The aim of the present study is to develop and evaluate an automatic system for real-time arc welding quality assessment and defect detection. The system research and development focuses on the identification of potential defects that may arise during the welding process by analysing the occurrence of any changes in the visible and near-infrared spectrum of the weld pool. Currently, the state- of-the-art is very simplistic being based on an operator observing the process. The work is subjective and the criteria of acceptance based solely on operators observations can change over time due to fatigue leading to incorrect classification.
Variations in the weld pool are the primary result of the chosen welding parameters and torch position, being at the same time the very first indication of the resulting weld quality.
The system under development in this research study consists of a camera used to record the welding process and a processing unit which analyse the frames giving an indication of the quality expected. The frames are analysed using image processing techniques by extracting the weld pool characteristics (i.e. width, area, etc.) and categorise the weld into correct or defective. If defective weld, the target is to identify the type of defect. The categorisation is achieved by employing Artificial Neural Networks and correlating the weld pool appearance with resulting quality.