Online flaw detection in metal additive manufacturing using deep learned acoustic features

Developed a pipeline for training deep learning models to detect micro-scale flaws in additive manufacturing processes.
Implemented two deep learning techniques: Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks.
Focused on detecting specific flaw formation mechanisms: keyhole porosity, lack of fusion, and bead up.
Achieved over 99% classification accuracy on unseen test sets for both CNN and LSTM approaches.
Utilized acoustic process monitoring data as input for the machine learning models.
Demonstrated the potential of machine learning-enabled acoustic monitoring as a viable replacement or complementary tool for traditional quality assurance methods in additive manufacturing.
This project showcases the application of advanced deep learning techniques to solve a critical challenge in additive manufacturing, potentially revolutionizing quality assurance processes in the industry.

Flaw Detection in Metal Additive Manufacturing Using Deep Learned Acoustic Features. Zhang, W., Hanson-Regalado, J., Koz, C., Duvvuri, B., Beuth, J., Shimada, K., & Kara, L. B. In Machine Learning for Engineering Modeling, Simulation, and Design Workshop at Neural Information Processing Systems 2020, 2020. NeurIPS.