"Automated 3D Printing Defect Detection and Correction Using Machine L" by Taylor King

Automated 3D Printing Defect Detection and Correction Using Machine Learning

Author

Taylor King

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science and Info Sys

Date of Award

Fall 2024

Abstract

Three-dimensional (3D) printing is an innovative additive manufacturing technique that creates a 3D physical product from a digital design using layer-by-layer addition, enabling the possibility of the creation of complex geometries with high precision and customization. In recent years, it has been developing very quickly, with new technologies and applications emerging across a variety of industries including Aerospace, automotive, manufacturing, robotics, and education. Along with the growth in the adoption and integration of 3D printing across various industries, there is a pressing need to have high-quality 3D printing. Even a minor printing error and/or defect can compromise the project quality and result in substantial waste and added costs from reprints, material losses, and production delays. However, less attention has been given to the understanding of 3D printing defects and defect detection. Additionally, the absence of defect data has created a critical research gap, limited insights, and hindered the progress of research efforts. This research focuses on developing an automated defect detection system using machine learning. The study defines key 3D printing defects and correlations with their underlying causes, including hardware issues (e.g., nozzle clogs, misalignment), material problems (e.g., filament moisture), software misconfigurations (e.g., incorrect print speeds, G-code errors), environmental factors (e.g., temperature fluctuations), and user errors (e.g., improper calibration). Based on it, we then build a 3D printing system as a prototype to fill the identified gaps by collecting real 3D printing data and enabling automated 3D printing defect detection. Using both newly collected data and publiclyavailable datasets, eight machine learning algorithms (Alexnet, ConvNeXt, EfficientNet, GoogleNet, Resnet, Swin Transformer, VGG Net, and Vision Transformer) are trained and tested to detect defects, even on incomplete prints. The results demonstrate the potential for real-time error detection and correction, shedding light on future autonomous 3D printing systems for defect correction with minimal human intervention.

Advisor

Yuehua Wang

Subject Categories

Computer Sciences | Physical Sciences and Mathematics

Share

COinS