Databased Welding Technologies – Uprgrading Quality Management
When it comes to welding, mistakes can be expensive and time-consuming. While they can be detected and corrected in Quality Management, ensuring fewer mistakes happen in the first place is the better option to save resources.
In collaboration with specialists from DXC Technology, employees of the University of Applied Sciences Kiel and doctoral students from Heriot-Watt University (Scotland), the project team led by thyssenkrupp Marine System‘s experts team is working on a model for data-driven welding optimization. The solution is to be used for all welds in the future, the pilot is currently still focusing on section joints in shipbuilding.
Digitalization or databased welding provides the ability to avoid mistakes during the welding process. With a combination of data streams from the new EWM-welding machines and position sensors, welding errors will be detected or completely prevented in the future. Currently, errors are detected by the quality management department and passed on to the specialist department. There, correction takes place, and the quality management inspection starts all over again. Technical NDT (non-destructive testing) is performed using ultrasound and X-ray technology. This inspection will not be abolished with the new solution. However, error rates will significantly be reduced.
Supervised machine learning
But how does it work in practice? For the first recording, four EWM-welding machines were combined in a network and connected to a local notebook computer. The data is transmitted directly from the machines and recorded. In addition, position sensors are used. Sensors are placed for use on the welding gun and some are attached to the welder's wrist. This approach was chosen in order to obtain the broadest possible data base in as little time as possible. The sum of the data from the sensors is combined with the quality features from the welding logs and is then analyzed. The resulting data sets serve as a training basis for a model from the field of artificial intelligence and machine learning. The goal of the first part of the project is to train the model in such a way that it reports possible problem areas on the weld seam after a successful weld and thus relieves quality management.
An outlook: chances of databased welding
Through artificial intelligence, in the future the welding process will virtually be error-free – except for external errors due to dust, etc. In the medium to long term, most welding processes will be executed automatically with robots or with robotic assistance. By 2030, collaborative systems – meaning robots plus AI & humans – can take over most welding tasks. Exclusively complicated welds will be handled by manual welding.