Hybrid Supervised and Self-Supervised Learning for 3D Printing Optimization: A Masked Supervised Bootstrap Your Own Latent Approach

Abstract

This study introduces an advanced method that combines supervised and self-supervised learning to optimize the 3D printing process, based on the Bootstrap Your Own Latent technique. The approach utilizes two neural networks: an online network and a target network. The target network is provided with complete data, while the online network processes incomplete data through a “masking” technique applied to certain parameters. The main goal is to ensure that both networks can accurately predict 3D printing time, even with missing data, achieving over 90% accuracy. Experiments show that masking the “faces” parameter of the 3D model reduces prediction accuracy by 56.93%, highlighting the importance of this parameter in optimizing printing time. By combining two loss functions, cross-entropy and consistency loss, this approach enhances interaction between the networks while maintaining high prediction accuracy despite incomplete input data.