Metrology automation and geometric analysis for additive manufacturing

Additive manufacturing (AM) enables flexible, scalable production of parts that can have complex geometries; however, inspection of these parts is challenging with traditional measurement methods and technologies, which often have low sampling density and, in some cases, high labor intensity. This research introduces a machine vision measurement system and analysis methods capable of automatically extracting high density measurements from optical scan data. The system utilizes Canny edge detection and Hough transforms to detect and measure geometric features of interest. We employ the measurement system to extract strut geometries from lattice parts, and accurately predict their compression properties, which largely depend upon strut geometry. In a second study, we measure over 21,000 individual features from lattice parts fabricated using multiple, identical AM machines. We apply variance decomposition on part-level geometric features, and quantify the effects that using different machines, tools, and locations within the print area have on the geometric variance of strut length, thickness, and part height. We extend these methods to analyze the internal geometry of a batch of AM nozzle parts inspected using three-dimensional X-ray computed tomography (CT). The parts were made using 11 polymer materials and 3 AM processes. We demonstrate scalable CT metrology and extract over 100,000 measurements from 69 CT scans. We find that AM can have very high part-to-part repeatability, while feature accuracy can be poor. The manufacturing accuracy of internal and external features can differ substantially and highlights the need for analysis methods capable of automatic batch processing of CT and other types of scan data. The research concludes by investigating how automatic geometry measurements can be combined with machine learning (ML) to predict AM quality in the production of parts of different designs. We collect measurement data from 405 parts having three different designs with common features. Using recorded manufacturing conditions and feature descriptors, ML models predict feature geometries to within 55 µm, which is close to the limit of random manufacturing variation and measurement uncertainty. In addition to predicting geometry, tunable thresholds convert the regression predictions into classification models and enable more accurate acceptance or rejection of features and parts than traditional ML classification strategies. This research demonstrates opportunities for metrology automation in AM applications, and the methods are extensible to numerous processes, materials, part designs, and types of scan data. Flexible AM currently exhibits large variability and low accuracy, and detailed measurement data can help address and control these issues.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2022

Author dc:creator

• McGregor, Davis J

Contributors dc:contributor

• King, William P
• Tawfick, Sameh
• Lambros, John
• Shao, Chenhui

• Additive manufacturing
• Metrology
• Machine Learning
• Dimensional accuracy
• X-ray computed tomography
• Lattice
• Quality
• Part qualification
• Honeycomb
• Architectured materials

Statement dc:rights

• Copyright 2022 Davis McGregor

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/115653

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F115653