Neural networks accelerate quality control in metal 3D printing

Additive manufacturing of metals is considered a key building block for producing complex components, for example in aerospace or the energy sector. Despite technological progress, ensuring component quality remains a central hurdle. In particular, internal defects that arise during the printing process are difficult to detect with conventional methods and slow down the industrial adoption of metal additive manufacturing processes.

Research groups in mechanical engineering are therefore increasingly relying on high-resolution X-ray imaging to observe the melting process in real time during printing. These experiments provide detailed insights into melt pool dynamics, pore formation, and solidification processes. At the same time, they generate enormous volumes of data that are hardly manageable through manual analysis. This is precisely where the neural network AM-SegNet comes into play, developed by a team led by Prof. Peter Lee and Dr Alex Chu Lun Leung at University College London.

AM-SegNet is designed as a lightweight deep-learning model that has been specifically trained to segment X-ray images from metal manufacturing. It is based on a dataset of more than 10,000 annotated images from international large-scale research facilities. According to the developers, the model achieves an accuracy of around 96 percent and processes individual images in less than four milliseconds. This makes the approach suitable for near-real-time analysis during ongoing experiments.

Through automated analysis, physical effects in the additive process can be identified more quickly and investigated systematically. This supports both process development and quality assurance strategies, for example by enabling early detection of unstable printing conditions. The full release of the source code via GitHub is also intended to facilitate further development by other research teams. Overall, it is becoming clear that data-driven methods are increasingly becoming an integral part of metal additive manufacturing.