Co-Learning of Extruded Deposition Quality for Developing Interconnected Additive Manufacturing Systems

Fused deposition modeling (FDM) is a cost-effective way of flexibly producing a high variety of 3D structures. However, the deposition non-uniformity or variation in the FDM process leads to dimensional errors and quality problems such as infill defects and non-uniform material distribution. It is essential to calibrate a printer, aiming to adjust its printing setup such as G-code that controls extruders’ kinematics to compensate for the errors due to deposition non-uniformity. Such a calibration process requires extensive experiments to collect data and thereby estimate the relationship between the printers’ setup and printing quality. This thesis research explores the use of multiple inter-connected printers that can exchange and share quality data to co-learn quality problems to expedite the printer calibration process and reduce ramp-up time. State-of-the-art research focuses on shape error compensation and between-printer transfer learning to leverage knowledge from data-rich printer to data-lacking printer. Very limited research has been developed to address the multi-printer co-learning problem that concerns with fusing the information from many data-lacking printers for in-process quality control. The proposed work in this thesis includes (1) validation of between-printer similarity patterns to justify the multi-printer co-learning, (2) estimation of non-uniformity of material deposition induced by extruders’ kinematics to capture the between-printer relatedness for co-learning, (3) development of the co-learning of material deposition quality for a pair of data-lacking printers and study the applicability conditions, and (4) extension of the co-learning algorithms to multiple data-lacking printers for improved prediction printing errors. The success of this research establishes the theoretical grounds to support reconfigurable additive manufacturing systems and cost-effective mass personalization deployed on a cloud platform that can responsively cope with high product variety. In addition, the broader impact of the study can also be expected to lower the entry barrier for new manufacturers by leveraging online data. Keywords: Fused deposition modeling (FDM), additive manufacturing, quality control, multi-printer co-learning, transfer learning, extruders’ kinematics