Bayesian approach for extracting and identifying features

The paper introduces a new uncertainty reasoning based method for the identification and extraction of manufacturing features from solid model descriptions of objects. A major difficulty faced by previously proposed methods for feature extraction has been the interaction between features. In interacting situations, the representation for various primitive features is nonunique, making their recognition very difficult. The paper develops an approach based on generating, propagating, and combining geometric and topological evidence in a hierarchical belief network for identifying and extracting features. The methodology combines and propagates pieces of evidence to determine a set of corect virtual links to be augmented to the cavity graph representing a depression of the object so that the resulting supergraph can be partitioned to obtain the features of the object. The hierarchical belief network is constructed on the basis of the hypotheses for the potential virtual links. The pieces of evidence, which consist of topological and geometric relationships at different abstraction levels, impacts the belief network through its (amount of) support for different hypotheses. The propagation of the impact of different pieces of evidence updates the beliefs in the network in accordance with the Bayesian probabilistic rules.