About this Abstract |
Meeting |
2021 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2021)
|
Symposium
|
Special Session
|
Presentation Title |
Autoencoder-based Anomaly Detection for Laser Powder Bed Fusion |
Author(s) |
Bumsoo Park, Aleksandr Shkoruta, Sandipan Mishra |
On-Site Speaker (Planned) |
Bumsoo Park |
Abstract Scope |
This research proposes a convolutional autoencoder for anomaly detection in melt pool images for laser powder bed fusion (L-PBF). Generally, image-based anomaly detection requires image filters that are manually engineered, and thus may require a large amount of engineering time. With autoencoders however, manually pre-selected features or laborious labelling of a large dataset are not required, as this machine-learning approach enables the unsupervised dimensionality reduction of high-speed melt pool images. Moreover, the distribution of lower-dimensional encoded image values forms clusters within the lower-dimensional feature space. Thus, anomaly detection can be performed by evaluating the encoded values of a newly acquired image with respect to existing clusters in data. The proposed algorithm is validated using experimental data from an instrumented L-PBF testbed, demonstrating capabilities of detecting process and sensor anomalies such as powder spattering or overheating. |
Proceedings Inclusion? |
Definite: Post-meeting proceedings |