|
Scheduling
many tasks properly is an important problem to construct
CIM. In solving this problem, precise calculation
of the time required for each task is essential.
In this article, we take up the problem of estimating
the time required for processing resin molds. In
this field, rapid technology progress necessitates
processing molds by the state-of-the-art technology,
resulting in so many kinds of molds needed. Therefore,
the estimation of the processing time has been difficult
for a computer system and had to be done by processing
experts.
We applied a neural network to this estimation.
We defined the processing time to include not only
the time for processing molds with machines but
also the time before and after processing. Also,
we introduced 2 types of input variables for the
network; dimensional variables for mold design and
processing variables.
In applying a neural network to the solution of
such a problem, it has previously been criticized
that the network looks like the black box and that
the value estimated by the network is sometimes
very different from the expected value. Therefore,
the network has not been used with full confidence.
To solve this problem, we added the weight-decay
term to the learning criterion and introduced
a selective layer into the network configuration
method. As a result, we obtained an estimation system
whose qualitative meaning can be easily understood
and whose quantitative precision satisfies the requirements.
|
