in the part after interaction with
other features. For example, if a feature is a linear slot, then the slot feature must have some
subset of both opposite faces remaining. Once the feature hint is established, the algorithm
completes the feature to produce the largest volumetric feature that is compatible with the other
information in the part model. The algorithm reunites the features that were separated by feature
interaction just before completing the process. The completed features are marked as ‘‘optional’’
and ‘‘required’’ to represent the feature interactions. The discussed method has been represented
as shown in Figure M8.2.3.
Figure M8.2.3: Representation of feature interactions (Shah et al. (2001))
Convex Decomposition
Convex decomposition method is also called Alternating Sum of Volumes (ASV)
Decomposition. ASV method was developed by Woo (Woo, 1982). In this method, the author
has proposed a way to obtain either negative or positive volumetric feature decomposition in a
selective manner. The first step in this method is to compute the convex hull of a non-convex
part. To obtain the convex deficiency the part model is subtracted from the convex hull. This
decomposition is achieved through recursive steps.
In the next step, the original part is represented by a Boolean combination of convex
components. The first convex hull represents positive volume and the convex hull of the first
deficiency represents the negative volume. But the problem in the above approach is that the
recursive steps never terminate which is a non-convergence problem. Kim [Kim, 1992] proposed
an alternate method to solve the non-convergence problem called Alternating Sum of Volumes
with Partitioning (ASVP). This method is explained in the Figure M8.2.4.
For the part given in Figure M8.2.4 (c), its polyhedral abstraction and ASVP decomposition is
shown in Figure M8.2.4 (a). After applying combination operation, form feature decomposition
is obtained as shown in Figure M8.2.4 (b). Part shown in the Figure M8.2.4, is recognized to
have a base block, a hole, a step, a slot, two open pockets, and a rectangular pocket.
Figure M8.2.4: Method of convex decomposition (Shah et al. (2001))
Volume Decomposition and Re-composition
In this method, the part is subtracted from the stock to get one or more separate volume or lump.
These lumps are volumes to be removed by machining from the stock as shown in Figure M8.2.5
(a). Three main stages in this decomposition process are
1. Lumps represent volumes to be
2. Volume decomposition into ‘‘atomic’’ cells (shown in Figure M8.2.5 (b))
3. Concatenation of cells into macro volumes (shown in Figure M8.2.5 (c) & (d))
Figure M8.2.5: Method of volume decomposition and re-composition (Shah et al. (2001))
Free Form Features.
In freeform feature extraction, there is a general method in which two principal curvatures are
used. The curvature is classified as concave (+), convex (-), and flat (0). The surfaces are
partitioned into curvature regions with identical curvature types for both the principal curvatures.
Based on the above, curvature regions are classified into Pure Protrusion ([-,-][-,0]), Pure
Depression([+,+][+,0]), Transition([+,-]) and Flat([0,0]). The two signs represent the sign of two
principal curvature.
Hybrid Techniques
In the techniques mentioned above there are many advantages as well as limitations. To
overcome the limitations, different researches used hybrid recognition techniques by combining
the concepts of several basic techniques.
Reference
Woo, T., 1982, ‘‘Feature Extraction by Volume Decomposition,’’ Proc. Conf. on CAD/CAM
Technology in Mechanical Engineering, MIT, Cambridge, MA, pp. 76–94.
Kim, Y. S., 1992, ‘‘Recognition of Form Features Using Convex Decomposition,’’ Comput.-
Aided Des. 24, No. 9, pp. 461–476.
Sonithi, R., Kunjur, G., and Gadh, R., 1997, ‘‘Shape Feature Determination Using the Curvature
Region Representation,’’ Fourth ACM Symposium on Solid Modeling and Applications, pp.
285–296.
Shah, J. J., Anderson, D., Kim, Y. S., & Joshi, S. (2001). A discourse on geometric feature
recognition from CAD models. Journal of Computing and Information Science in
Engineering, 1(1), 41-51.
