Production of corrugated board

30 years of experience - high-quality product range

Packwell Logo

Corrugated board is a versatile and above all ecological packaging material and therefore the first choice for shipping goods. Every year, more than 250,000 km² of corrugated board is produced worldwide in a very complex and demanding industrial process consisting of more than 7 individual steps. The production is realized in large corrugated cardboard production plants, which produce up to 1,500m² of corrugated cardboard per minute.

High-quality packaging has been produced in Schwepnitz near Dresden in Saxony since 1994. Packwell Schwepnitz is an attractive and solid employer in the region and one of the most modern and productive packaging plants in Germany. As a member of the Palm Packaging Group, Packwell Schwepnitz offers a broad and high-quality product range.

Decisive parameter: the evenness of the corrugated board

The greatest challenge in production is the evenness (flatness) of the corrugated board. Deviations from this are the cause of most rejects (> 2%). The total loss is around €1.25 billion per year.

Uneven corrugated board is caused by inhomogeneous and inappropriate moisture in the mainly recycled paper, whose properties vary greatly.

The difficulty lies in measuring the flatness across the entire width (up to 3.35 m) of the paper web moving at up to 450 m/min and, based on this to optimize and automatically control the process in real time.

DENK VISION AI signals “OK” or “not OK” with light signals

Packwell Schwepnitz now uses AI in its production to minimize waste and to be able to act immediately in the event of any problems (e.g. bulges).

Normabweichungen werden Mitarbeitern sofort über ein optisches Signal mitgeteilt, sodass schnell in die Fließbandproduktion eingegriffen werden kann.

Green screen: Flatness Ok! Red screen: Flatness not Ok! (unevenness recorded)

With individual AI to a stable solution - training in the DENK VISION AI Hub

To ensure reliable AI results, the customer himself trained the system using individual, real images. The customer trained the AI as an “expert” with regard to correct and unacceptable plan positions. As an “expert”, only the customer himself can make congruent statements about the planning situation and consistently train the AI.

A reliable network was trained with fewer than 700 input images. In the case of classification, the training itself was quick and easy: all the customer had to do was decide whether the flatness of an image was “good” or “bad” by pressing the number 1 or 2 on their keyboard.

Only by differentiating between these two classes was a stable AI trained, which was easily integrated offline into the customer's computer environment using an API.

DENKi wants your email address

DENKblatt requested?

Grab our newsletter faster than your machines can sort components.

Click to join our webinars to learn about possibilities and applications of our Vision AI technology and to see it run in practice!