Machine Learning in AM: Layer-by-Layer Defect Detection in Powder Bed Fusion

Summary:
In powder bed fusion (PBF), each printed layer can be imaged and analyzed, providing the basis for data-driven assessment for both process stability and part quality. However, there are several limitations to current layer imaging techniques, which can results in product escapes and production delays. In situ Backscattered Electron (BSE) imaging layer by layer serves as a highly effective inspection technique for PBF quality control, offering imaging capabilities that utilize direct atomic interaction with the melted layer to detect density variations, cracking and surface features.

Description:
Artificial intelligence (AI) techniques such as machine learning are increasingly being integrated into additive manufacturing workflows to interpret process data and analyze large datasets such as layer by layer powder bed images. In this presentation, a supervised learning approach for automated defect classification using sequential BSE image datasets will be shown, which ultimately can be used for defect detection for serial production. A manually annotated training set from which the model learned to associate BSE data with defect likelihood. During inference, the trained classifier evaluates new layer images to determine the likelihood of a feature being a true defect. Model performance is assessed through statistical indicators of detection accuracy and sensitivity, ensuring reliable identification of actual defects.
Attendees will learn the details of this approach and how it demonstrates the potential for scalable, in-situ defect detection in additive manufacturing, with combining machine learning and electron beam based imaging. The approach establishes the groundwork for future remelting strategies such as using the electron beam melt pool depth for closing defects above a certain threshold in the next layer.