Summary: | 碩士 === 中原大學 === 工業工程研究所 === 81 === Coding and classification scheme and production flow analysis
are the two main approaches to implement Group Technology
principles. In this research, a hybrid method for coding and
classifying rotatory mechanical parts by integrating the
bidirectional associate memory model of neural networks and a
rule based feature extractor was presented. In additional, a
new type of competitive neural networks model to solve
production flow analysis problems. To perform the coding and
classification, the KK-3 system was used to develop the system.
Based on the DXF file of CAD , the operational features of
parts are recognized by the rule feature extraction method
developed in this research. Weight of our BAM neural network is
then built according to the standard definition of KK-3 to
obtain the part classification code. results show that:(1)If
the input data is complete and correct, output code conforms to
KK-3 exactly, and(2)If the input data is insufficient or
ambiguous, the output might provide several KK-3 codes from
which coding experts can select a suitable one. can accept this
new knowledge. This learning feature could system''s performance
and meet the needs of realistic system. On the part of
production flow analysis, the competitive neural network
proposed in this research use the processing operation
information of parts as input data and classified parts into
groups with balanced sizes and avoid improper classification
which is generated by input patterns closely clustered and high
similarities among input patterns. The testing results can be
divided into two different types: (1)Input patterns are not
well-organized (2)Input patterns are well-organized. All the
simulation results show that the model can generate correct,
consistent and reasonable answers. With the aids of the
algorithm, system can handle large scale problems and solving
them correctly.
|