A Data-Driven Approach to Complex Voxel Predictions in Grayscale Digital Light Processing Additive Manufacturing...

Area of Science:

• Materials Science
• Computer Science

Background:

• Digital Light Processing (DLP) is a key additive manufacturing technique.
• Predicting and controlling voxel geometry is crucial for precision in DLP.

Purpose of the Study:

• To develop and validate machine learning models for predicting 3D printed voxel geometry in DLP.
• To explore the application of U-net and pix2pix conditional generative adversarial networks (cGANs) for enhanced precision.

Main Methods:

• Utilized a confocal microscopy workflow for high-throughput data acquisition of voxel interactions.
• Trained pix2pix cGAN models on data from randomly gray-scaled digital photomasks.
• Validated model predictions against actual 3D prints, assessing sub-pixel resolution accuracy.

Main Results:

• Machine learning models demonstrated accurate predictions of 3D printed voxel geometry with sub-pixel resolution.
• The trained cGAN successfully performed virtual DLP experiments, including cure depth and anti-aliasing.
• The pix2pix model showed applicability to larger masks than those used in training and could inform print failure analysis.

Conclusions:

• Data-driven machine learning, particularly U-nets and cGANs, shows significant promise for predicting and correcting photomasks in DLP additive manufacturing.
• This methodology can lead to increased precision and improved quality control in 3D printing processes.