Deep Neural Networks for Defects Detection in Gas Metal Arc Welding

Welding is one of the most complex industrial processes because it challenging to model, control, and inspect. In particular, quality inspection process critical a time-consuming activity. This research aims propose system online welded items with gas metal arc welding (GMAW) technology through use neural networks speed up process. following experimental tests, deviations parameters—such as current, voltage, speed—from Procedure Specification was used train fully connected deep network, once labels have been obtained for each weld seam multi-pass procedure non-destructive testing, which made possible find correspondence between defects (e.g., porosity, lack penetrations, etc.) parameters. The final results shown an accuracy greater than 93% in classification inference time less 150 ms, allow us this method real-time purposes. Furthermore work were trained reach smaller false positive rate task on test data, reduce presence faulty parts among non-defective parts.