Deep-Learning Quantitative Structural Characterization in Additive Manufacturing
This addresses the need for precise microstructure characterization in additive manufacturing, but it is incremental as it builds on existing image-to-image translation techniques.
The researchers tackled the problem of accelerating additive manufacturing by developing a deep learning method for fast and accurate prediction of geometric features like melt pool boundaries and defects from optical images, enabling real-time control.
With a goal of accelerating fabrication of additively manufactured components with precise microstructures, we developed a method for structural characterization of key features in additively manufactured materials and parts. The method utilizes deep learning based on an image-to-image translation conditional Generative Adversarial Neural Network architecture and enables fast and incrementally more accurate predictions of the prevalent geometric features, including melt pool boundaries and printing induced defects visible in etched optical images. These structural details are heterogeneous in nature. Our method specifies the microstructure state of an additive built via statistical distribution of structural details, based on an ensemble of collected images. Extensions of the method are proposed to address Artificial Intelligence implementation of developed machine learning model for in real time control of additive manufacturing.