I've seen a similar paper to this one using a case-based reasoning approach to the process planning for cylindrical components in what way do think this approach is better or worse, and how do you think you could further extend this work.

In the pdf file state that “In the absence of any theory or guide in the selection of the most appropriate configuration of the neural network, the optimum architecture can be only reached through trial and error …”

1) Do you consider that is enough the “trial and error” method to perform the network optimization and

2) How deal the system to avoid the problem of premature ending when a local minimum is founded?

3) Why were selected the sigmoid function as a Transfer Function?

4) What is the name of the software that was used?

It seems that many efforts are required in order to build the examples allowing to teach the neural network, even on a rather simple case. If rules are already available, containing some knowledge about the sequence of operations, what is the added value of the neural network ? In my opinion, NN are interesting when no formalised knowledge is available but when solved examples can easily be obtained. It does not seem to be the case here.