Qualitative and quantitative characterization of powder bed quality in laser powder-bed fusion additive manufacturing by multi-task learning

Poor quality consistency is a challenge in the field of additive manufacturing, necessitating the development of online monitoring for the additive manufacturing process. In the process of laser powder-bed fusion additive manufacturing, powder spreading is one of the most basic and important steps. The powder bed quality directly impacts the printing process. Besides, it’s needed to be aware that not all powder bed defects have the same impact on printing quality, and that some defects should be treated specially. To accomplish this, a multi-task deep learning model is trained and tested using a data set comprised of industrial powder bed images to analyze the quality of the powder bed qualitatively and quantitatively. In detail, its classification module is meant to carry out high-precision qualitative analysis for the defect that significantly affects the printing process, while its segmentation module is meant to quantitatively determine the degree of general defects. Experimental results indicate the proposed multi-task model performs well in different tasks. During the training, the test set's classification accuracy can reach 100%, and the segmentation MIoU (mean intersection over union) can reach 75.23%. Furthermore, it’s found that the multi-task learning model has obvious advantages over single-task models. In addition to obtaining multi-dimensional defect analysis results, the multi-task deep learning strategy also yields improved analysis accuracy, which is highly valuable for powder bed quality monitoring in laser powder-bed fusion additive manufacturing.