About this Abstract |
Meeting |
Materials Science & Technology 2020
|
Symposium
|
Additive Manufacturing: Equipment, Instrumentation and Measurement
|
Presentation Title |
Machine Learning Enabled Acoustic Monitoring for Flaw Type Detection in Laser Powder Bed Additive Manufacturing |
Author(s) |
Brandon Abranovic, Wentai Zhang, Haiguang Liao, Jack Lee Beuth, Levent Burak Kara, Qingyi Dong |
On-Site Speaker (Planned) |
Brandon Abranovic |
Abstract Scope |
This work focuses on the analysis of acoustic data as a means to monitor laser powder bed additive manufacturing processes for key outcomes. This is of interest as it enables robust quality assurance, control and optimization of component properties, and improvement of process stability while reducing operator burden. Process mapping for Ti-6Al-4V was employed in determining parameter sets that would reliably induce keyholing, lack-of-fusion, bead-up, as well as a fully dense component. Using acoustic data collected during builds using these parameter sets, bag of words (BOW), support vector machines (SVM) and convolutional neural networks were evaluated for their performance in effectively classifying porosity flaws. Preliminary results have shown that these methods are able to reliably distinguish between the classes of interest. In future work, the application of recurrent neural networks (RNN) such as long-short term memory (LSTM) networks will be assessed for their viability against CNNs for baseline testing. |