Abstract: The neuro-fuzzy network
(NFN) is used to model the rules and
experience of the process planner. NFN is to select the
manufacturing operations sequences for the part features. A
detailed description of the NFN system development is given. The
rule structure utilizes sigmoid functions to fuzzify the
inputs, multiplication to combine the if part of the rules and
summation to integrate the fired rules. Expert knowledge from
previous process plans is used in determining the initial
network structure and parameters of the membership functions. A
back-propagation (BP)training algorithm was developed to fine
tune the knowledge to company standards using the input-output
data from executions of previous plans. The method is illustrated
by an industrial example.
Key words: Neuro-fuzzy
networks Training Semi-generative systems CAPP
Manuscript received on
April 26, 1998; revised manuscript May 24, 1999
|