Intelligent computer -aided process planning for rotationally symmetrical parts using neural network and expert system

The research work reported in this thesis is aimed at exploring possible applications of different Artificial Intelligence (AI) techniques for automating two of the important decision making tasks in generative Computer-Aided Process Planning (CAPP) systems, namely the selection of machining operations and the set-up planning in the case of rotationally symmetrical parts.

The relevant literature on previous research work for automating the machining operations selection by decision trees, expert systems and neural networks has been reviewed, highlighting their contributions and shortcomings. The literature review indicates that the neural networks based approaches can overcome several limitations of the decision trees and the expert system based approaches. However, inspite of the above advantages, a limitation observed with the neural network based approaches developed by previous researchers is that there are no guidelines for choosing the input patterns of training examples. Also an issue that has not been adequately addressed by previous researchers is that whether any prior domain knowledge on machining operations selection could be taken advantage of. Further, most of the previously developed neural network models tend to recommend a single machining operation sequence for a given feature of the part. Keeping the above in mind, a neural network based methodology has been developed for selection of all the possible alternative operation sequences for machining a given feature of the part. The PC based software package NeuFrame Version 4 (2000) has been used to simulate the neural network operation. The neural network has been pre-structured with prior knowledge on machining operations selection in the form of heuristic or thumb rules. It has been achieved by developing two forms of representation for the input data to the neural network. The external representation is used to enter the crisp values of the input decision variables (namely the feature type and its attributes such as diameter or width, tolerance and surface finish) to the neural network. The purpose of internal representation is to categorize the above crisp values into sets; these sets, in turn, correspond to all the possible different ranges of dimension, tolerance and surface finish, encountered in the antecedent 'IF' part of the thumb rules mentioned above. (Abstract shortened by UMI.)