Comparative Study of Machine Learning Models for Optimal Prediction of Printed-Line Features in Material Extrusion...

Area of Science:

• Additive Manufacturing
• Materials Science
• Machine Learning

Background:

• Material extrusion (MEX), or fused deposition modeling (FDM), is a popular 3D printing method due to its affordability and material versatility.
• However, defects like edge non-uniformity and variable line width limit its use in high-precision applications.
• Improving geometric fidelity is crucial for advancing MEX technology.

Purpose of the Study:

• To develop a machine learning framework for predicting printed line characteristics in MEX.
• To analyze the impact of process parameters on line width and edge non-uniformity.
• To establish a foundation for closed-loop quality control in additive manufacturing.

Main Methods:

• Utilized four machine learning algorithms: XGBoost, BPNN, GPR, and SVR.
• Optimized model hyperparameters using Particle Swarm Optimization.
• Conducted SHAP-based interpretability analysis to identify key process parameters.

Main Results:

• Gaussian Process Regression (GPR) was identified as the optimal predictive model.
• Nozzle temperature was found to be the dominant factor for line width.
• Material flow rate significantly influenced edge non-uniformity.

Conclusions:

• The proposed machine learning framework enables rapid prediction and analysis of printed line quality.
• Interpretable surrogate modeling enhances understanding of process-quality relationships.
• This approach supports closed-loop quality control and inverse process design for precision additive manufacturing.