Decision tree based weld defect classification using current and voltage signatures in GMAW process

The quality of the weld is most important in industries manufacturing boilers and pressure vessels which will work in severe operating conditions. In an automated environment, developing a process monitoring and control system will ensure the weld quality and prevent the occurrence of defects. In this paper, an attempt is made using the decision tree algorithm to establish a correlation between the current and voltage signatures with the quality of the weld. Carbon steel plates are welded using GMAW processes and experimental design is established to obtain weld without any defects (good weld) and weld with porosity and burn-through defects. "KUKA" robotic GMAW welding setup integrated with "Fronius" power source is used in this study for experimentation. "TVC" data acquisition system is used to capture the current and voltage signatures. Statistical features are extracted from the current and voltage signatures. Decision tree algorithm with split criterions such as "gini index", "towing", and "deviance" are used to classify the weld defects. Results indicate the effectiveness of decision tree algorithms in classifying the weld defects using the current and voltage signatures. (C) 2017 Elsevier Ltd. All rights reserved. Selection and/or Peer-review under responsibility of International Conference on Emerging Trends in Materials and Manufacturing Engineering (IMME17).