A defect is more than just bits and bytes - Paper Technology International 2023 - Paper Technology International 2023 - Flipbook - Page 55

A defect is more than just bits and bytes - Paper Technology International 2023

PAPERTECHNOLOGYINTERNATIONAL
depending on the defect type, size or frequency, actions can be
initiated if desired. For example, defects at the edge of the web are
evaluated differently, as they are more critical, than defects in the
middle of the web. This reliable classification of all defects found by
the web inspection system is mandatory not only to deliver surface
quality information but as well to gain information concerning the
production process stability.
In the past and even in the present in rudimentary Inspection
Systems, the Classification is only based on decision rules also
known as rule-based classification or box classification. With this
approach, the user defines the rules for classification manually.
The next level is called image-based classification methodtoday‘s standard for most classification approaches used in modern
web inspection systems. Here, the user does not need to define
rules any more, but is defining defect classes and teaching those
classes with defect samples. The classifier would then define a
feature set for the classification which would distinguish the defects
in the best way.
With the growing demand regarding surface quality, more
data is needed to find process relevant defects. This makes the
classification more complex, time consuming and error prone
needing more advanced classification methods.
Therefore, ISRA is now providing an AI-based classification
approach based on deep-learning and multi-layer neural networks,
making classification more reliable and less time consuming for
the user. The Inspection system does recognize different defects
completely on its own, without any defect training or rule generation
by a user. ISRA´s system defines defect classes based on the defect
appearance. The only work left is naming the defect classes. Even if
the detection run is still extracting features, the classification system
uses optimized multi-parallel layer neural networks to distinguish
defects and asign them to groups (classes). Processing resources
to define the right set of selected features to distinguish classes
from each other are now allocated to the AI algorithms. This way the
classification of defects is much more precise; reliable and quicker.
Benefits to the production line
PAPER MASTER 4.0 - ISRA’s latest iteration of the
inspection management platform – is the first browser-based
platform for surface analysis to access the data of the ISRA system
from anywhere, whether it be directly at the production line, in a
meeting room, or on the road, by simply using mobile devices such
as a tablets or smartphones.
Another new feature is sync code marking based on laser
technology. This mark, done on the paper machine, can be detected
on the winder, which then specifies the position in the web. For
example, if there is an edge crack or other defect, the laser mark
coding tells the winder control to slow down to avoid a web break.
If a reel is going from a rewinder to a coater, it can be stopped;
the defect patched and then sent on to the coater. Today, about
25 installations with the laser marking technology benefit from the
technology where defects can be assigned by location on the sheet
in the next production stage, regardless of paper losses incurred in
the process.
With the Reel Release module ISRA also offers a solution for
winders that can give an exact quality report for each roll cut. This
setup can also tie in other data to ISRA´s software (e.g., moisture
measurement), so a producer can decide whether or not the end
product is saleable to a specific customer.
An additional new development is the introduction of color
cameras. They provide significantly more information about the
defect. For example, color cameras can distinguish between oil
and water drips on the web. For recycled mills, it can also identify
bacteria to identify whether a mill’s deinking system is working well
or not.
Figure 1b: With no actions taken, the web break has ultimately developed.
55

Paperturn flipbook system