Part Recognition with Neural Network Implemention

• Main Authors: Lee,Shung Chung, 李淑瓊
• Other Authors: Su,Chwen Tzeng; Jiang,Ruey Ching
• Format: Others
• Language: zh-TW
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/60265695130418131986

碩士 === 中原大學 === 工業工程研究所 === 81 === The objective of this research is to develop a three- dimensional (3D) part recognition method by implementing neural network. The recognition procedures developed in this research has two phases. The first phase is trying to extract the data structure from Computer Aided Design (CAD) system and convert it into syntatic patterns which are then feed to a neural network system to perform two-dimensional part feature recognition. In order to recognize three- dimensional compound feature of parts, the major tasks in second phase consist of developing some feature recognition rules and combining the results from phase one to develop an adjacent matrix for each two dimensional pattern. The adjacent matrix is the input into 3D neural network recognizer to recognize the 3D compound part features. Basically, the features can be recognized in this research are limited on steps and slots which include standard slot, i-directional blind slot, blind slot, standard step, tri- directional blind step, and blind step. In addition, an experiment was conducted in this research to study the relationship between system performance and neural network parameters such as batch learning sample, learning rate, and number of hidden layer.