Co-learning of Extrusion Deposition Quality for Supporting Interconnected Additive Manufacturing Systems

Abstract

Additive manufacturing systems are being deployed on a cloud platform for providing networked manufacturing services. This paper explores the value of interconnected printing systems that share process data on the cloud in improving quality control. We employed an example of quality learning for cloud printers by understanding how printing conditions impact printing errors. Traditionally, extensive experiments are necessary to collect data and estimate the relationship between printing conditions vs. quality. This research establishes a multi-printer co-learning methodology to obtain the relationship between the printing conditions and quality using limited data from each printer. Based on multiple interconnected extrusion-based printing systems, the methodology is demonstrated by learning the printing line variations and resultant infill defects induced by extruder kinematics. The method leverages the common covariance structures among printers for the co-learning of kinematics-quality models. This paper further proposes a sampling-refined hybrid metaheuristic to reduce the search space for solutions. The results showed significant improvements in quality prediction by leveraging data from data-limited printers, an advantage over traditional transfer learning that transfers knowledge from a data-rich source to a data-limited target. The research establishes algorithms to support quality control for reconfigurable additive manufacturing systems on the cloud.