Discontinuity Detection in the Shield Metal Arc Welding Process

• Published in: Sensors (Basel, Switzerland) Switzerland), 2017-05, Vol.17 (5), p.1082
• Main Authors: Cocota, José Alberto Naves, Garcia, Gabriel Carvalho, da Costa, Adilson Rodrigues, de Lima, Milton Sérgio Fernandes, Rocha, Filipe Augusto Santos, Freitas, Gustavo Medeiros
• Format: Article
• Language: English
• Subjects: Accuracy; Acoustic emission; Discontinuity; Feasibility; Learning theory; Neural networks; Sensors; Shielded metal arc welding; Support vector machines
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s17051082

This work proposes a new methodology for the detection of discontinuities in the weld bead applied in Shielded Metal Arc Welding (SMAW) processes. The detection system is based on two sensors-a microphone and piezoelectric-that acquire acoustic emissions generated during the welding. The feature vectors extracted from the sensor dataset are used to construct classifier models. The approaches based on Artificial Neural Network (ANN) and Support Vector Machine (SVM) classifiers are able to identify with a high accuracy the three proposed weld bead classes: desirable weld bead, shrinkage cavity and burn through discontinuities. Experimental results illustrate the system's high accuracy, greater than 90% for each class. A novel Hierarchical Support Vector Machine (HSVM) structure is proposed to make feasible the use of this system in industrial environments. This approach presented 96.6% overall accuracy. Given the simplicity of the equipment involved, this system can be applied in the metal transformation industries.