Model reconstruction of existing products using neural networks for reverse engineering

In many cases, the task of manufacturing a new product starts with a prototype. Then the detailed physical characteristics of the product is extracted to facilitate its redesign or manufacture. Reverse engineering is the process by which this is done. In general, reverse engineering is accomplished through two stages-part digitizing and surface modeling. Techniques for part digitizing are well established and commercial systems are available. The area that is less developed is to generate models based on the digitized points. The paper emphasizes the applications of neural networks in reconstruction of computer models for existing surfaces. Three neural networks which use the value of defined surface parameters as input and identify the correspondent points are designed and trained using a backpropagation algorithm. In order to solve the problem of the network visual feasibility, functions of OpenGL are introduced in the training program to monitor changes of the parameters and errors of the networks. To evaluate the effectiveness of the approach, a mathematically known surface, a non uniform B spline surface is used for generating a number of samples for training the networks. The trained networks then generate a number of new points which can be used to reconstruct computer representations of the existing surfaces and to manufacture these surfaces.