DIADEM: Defect Identification ADditivE Manufacturing

The widespread adoption of Additive Manufacturing (AM) particularly Selective Laser Melting (SLM) has the potential to disrupt the conventional manufacturing supply chain employed by automotive, aerospace and industrial product manufacturers. There are many parameters that impact on AM cost, time, and quality of build and furthermore for aerospace, considerable time and cost is expended on QA (particularly X-ray and computer tomography (CT) scanning and testing of parts after each build) because some defects are difficult to detect with conventional non-destructive testing (NDT) techniques due to part complexity. State-of-the-art AM machines with sophisticated melt-pool monitoring systems promise improved quality control but require exceptional computing power to process the terabytes of recorded sensor data.

Our approach is different and propose to collect high-resolution images of the powder layer surface taken before & after recoating, perform quality testing of the build parts using X-ray computed tomography (xCT), label different types of defects on the layer-based xCT images and then use them to train machine learning (ML) model to automatically identify these defects from images of the powder layer surface. This means that the development of defects can be identified in real-time from the captured powder layer images and the developer of the part can be alerted as the build progresses. This implies that developed parts will be more reliable and the development of defects during serial production identified thereby limiting the production of rejected parts.