• artificial intelligence
• machine learning
• virtual sensor
• soft sensor
• moisture prediction
• corrugated cardboard
• manufacturing
• Industry 4.0

Nowadays, Artificial Intelligence is playing a crucial role in the manufacturing sector,
and it can help ensure quality, waste reduction and cost optimization. Using
various real-time IIoT (Industrial Internet of Things) sensors, industries can collect
a lot of data and allow to improve the levels of operational and business efficiency,
integral parts of Industry 4.0.
This project thesis was born from a collaboration between the University of Pisa
and Fosber S.p.A., an international group leader in the design, production and installation
of complete corrugating lines and machinery for corrugated cardboard,
a material used mainly in the packaging sector. This thesis is applied to a case
study in this area and related challenges with the help of Machine Learning, which
is specifically suitable given the enormous amount of data available from these sensors.
In particular, the study focuses on the problem of the estimation of the initial moisture
value of the paper (reel) by means of the analysis of the measured parameter’s
variation in different points on the corrugator to avoid the installation of high-cost
moisture sensors. There is currently no state-of-the-art solution to face this task for
corrugated cardboard.
The thesis work involved all phases of the Machine Learning workflow, starting from Data Understanding to the evaluation of the models implemented.
Applying a correct Machine Learning pipeline and rigorous validation schemes
made it possible to tackle the problem, obtaining performances in line with those
expected by domain experts.