The Vega shrink-wrapper from OCME is the machine that is used in food industry production lines. It groups bottles or cans into sets, wraps these sets into plastic film and then heat-shrinks the plastic film to combine them into package. The film is fed into the machine by large spools and then cut to proper length by blade inside machine. The cutting component is one of vital parts of the machine as the quality of cut may later determine quality of package itself. The blade cannot be inspected visually during work due to fast rotation speed and being enclosed in machine’s housing. Monitoring of cutting parameters will improve machine reliability and reduce downtime caused by failed cuts.
The paper presents method of wear classification through combining both supervised and unsupervised machine learning models. In the dataset it is known which measurements are done by using worn component and which are using the new one. The unsupervised learning model and blade labels will be used to determine which measurements were defective and then supervised learning model will be trained to detect when blade is producing faulty cuts. Dataset used in training those models is publicly available under 3.0 Unported (CC BY-SA 3.0) license.