Automatic quantification of porosity using an intelligent classifier

Abstract

The porosity in manufactured components by additive manufacturing is a latent problem that leads to adverse effects on components such as fatigue. A correct segmentation and its subsequent classification allow us to identify its cause, either lack of fusion of particles or gases trapped during the process. There are several pores classifications described in the literature, but it is difficult to give a global classification. The present work describes a classification based on size, distribution, and origin of pores. For this purpose, the development of an artificial vision methodology is described that allows segmentation and classification of porosity with a high-accuracy rate in tracks manufactured by the laser metal deposition technique using commercial Al-5083 powders. The methodology is divided into 3 steps. (1) The first step consists of the image smoothing and denoising using a bilateral filtering. (2) A variant of the Hough transform has then implemented to segment the pores, and finally, (3) the automatic classification is performed by quadratic discriminant analysis (QDA) and Kohonen maps. The results obtained are compared with the manual classification of two materials experts. These results show an accuracy of + 95%. Our approach has the potential to be used in the analysis of any additively manufactured component.