Related Experiment Video
Updated: May 20, 2025
08:15
Bead Aggregation Assays for the Characterization of Putative Cell Adhesion Molecules
Published on: October 17, 2014
10.5K
Machine learning image-based analysis for bead geometry prediction in fused granulate fabrication for large format
Daniele Vanerio1,2, Mario Guagliano1, Sara Bagherifard1
1Department of Mechanical Engineering, Politecnico di Milano, Milan, Italy.
NPJ Advanced Manufacturing
|March 25, 2025
Summary
This study uses an artificial neural network (ANN) to accurately predict cross-sectional geometry in fused granulate fabrication (FGF), a large-format additive manufacturing (LFAM) process. The ANN model enhances geometric precision for complex shapes in polymer-based additive manufacturing.
Area of Science:
- Materials Science
- Mechanical Engineering
- Computer Science
Background:
- Large-format additive manufacturing (LFAM) relies on polymer-based fused granulate fabrication (FGF) for producing large structures.
- Predicting and controlling the cross-sectional geometry of FGF parts is crucial for ensuring dimensional accuracy and part integrity.
- Existing methods often focus on contour prediction, limiting the ability to capture complex geometries.
Purpose of the Study:
- To develop and validate an artificial neural network (ANN) model for predicting the full cross-sectional geometry in FGF processes.
- To investigate the influence of critical process parameters on bead morphology.
- To advance geometric prediction capabilities in LFAM beyond contour-based approaches.
Main Methods:
- A full factorial design was employed to generate a comprehensive dataset by systematically varying layer height, transverse speed, and screw speed.
- Cross-sectional images of fabricated parts were acquired and processed for training the ANN.
- An ANN with two hidden layers was designed and integrated with image processing techniques to predict cross-sectional geometry.
Main Results:
- The ANN model demonstrated strong agreement with experimental cross-sections, achieving a mean absolute error of 8.88%.
- The model effectively captured the complex geometric features of the FGF bead profiles.
- The approach successfully predicted full cross-sectional images, outperforming methods limited to contour point prediction.
Conclusions:
- The developed ANN is effective for predicting FGF profiles, significantly enhancing geometric precision in LFAM.
- This predictive capability holds potential for generating more complex shapes and improving overall quality in polymer-based additive manufacturing.
- The study provides a foundation for advanced geometric control in large-format additive manufacturing technologies.
