Gan Enables Outlier Detection and Property Monitoring for Additive Manufacturing of Complex Structures

Additive manufacturing technologies have seen significant economic growth over the past years. Although additive manufacturing processes have matured in many areas, difficulties with regard to printing accuracy persist. Possible defects are, e.g., the generation of unwanted internal pores or a lack of fusion between layers, to name only a few. In general, defects result in a deviation between as-planned and as-built geometries, all of which can be difficult to detect in an automatized fashion. Previous work has shown that image-based simulation can assist in quality monitoring of produced parts and may complement experimental testing. Yet, both experimental testing and simulation-based approaches are involved and not yet directly applicable to all manufactured parts in a series production. The paper at hand suggests a remedy to this problem by using a generative adversarial network (GAN).Generative adversarial networks have shown to be able to emulate as-built geometries of engineering relevance. Moreover, they can realistically reproduce the distributions of such deviations. To this end, we present how this feature can be harvested to employ generative adversarial networks for outlier detection. To this end, we use the discriminator of a GAN as a classifier on as-built parts to judge whether an as-built structure is acceptable or defective. The viability of the approach is demonstrated on basic artificial structures with spherical voids as well as additively manufactured lattice structures whose geometry is acquired after production via computed tomography (CT). The methodology is not only applicable for automated property monitoring but potentially also for reliability estimates of neural networkbased property predictors.     «

Additive manufacturing technologies have seen significant economic growth over the past years. Although additive manufacturing processes have matured in many areas, difficulties with regard to printing accuracy persist. Possible defects are, e.g., the generation of unwanted internal pores or a lack of fusion between layers, to name only a few. In general, defects result in a deviation between as-planned and as-built geometries, all of which can be difficult to detect in an automatized fashion. P...     »