One of the major problems in the production of biaxially stretched films is that almost no film properties can be measured during production. This makes it necessary to test film samples of a produced roll in a laboratory.
These tests can take up to three hours, depending on the scope of the tests and equipment of the film laboratory. If a produced roll turns out to be of inferior quality, it is not only that that roll needs to be scrapped, but at least also the subsequent three hours of film production.
Even worse, if a roll turns out to be bad, operators need to take the right decision to get production back on track. Ideally, this would be an immediate fix, but occasionally it can take days, if not weeks, to find the right settings for a satisfactory production.
A system that can guide operators to take the right decisions ideally would avoid bad rolls in the first place. Furthermore, it could still be of great help for the subsequent task of troubleshooting once a bad roll has come up.
An important building block for such a system could be the systematic acquisition of knowledge by experienced plant operators in the form of cause-effect graphs combined with data mining techniques. This data, prepared in a suitable user-friendly system, could show a novice operator what actions a skilled operator would take in the specific context.
Another point we only considered marginally at the beginning of the project was the simulation of plant components. We considered this to be an issue that – at least in context of film stretching plants – can be seen to be far off in the future. Surprisingly, during the course of the project, it has been shown that individual parts of our systems can already be simulated today to a sufficient depth in order to run simple test scenarios, in particular with regard to the testing of PLC software.