Search for a command to run...
Additive manufacturing (AM) has changed the entire manufacturing enterprise by offering unique features for the fabrication of complex-shapes with superior mechanical properties. In the last decades, through an exponential advancement, AM has been promoted from a prototyping to a series and mass production platform. Like all conventional techniques, quality assurance procedures/tools are of the utmost importance in aiding manufacturers in quality management and certification. For this purpose, in-line melt pool monitoring devices, installed in laser-based AM systems, provide vital real-time information about process characteristics, implicitly or explicitly leading toward understanding the quality of printed parts. This research aims to develop a defect-detection platform using in-situ monitoring of light intensity emitted from the melt pool of laser powder bed fusion (LPBF) to detect pores initiated from the lack of fusion phenomenon. This platform is driven by correlating disturbances in the light intensity emitted from the melt pool to actual pores identified through a post-processing micro-computed tomography (CT) scanning. Two sets of experiments were devised: one with embedded micro-voids to purposefully mimic the lack of fusion in printed parts composed on Hastelloy X to assess the sensor response and develop the analysis algorithm. The second set was included printed parts with stochastic/randomized distributions of pores to evaluate the proposed approach. The recorded data were extracted from an Absolute Limits algorithm and were analyzed offline through image processing. Next, the printed samples were CT scanned, and the data from both steps were analyzed by the segmentation method and confusion matrix to examine the correlation. The results of the intentionally seeded defects demonstrated that voids larger than 120 µm were detectable through the collected photo-diodes signals. The evaluation matrices to validate stochastic/randomized distributions of pores also showed that for two sets of process parameters with a high laser power of 200 W, hatching distance of 150 and 90 µm, and process speed of 1000 and 1500 mm/s, the sensor prediction from randomized defects is about 70.14 ± 2.24% and 72.82 ± 1.39%, respectively. However, for the low laser power cases with laser power of 100 W, hatching distance of 90 µm, and process speed of 1000 mm/s, the correlation was less than 30%.