Search NASA⌕ Search

NASA NTRS · 20230008993

Convolutional Autoencoder for Defect Detection in Additive Manufacturing

Abstract

The core idea behind using machine learning (ML) for defect detection is that it can be used to detect flaws as they are being formed in an AM part. As the part is being made, a near-infrared (NIR) sensor records each layer and creates an image of the entire build layer. These images, usually thousands, can be compiled into a ‘3D’ array of the entire part. ML tools, such as a convolutional autoencoder (CAE) can go through these images and highlight potential anomalous regions of your part.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

James Mavo, Aidan Faino, Daniel Vaughan, Raymond Carter, Erin Lanigan, Delphine Duquette. Convolutional Autoencoder for Defect Detection in Additive Manufacturing. https://ntrs.nasa.gov/citations/20230008993

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

On the Nature of Recoater Damage Powder Spreading Mechanics

Laser powder bed fusion (PBF-LB) additive manufacturing gained popularity for the creation of high mix, low volume parts for defense and commercial applications. Parts made via PBF-LB can be difficult to qualify due to their variation in performance, even within the same build. During the build process, regions of the PBF-LB recoater blade could be subject to damage by spatter or superelevation in the powder bed. As a result, the worn recoater can potentially cause spreading defects in the localized topography profile, which could in turn cause porosity or form deviations. These process concerns are not well understood for their criticality and subsequent impact on part quality. This study will investigate the mechanics of recoater damage through two experimental builds. Post-mortem inspection of the failure region will be conducted with laser line profilometry and will help propose a better understanding of the criticality of extreme recoater damage.

In-Situ Monitoring↗