Image sensors, e.g., IR camera, can be installed into advanced manufacturing processes for real-time imaging data collection. The acquired data are form a thermal video, which reflects the entire build process of the manufactured object. The spatial structure of the object, as well as its temporal evolution (until completion), are all reflected by the video. The video data are therefore crucial resources for learning the defect formulation and process anomalies. It is noteworthy, however, defect and anomaly prediction in such thermal video data cannot be done with conventional statistical/machine learning methods. During the build process, the likelihood of having defects and anomalies evolves with time. This is a physical phenomenon that is not considered in conventional use of machine learning. To handle this issue, an online learning method must be developed for defect/anomaly prediction. When used for an ongoing manufacturing process, this method should be able to update itself with incoming thermal images and learn new information about defects, thus ensuring accurate predictions.
Please apply through FURI or MORE and then contact the faculty via email. Resume requested.