Towards efficient powder quality control in additive manufacturing via an in situ capable device and methodology leveraging multispectral machine learning
Clemens Maucher, Jonas Gerold, Hans‐Christian Möhring
Abstract
Additive manufacturing (AM) processes enables the fabrication of highly complex parts that cannot be manufactured using conventional manufacturing methods. Constant and specified material properties are of crucial importance for these highly optimized components. AM processes like laser-based powder bed fusion for metals or laser-based direct energy deposition use fine metallic powder as the feedstock material. Powder properties can be influenced due to, for example, reuse cycles or environmental influences. These changes have a significant impact on the properties of the components to be manufactured. It is therefore particularly important to monitor the quality of the powder used throughout the entire process. However common methods for powder qualification often are expensive, time consuming and cannot be applied to powders in situ the built process. In this paper a new qualification method with a corresponding device is introduced, that could easily be transferred to in situ process monitoring of powder characteristics, offering consistent production of high-quality components, and enhancing the efficiency of the overall process. This developed device consists of a system for automated acquisition of images using a digital light microscope along with multispectral and multidirectional illumination. For testing and to simulate the process of image acquisition after recoating in the AM machine a recoating test stand has been build. Image data was evaluated using image segmentation methods and machine learning algorithms for classification. It was shown that particle size distributions obtained from image segmentation provide good matches with comparative data. In addition, machine learning algorithms were able to correctly assign eight different powder classes with a test data accuracy of 91.3 %, and the influence of illumination was demonstrated. Furthermore, it could be shown for titanium (Ti-6Al-4 V) powder, that the proportion of particles with colored oxidation layers increased with increasing degree of powder degradation. Lastly, the use of the developed methods for characterization of an unknown powder with unknown degree of degradation was demonstrated.