Demonstrating Paraflow: Interactive fluid dynamics simulation with real-time visualization for augmented resin 3D printing

Access full-text files

Date

2023

Journal Title

Journal ISSN

Volume Title

Publisher

University of Texas at Austin

Abstract

While resin 3D printers are seeing growing adoption in both manufacturing and personal fabrication settings, detecting print failures in real time remains challenging. Object-detection neural networks have shown benefits in a variety of extrusion-based 3D printing methods. Here, we extend such work to resin printing using a physics-informed machine learning data generation pipeline. Our approach leverages our models of the fluid dynamics of the printing process at every slice, in order to synthetically generate a library of print defects. We show such an approach is capable of providing data sufficiently resembling real-world failures to fine-tune a pre-trained custom defect detection neural network that can alert users of failure in real-time. Finally, to allow novice users to take advantage of our simulation platform, we integrate our tool into an interactive augmented reality interface, which displays simulation predictions to provide guidance on design and machine parameters prior to printing.

Description

LCSH Subject Headings

Citation