About this Abstract |
Meeting |
2021 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2021)
|
Symposium
|
Special Session
|
Presentation Title |
Automated Anomaly Detection of Laser-based Additive Manufacturing Using Melt Pool Sparse Representation and Unsupervised Learning |
Author(s) |
Xiyue Zhao, Aidin Imandoust, Mojtaba Khanzadeh, Farhad Imani, Linkan Bian |
On-Site Speaker (Planned) |
Farhad Imani |
Abstract Scope |
Advanced thermal imaging is increasingly invested in additive manufacturing (AM) to improve the information visibility of meltpool and cope with process inconsistency. However, there are key challenges regarding the feasibility of current data-driven monitoring methodologies. First, high-resolution thermal images consist of millions of pixels captured by hundreds of frames, thereby leading to the curse of dimensionality in analysis. Second, generated data lack labels, which is a stumbling block in training data-driven models. The objective of this research is to advance the frontier of meltpool monitoring in metal AM by designing an automated and unsupervised anomaly detection on high-dimensional thermal data. We develop a deep variational autoencoder to generate a low-dimensional representation and reconstruction error of each input thermal image data. A novel Gaussian mixture model is integrated with the generative model to split latent space into homogenous regions and detect information of anomalies. |
Proceedings Inclusion? |
Definite: Post-meeting proceedings |