Abstract Scope |
Machine Learning (ML) using big data and physics knowledge provides Additive Manufacturing (AM) with high potential to extract newfound knowledge, such as causality between processes, structures, and properties, from real-world data. However, this potential often relies on increased model complexity for a surge in performance, which turns ML models into “black boxes”. Such black-box models lead to uncertainty regarding the knowledge ML extracts and, eventually, decisions the knowledge supports for AM. This uncertainty is a challenge in adopting ML in high-value yet critical AM applications. To tackle the challenge, this study addresses interpretation of ML models for AM in the perspective of explainable AI. This study focuses on a literature review to classify related methods and identify their potential in design, process control, and part evaluation for AM. This study sheds light on extraction of knowledge from ML models and transfer of the knowledge at multiple scales in various applications. |