Anomaly detection for in-production systems diagnosis: a replicable model applied to a manufacturing coextrusion process

The recent trend in the industry field is to push forward well-designed and efficient production systems. To reach this goal, one of the main challenges, that has to be faced and overcome, consists in achieving an as much as possible continuous and clean process. The aim of this thesis is to apply data scientist tools, in order to develop data-driven models that may successfully monitor an in-production process and detect straight away suspicious behaviors. The whole project focuses on a real-life use case of plastic extrusion, enhancing its reliability and validity. Even though the designed models are meant to cope with the specific case circumstances, they have been thought in such a way to be generalizable and applicable to similar problems, where their employment may be part of a winning strategy.