Automatic sequence of 3D point data for surface fitting using neural networks

Abstract

In this paper, a neural network-based algorithm is proposed to explore the sequence of the measured point data for surface fitting. In CAD/CAM, the ordered data serves as the input to fit smooth surfaces so that a reverse engineering system can be established for 3D sculptured surface design. The geometry feature recognition capability of back-propagation neural networks is also explored. Scan number and 3D coordinates are used as the inputs of the proposed neural networks to determine the curve which a data point belongs to and the sequence number of the data point on the curve. In the segmentation process, the neural network output is segment number; while the segment number and sequence number on the same curve are the outputs when sequencing those points on the same curve. After evaluating a large number of trials, an optimal model is selected from various neural network architectures for segmentation and sequence. The neural network is successfully trained by the known data and validated the unexposed. The proposed model can easily adapt for new data measured from the same part for a more precise fitting surface. In comparison to Lin et al.’s [Lin, A. C., Lin, S.-Y., & Fang, T.-H. (1998). Automated sequence arrangement of 3D point data for surface fitting in reverse engineering. Computer in Industry, 35, 149–173] method, the presented algorithm neither needs to calculate the angle formed by each point and its two previous points nor causes any chaotic phenomenon of point order.