Enhancing the 3D printing fidelity of vat photopolymerization with machine learning-driven boundary prediction

Abstract

Like many pixel-based additive manufacturing (AM) techniques, digital light processing (DLP) based vat photopolymerization faces the challenge that the square pixel based processing strategy can lead to zigzag edges especially when feature sizes come close to single-pixel levels. Introducing greyscale pixels has been a strategy to smoothen such edges, but it is a challenging task to understand which of the many permutations of projected pixels would give the optimal 3D printing performance. To address this challenge, a novel data acquisition strategy based on machine learning (ML) principles is proposed, and a training routine is implemented to reproduce the smallest shape of an intended 3D printed object. Through this approach, a chessboard patterning strategy is developed along with an automated data refining and augmentation workflow, demonstrating its efficiency and effectiveness by reducing the deviation by around 30%.