About this Abstract |
Meeting |
6th International Congress on 3D Materials Science (3DMS 2022)
|
Symposium
|
6th International Congress on 3D Materials Science (3DMS 2022)
|
Presentation Title |
Machine Learning Framework for Spiking Defect Detection in Electron Beam Welding |
Author(s) |
Sanjib Kumar Jaypuria, Bondada Venkatasainath, Santosh Gupta, Dilip Kumar Pratihar, Debalay Chakrabarti |
On-Site Speaker (Planned) |
Sanjib Kumar Jaypuria |
Abstract Scope |
Although electron beam joints are known for the large aspect ratio, the inherent spiking defects in the weld is inevitable. Spiking is one of 3D weld defect and it is characterized as non-uniform penetration along the welding direction. The fluctuation in penetration in these partial penetration weld act as stress raiser and becomes prone site of cracking. Therefore, industries need a trade-off between penetration and spiking to accept the joint with acceptable penetration and spiking level. In this study, machine learning based unsupervised clustering approaches are used to classify the acceptable and unacceptable joints. Fuzzy c-mean clustering (FCM) and density-based spatial clustering of applications with noise (DBSCAN) clustering algorithms have been used for the classification of these joints. Comparison of these approaches are done through Silhouette coefficient (SC), Calinski-Harabasz index (CHI), and Davies-Bouldin index (DBI). Fuzzy c-mean clustering is found to be more accurate in classification and gives necessary information about the process variables of electron beam welding. |
Proceedings Inclusion? |
Definite: Other |