The batch sizes in the production of consumer goods are constantly increasing. Therefore, there are more flexible and faster machines with complex and susceptible processes. Even modern measurement and automation systems cannot prevent these problems. The automatic detection of every possible reason for a breakdown is too expensive and often even impossible. That is why current diagnosis systems detect only consequences (e.g. product build-up, web break) but not their specific causes (e.g. incorrectly folded carton trays).
Stable process control with a minimum of downtime and waste requires longstanding expertise and the operator’s extensive process knowledge. Companies have more and more problems because of high level fluctuation in workforces and are missing skilled workers. In addition to missing possibilities for saving operator knowledge, there are high production losses, production waste and high machine wear.
A possible solution of these problems is to make implicit knowledge explicitly available. This aim can be reached by making operators use learning assistance systems. These systems learn through matching the machine’s condition and breakdown description of operators and can autonomously diagnose new breakdowns. A prerequisite for autonomous diagnoses is the autonomous recognition of machine conditions without specific additional sensor systems. This becomes possible by matching specific breakdowns to specific patterns of condition of whole internal sensor systems.
The potential of this approach will be investigated in the IVLV-Project. The first work package is the research of detection and reduction algorithm, software tools, and real time data processing systems. The recognition of sequence patterns will be exemplary examined on laboratory machines. The result contains statements for the feasibility of learning assistance systems.
Fehler: Die ID ist kein Bild.