Date of Award
Thesis - Open Access
Master of Science in Aerospace Engineering
First Committee Member
Second Committee Member
Additive manufacturing (AM) is a rapidly growing industry with numerous applications in the aerospace industry such as aircraft parts and emergency tools on the International Space Station. Defects in additively manufactured structures, however, can waste a lot of time and money. Being able to monitor the manufacturing process for defects is one of the first steps which can be taken to mitigate these losses. This study focuses on the use of thermography in conjunction with deep learning to identify flaws during 3D printing of composite structures made using Onyx, a mixture of chopped carbon fiber and nylon, composite prints. In addition, polymeric structures using polylactic acid (PLA) were analyzed using thermography and digital image correlation (DIC) to understand the interactions between the thermal variations and resulting deformation.
The inclusion of a zero-bias deep neural network (ZBDNN) to classify given images can also show real-time monitoring of defects in composite prints as a realizable goal. The ZBDNN was trained to classify thermal images of undamaged prints based on which layer of the print they occurred on and to set aside any of these images containing defects. The addition of a non-bias layer in the deep neural network ensures the classifications of these images remain consistent and accurate, with a learning accuracy of over 90%. The algorithm was also used to analyze grayscale images from multiple angles of the prints and compared these images to thermal images as another means of detecting defects in each print. The use of these multiple data sources may be used as the basis for an early-warning detection system for real-time analysis.
Scholarly Commons Citation
Phillips, Nicholas, "In-situ Thermal and Deformation Characterization of Additive Manufacturing Processes" (2023). Doctoral Dissertations and Master's Theses. 756.