Fine detection of process state and secondary process emissions in laser powder-bed fusion (LPBF) by dual acoustic sensor acquisition and machine learning

Abstract

Powder Bed Fusion – Laser Based/Metal (PBF-LB/M), respective Laser powder-bed fusion (LPBF) has become a key additive manufacturing method, driving the need for robust monitoring techniques to ensure process stability and reproducibility. In-situ monitoring methods are broadly categorized into optical and acoustic approaches, each with distinct advantages. Acoustic methods, unlike optical ones, can capture emissions from within the substrate and are often more cost-effective. Most PBF-LB/M monitoring systems rely on either airborne or structure-borne sensors, rarely on both.

This work demonstrates a dual-sensor system integrating airborne and structure-borne sensors to leverage their complementary strengths for predicting both the process state of a PBF-LB/M-machine and defects (e.g. cracks or delamination, visible via secondary process emissions) occuring during the build. Using a fine-tuned neural network, both sensor types individually, as well as their combination, achieved a classification accuracy of ≈ 99 % for predicting the process state based on acquired emissions. Additionally, the structure-borne sensor proved superior in detection of secondary process emissions and slightly more accurate in process state classification. The compact form factor of the structure-borne sensor further enhances system flexibility, making this dual-sensor approach a versatile solution for PBF-LB/M monitoring.