Generation of manufacturing process knowledge for process optimization : a case study on milling

• Κύριος συγγραφέας: Σούφλας, Αθανάσιος
• Άλλοι συγγραφείς: Souflas, Athanasios
• Γλώσσα: English
• Έκδοση: 2021
• Θέματα: Milling; Chatter detection; Milling modelling; Machine learning; Knowledge generation; Κατεργασία φρεζαρίσματος; Ανίχνευση chatter; Μοντελοποίηση κατεργασίας φρεζαρίσματος; Μηχανική μάθηση; Παραγωγή γνώσης

Industry 4.0, also called the fourth industrial revolution, is aiming to push the digital transformation of the manufacturing sector. The usage of sensors in the machines, which are interconnected with one another, as well as with other levels of the product lifecycle, such as design and operating life of the product will be introduced in a wide audience. Therefore, technologies that provide the ability to manage this amplitude of data and process them are key elements of this era. As a result, the automated generation of manufacturing process knowledge is highly pursued. To this end, structured approaches to design and develop robust and high-performance knowledge generation systems are required. This work presents an approach based on the DIKW pyramid, where the application of each level of the pyramid is mapped into the manufacturing process context, in order to define the requirements for the knowledge generation system. Physics-based modelling of the process is utilized to enable the identification of the required transformations to climb the DIKW pyramid. The proposed approach is validated in a case study on chatter detection in milling. An acceleration sensor is integrated in the milling machine and Variational Mode Decomposition is utilized to decompose the acceleration signal and keep the component that holds chatter rich information. Features that are sensitive to the phenomenon of chatter are extracted from the signal, in order to capture the transient nature of the phenomenon, compared to the steady-state, stable process, as well as the energy shift from the tooth passing frequencies towards the structural modes of the machine. A Support Vector Machine classifier is trained and utilized towards chatter detection, enabling high classification accuracy and excellent detection speed, which can facilitate the implementation of the knowledge generation system towards real-time monitoring and control of chatter during the milling process.