Research On The Application Of Radical Basis Functions And Neural Network Technology In Reverse Engineering

Although reverse engineering technology has been widely applied in many fields, there are still many problems to be researched. To provide new effective approaches and to improve the behaviors of present methods, the problems of model reconstruction from dense scattered points and triangle mesh repairing were investigated thoroughly and systemically based on available investigations and the novel mathematic tools, namely radial basis functions (RBF) and artificial neural network. The main contents and achievements of this dissertation are as follows:1. Combined with the RBF neural network, a novel repairing method for triangle mesh which contains holes was presented. The RBF neural network was trained by the coordinates of the vertices around the holes, and then used to create the coordinates of the new triangle vertices to repair the unexpected holes in the mesh. Compared with Geomagic, one of the famous software in reverse engineering, the present algorithm has good repairing effects and high efficiency.2. An innovative holes repairing algorithm for triangle mesh was put forward and implemented based on volumetric data field. The ordinary hole, island hole and unclosed hole were repaired by establishing an implicit surface equation which was fitted to the vertices near the holes based on RBF and the vertices of added triangles were mapped further to the implicit surface. The algorithm guarantees the geometry consistency between the mending surface and the original surface around the hole. Excellent results which are better than those in the domestic references have been achieved in repairing the holes in large curvature change area. Compared with Geomagic software, the repairing results of big holes were not in the shade. The algorithm was extended skillfully to deal with holes in colorful triangle meshes and showed good repairing results as well.3. The model reconstruction technology based on compactly supported radial basis functions (CSRBF) was studied. A new cloud point simplification algorithm was presented. The implicit surface reconstruction from the simplified point set and the reconstruction error analysis based on CSRBF were realized. A novel idea to reconstruct the implicit surface of the vertices of triangle mesh by using CSRBF, and the offset surface of arbitrary topology mesh based on the volumetric data field and the marching cube method were implemented. As a result, the self-intersection of offset was avoided, which has favorable integration with rapid prototype manufacturing (RPM) and numerical controlled machining.4. The smooth B-spline surface reconstruction technology based on self-organizing map neural network (SOM) was studied. A new neurons initialization method and a divide-and-conquer training scheme for scattered points were introduced. As a result, the training time was reduced and the training effect was improved obviously. Then the parameterization from scattered points in quadrilateral patches network of triangle meshes was realized by using adaptive parameterization method based on SOM. With the fitting precision, fairness and continuity in full consideration, an applied approach was realized to smoothly fit piecewise B-spline surfaces to the triangle mesh.The algorithms proposed in this dissertation were verified by lots of experimental examples.