Research On Freeform Feature Reconstruction Based On RBF Mixture Neural Network

Feature modeling is one of the key techniques in integrated system of CAD/ CAM, which is the core of the design to product model. During the process of product design and modification with the feature modeling system based on the current course, the technique of feature reconstruction is particularly important. The efficiency of Reconstruction will be increased with the linear growth of complexity of the model, The Reconstruction in the existing feature modeling system are basically ruled shape features. With the development of feature modeling technology, involving irregular increasing problems, the time of reconstruction for the geometric model will be longer, and model efficiency is not high enough, this question needs to be solved timely.In order to maintain correctly the intent of designers and satisfy them, the view of this question that the current feature modeling contain only a small number (such as cylindrical and prismatic) rules graphics, and the accuracy of surface modeling to the measurement scattered is not high under large sample space, this paper depth study with neural network technology and features reconfigurable technology, the main activities include the following:1. This paper analyzed the basic theory of semantic feature modeling and the said-method of feature, introduced the free-form surface features, on the basis of HUST-CAID (Computer Aided Industry Design System developed by Harbin University of Science and Technology) semantic feature modeling system as an example, and detailed descript the freeform feature of the design, create process and reconstruction order.2. It proposed the reconstruction mechanism of the freeform feature based on neural network in feature modeling with the thinking of the level-divide and modular, created mixture neural network model based on RBF (RBFMNN). Greatly improving the surface to the accuracy of modeling through by RBFMNN extraction achieve feature of large sample set trained, enhanced identification robustness and fault-tolerance.3. In order to reduce the reconstructed data redundancy of freeform feature and improve the efficiency of reconstruction, and convert the topology of three-dimensional model describing to vector data of the neural network this paper proposed improved the algorithm or the methodology of the data pretreatment reconfiguration;4. It optimized the method of reconstruction to the freeform feature and to develop evaluation programmer to the accuracy of the reconstruction...