Giorgia Ficili
Digital Twin and Machine Learning in welding: implementation of a CNN model for image classification and quality monitoring.
Rel. Giulia Bruno, Emiliano Traini, Gabriel Antal. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Gestionale (Engineering And Management), 2023
|
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (11MB) | Preview |
|
Archive (ZIP) (Documenti_allegati)
- Other
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (102kB) |
Abstract: |
Digital twin technology is rapidly evolving within the Industry 4.0 realm and is emerging as a powerful tool for companies to transform themselves and proceed towards digitalisation of their operations. Often, digital twin technology embeds some artificial intelligent models, in the attempt of achieving a so-called ‘Intelligent Digital Twin’, with even more enhanced capabilities of data analysis, fault detection, decision-making support, prediction, and many more. In line with the advantages coming from the synergy between digital twin and advanced deep learning algorithms, a Convolutional Neural Network has been developed within this thesis work, thought as a component of a broader Digital Twin comprehensive framework for Resistance Spot Welding quality monitoring. Indeed, the purpose of the work performed is to give a contribution in trying to address the quality issues impacting Resistance Spot Welding process, which is often affected by some welding defects. Particularly, the focus is on a specific and quite common defect in RSW, i.e., expulsion: it consists in the ejection of molten metal out of the nugget area during the welding process and can strongly affect the quality of the welding in terms of joint strength and other structural issues. The proposed deep learning CNN algorithm has been designed for image classification and fed with post-Resistance Spot Welding workpieces images: the network has the ability to classify the images on the basis of whether the welded piece in the picture shows an expulsion or not. Despite the acquired dataset is much smaller than the ones proposed by other literature applications, the developed algorithm is surprisingly performant by giving rise to high prediction accuracy values. Such an algorithm could be a sub-model to be included within an overall digital twin framework, which might be adopted by companies for a more automated quality monitoring over the outcomes of a welding assembly: expulsion appearing on a welded piece, indeed, acts as a window into the welding process and is typically considered by manufacturers as an indicator for quality of the process. |
---|---|
Relators: | Giulia Bruno, Emiliano Traini, Gabriel Antal |
Academic year: | 2023/24 |
Publication type: | Electronic |
Number of Pages: | 135 |
Subjects: | |
Corso di laurea: | Corso di laurea magistrale in Ingegneria Gestionale (Engineering And Management) |
Classe di laurea: | New organization > Master science > LM-31 - MANAGEMENT ENGINEERING |
Aziende collaboratrici: | UNSPECIFIED |
URI: | http://webthesis.biblio.polito.it/id/eprint/29756 |
Modify record (reserved for operators) |