Neural networks - Signals, Processes, and Systems - page 459

Illustration 299: Production of an injection- molded casing for digital clocks (source: PKT )
The principle of the system is shown in Illustration 295. Initially, NEPRES was developed
for the monitoring of plastic injection molding processes and can reach its full potential
there due to the available preprocessing and the bundled process- knowledge in it.
Because of the universality of the implemented approach and the modular structure - the
entire functionality is included in modules integrated in DASY Lab - NEPRES is
generally applicable for various processes.
Before the monitoring process, the system has to be adapted to the respective target
process, meaning that e.g. the “generating” of pattern data and the training of the neuronal
network have to take place. In practice, the following systematic procedure has turned out
to be successful:
• The first step for any process is a detailed analysis of the planned target
process with intense involvement of the staff as an important carrier of
knowledge. By analysis the critical product features to be monitored are
found out as well as which process parameters are crucial for steering the
process. Based on this, the necessary sensors are deduced.
• After the process has been equipped with the sensors determined before,
and after the coupling of the entire measurement value acquisition, tests are
made systematically. These are necessary to optimally create the pattern
data later on needed for the training of the neuronal network.