Fast Image Segmentation Toward Automation of 3D Ice Printing

Area of Science:

• Additive Manufacturing
• Materials Science
• Computer Vision

Background:

• Freeform 3D ice printing is an emerging additive manufacturing technique with broad applications.
• Current manual design processes for complex ice geometries are time-consuming and empirical.
• Real-time visualization and control are crucial for improving ice printing precision and efficiency.

Purpose of the Study:

• To develop vision techniques for real-time monitoring and control of 3D ice printing.
• To enable closed-loop control for enhanced precision in ice additive manufacturing.
• To facilitate the automated design and fabrication of complex ice structures.

Main Methods:

• Implemented convolutional filters for ice segmentation to create training datasets.
• Utilized a hybrid optical flow algorithm (Farneback-FAST) for video frame segmentation.
• Trained a neural network (Icenet) using segmented video data for rapid frame analysis.

Main Results:

• Developed a machine learning approach for ice segmentation.
• Achieved frame segmentation in 25 ms, enabling single and low multi-droplet control.
• Demonstrated a viable method for generating actionable data from ice printing processes.

Conclusions:

• The implemented vision techniques enable future closed-loop control of 3D ice printing.
• This advancement is critical for automating complex ice structure fabrication.
• Potential applications include advanced additive manufacturing, organ-on-a-chip systems, and biomanufacturing.