Research On Key Technology In Feature-based Reconstruction Of Solid Model In Reverse Engineering

Reconstructing CAD model of existing physical object is an underlying task of reverse engineering. Compared with surface model, solid model could represent the 3D shape of object more integrally and unambiguously, which provides reconstructed model with the capacity of accrate volume, inertia and barycenter calculations. Nowadays, solid model reconstruction has become a new hot topic in reverse engineering. Feature is the elementary carrier of model design intent, and the implied feature information within physical object is very important to reengineer solid model accordant with original design representation. This dissertation studies feature-based solid model reconstruction in reverse engineering, and presents a good solution for the solid model reconstruction on the basis of solid feature combination and surface feature stitching. The main research subjects and achievements are as follows:Data segmentation technique is investigated. A robust edge-detection algorithm is proposed based on normal & area rule, which is applied to classify meshes with similar geometric attributes into the same partitions by"region-growing"from seeds.Then the partitioned model could be refined by some methods such as eliminating impure triangular meshes, combining small partitions with adjacent one, deposing narrow partitions into surrounding ones and smoothing jagged boundaries.The proposed algorithm could be well adaptive to models with different structures by adjusting its parameters.Solid feature recognition based on ANN is studied. A feature library consisting of common geometric features is established. Then an extensible feature recognizer is developed based on BP network. And an algorithm is presented to encode such attributes of feature as section shape, open & close and convex & concave automatically. After trained via sufficient training samples, the feature recognizer could effectively identify the features from triangular mesh model.Parameter extraction of solid features is addressed. A probability model concerning the errors from collected data to feature primitives is presented. Then, an approach is proposed to fit partitioned data by surface based on maximum likelihood estimation and calculate geometric, location and orientation parameters of solid features. The Mahalanobis distance is introduced to evaluate the closeness between collected data and reconstructed features. Furtherover, the expressions of Mahalanobis distance concerning plane, cylinder and cone are deduced. Finally, the feature parameters are refined by applying an appropriate similarity transformation to solid features.To reconstruct free-form objects with trimmed surfaces, an approach for trimmed surface reconstruction is proposed by fitting B-spline surface to cover all the data points and then trimming the B-spline surface by boundary curves or adjacent surfaces. An algorithm based on base plane is presented to parameterize the data points rapidly. A hole detection method based on valid points identification rule is proposed, and then the CNSP(Control Network Shape Preservation) constraints are selectively imposed upon control points around holes to guarantee the existence of least squares solution during B-spline surface fitting. The error-controllable approaches are designed for B-spline surface fitting and trimmed boundary reconstruction. The proposed approach is characterized by accordance with design intent, shape consistence around holes and high precision of resulting surface.Solid model reconstruction based on solid feature combination and surface feature stitching is studied. The interactive model tree is built to manage reconstructed features and constraints. The reconstructed solid features are capable of parameter-driven modification, which enables users to revise the features based on design intent. An algorithm of constraint recognition & collision detection between different solid features is proposed, and solid model reconstruction based on constraint and feature combination is achieved. Constraint recognition and constraint-based surfaces fitting simultaneously is studied, and an automatic surface intersection, trimming and stitching algorithm is proposed to reconstruct B-rep models of free-form objects.According to the achievements above, a solid model reconstruction system is developed and applied to reconstruction of aerial structural parts. Experimental results demonstrate the proposed methods.