Digital Signal Processing Reference
In-Depth Information
External disturbances
Production process
Part OK or defectiv
Material
Workpiece
Process parameter
Signals and data
Neuronal network
Quality
Prognosis
quality
Process parameters
Prognosis control
parameters
Neuronal network
Control
Illustration 296:
NEPRES in industrial use: schematic overview
The target here is to realize production with an error rate as low as possible. This requires methods that
allow a cycle- synchronized quality statement with practically no time delay. Due to the monitoring of the
quality fluctuations integrated in the process and a regulation of the adjustment parameters built on this,
this fast reaction to process- respectively quality variations is made possible. This is achieved by a model
monitoring of process data respectively process parameters, where by the means on neuronal networks
variations in product qualitiy can be detected (source: Fraunhofer IPA).
If already here no visual or only an uncertain distinguishability can be found, new
reference material should be implemented. Even more efficient is the according waterfall-
display of the parameters. If the clear distinguishability is given, the resulting neuronal
network will be working successfully with very high probability (see also Illustration
293).
Neuronal networks in industrial use
The connections between cause and effect in real procedures like industrial production
processes often are highly complex. This high complexity is also a result of the large
number of simultaneously interactive parameters. Furthermore, the great variety of
influencing factors as well as inevitable coincidental disturbances play a big role. Real
processes therefore evade analytic and exact formal descriptions of these connections to
the greatest possible extent.
So far, quality monitoring of manufactured products mostly happens by subsequent
inspection, by samples or even by a 100% inspection.
