Abstract
Recent advancements in Digital Light Processing (DLP) 3D printing technology have led to improved precision and speed, along with the development of specialized resin materials catering to diverse industry needs. Like other pixel-based additive manufacturing techniques, challenges persist in achieving DLP printing fidelity due to hardware limitations, particularly in reproducing structures smaller than one pixel. The integration of grayscale pixels has been a latest strategy to address this issue, but optimal allocation remains a challenge.Photopolymerization-based 3D printing methods often involve direct or predictive modeling of the cure process to anticipate ink solidification under varying UV intensities, typically utilizing inverse methods to determine the required light distribution. While effective, these models typically require measurement of ink cure degree or mechanical performance, necessitating additional time-consuming and intricate characterization steps. Moreover, the computational demands of these simulations often call for high-performance computing resources, adding complexity to the process.
To simplify the allocation of grayscale pixels for optimal light distribution, the application of machine learning (ML) has emerged as a promising solution. ML offers a data-driven approach to swiftly identify optimal configurations, reducing costs and enhancing printing efficiency. Researchers have successfully employed ML techniques to automate dataset refinement, mitigate scattering effects, improve printing accuracy for large-scale features, and efficiently obtain stress-strain curves. Recent advancements have demonstrated the efficacy of neural network-based ML models in predicting projection patterns, leading to significant enhancements in vat photopolymerization 3D printing precision. While these approaches show promise, some may require substantial characterization data and computational resources that may not be universally available in all research settings.
To overcome these challenges, a data-driven approach leveraging ML techniques has been proposed in this thesis. By using ML algorithms to predict grayscale pixel allocation, this approach aims to streamline the printing workflow, optimize material usage, and enhance printing precision. Through the implementation of a chessboard patterning strategy and an automated data refining and augmentation workflow, the proposed approach demonstrates increased efficiency and effectiveness. This advancement in DLP 3D printing technology, coupled with specialized resin materials catering to diverse industry needs, signifies a significant progression in the field. The integration of ML techniques not only enhances accuracy and efficiency in vat photopolymerization but also facilitates easy sharing and utilization of the trained model across multiple users without the need for extensive computing resources.
| Date of Award | Nov 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Yinfeng He (Supervisor), Haonan Li (Supervisor) & Gongyu Liu (Supervisor) |