Bayesian optimization with Gaussian-process-based active machine learning for improvement of geometric accuracy in...

Area of Science:

• Additive Manufacturing
• Machine Learning
• Optical Engineering

Background:

• Multi-photon polymerization (MPP) is a key micro/nanoscale additive manufacturing technique.
• Optimizing process parameters for dimensional accuracy in MPP is challenging and time-consuming.
• Existing methods often require extensive experimental trials.

Purpose of the Study:

• To develop an active machine learning framework for optimizing process parameters in high-speed, continuous projection 3D printing.
• To reduce the experimental data collection effort required for achieving high geometric accuracy.
• To create a surrogate model for predicting optimal parameters for target geometries.

Main Methods:

• Utilized an active learning framework with Bayesian optimization.
• Employed Gaussian-process-regression for the machine learning model.
• Tested the framework on three representative 2D shapes at different scales.

Main Results:

• Achieved significant improvements in geometric accuracy.
• Reduced errors to within measurement accuracy.
• Demonstrated effectiveness in just four Bayesian optimization iterations with limited data.

Conclusions:

• The active learning framework efficiently optimizes additive manufacturing processes.
• It drastically reduces experimental effort for parameter optimization.
• The approach is broadly applicable to various additive manufacturing techniques.