Data Analysis for Defect Monitoring in Additive Manufacturing – Applying Machine Learning to Predict Porosity in L-PBF

Abstract

Laser powder bed fusion (L-PBF) is an additive manufacturing technique that sees more and more use in industrial settings, but is held back by a lack of cost-effective quality validation of created products. One core attribute of high-quality additive manufactured products is a low porosity, i.e. a high ratio of solid to empty volume inside the object. This thesis provides an overview of the state of the art for in-situ monitoring of L-PBF manufacturing and investigates the use of outlier detection methods as a way of encoding optical tomography data from an L-PBF process. This is done using a commercial L-PBF machine with its accompanying in-situ monitoring camera. The results show that outlier detection methods can be used to detect porosity in created objects (0.94 - 0.99 ROC-AUC, receiver operating characteristics’ area under curve) and that it can generalize between similar object geometries. The thesis also provides a discussion of the limitations of the current research and suggests future work both building upon the methods introduced in the thesis and in the field of in-situ monitoring of L-PBF.