Weld Monitoring developments

Due to the increased industrial need for improved welding productivity, speed and quality, many components that were originally welded manually are now welded using mechanised or automated processes , eg. robots. These have the benefit of not being dependent on high skilled labour (which is potentially decreasingly available), but also are capable of producing consistent results, compared with the wider variation seen from welder to welder.

However, these automated systems do not have the full range of observation available to a manual welder, and as such are less capable of responding to a developing situation, to prevent the formation of defects or sub-optimal welds. This work attempts to simulate, to some extent, the observation capabilities of an operator, and allow the information collected to be fed back to a monitoring system. This system would then be able to either interrupt the process or perform some modification to the welding parameters in order to bring it back to an optimal condition.

Three different sensing systems were applied to an arc welding process. These were a vision system with a high dynamic range (HDR) camera, an acoustic sensor which measured through substrate noises and an electrical parameter monitoring system. Each system was used in parallel, and the signals collected by each were analysed to determine trends in the data that correlated with a defective or non-defective weld.

Image analysis was done using a neural network system that performed iterative self-training based on supplied images and user identification of artificial flaws in test pieces. The system was then tested for accuracy by provision of a testing dataset of example images. Consideration is also given to the accuracy of the system dependent on response time.