An Analysis of Lateral Dimension Growth in Stereolithography Additive Manufactured Alumina Using Computational Image Analysis

Stereolithography 3-D printing, a form of additive manufacturing, has been the focus of recent research in the field of materials science due to its ability to manufacture complex structures with little human input. SLA uses a liquid polymer that hardens upon exposure to ultraviolet (UV) or ultraviolet-visible (UV-Vis) radiation, a process known as photopolymerization. To be more detailed, a ceramic powder of choice, such as aluminum oxide (Al2O3) or silicon nitride (Si3N4) is suspended in a polymeric resin which crosslinks upon exposure to a beam of UV/UV-Vis light, locking up and solidifying the system, before loose powder is finally extracted from the print. One trait of stereolithography that little is known about is the nature of lateral dimension growth between the input 3-D model and the output print. In other words, the physical printed part will demonstrate larger dimensions in the lateral directions compared to the originally inputted dimensions. This study aims to take a systematic approach to identify any potential mechanisms or patterns that relate to lateral dimension growth in Al2O3 green bodies. The term "green body" refers to a stage in the stereolithography additive manufacturing process. After photopolymerization of the ceramic suspension, the resulting hardened print is considered a green body, and in practical applications, additional processing steps are taken beyond the green body stage. This study focuses exclusively on Al2O3 green bodies, using scanning electron microscopy to compare secondary electron images of the bulk microstructure to images of the microstructure of the edges of each layer in the samples to uncover the mechanisms behind lateral growth in stereolithography. To aid in this analysis, deep learning (DL), a subset of machine learning (ML), is used to expedite the data analysis process and make data-driven predictions on how the microstructure of other printed models may appear, leveraging algorithms like k-means and k-nearest-neighbors. Ultimately, some trends appeared in the data, but the statistics were not significant enough to draw definitive conclusions. Instead, these trends have provided potential areas that could be further explored in future research. Most importantly, this project has demonstrated the data analysis capabilities of ML, analyzing a myriad of samples in the fraction of the time a human alone would require.