HMQ-ES-Stack-GBR: A Hybrid Ensemble Learning Model for Mechanical and Physical Quality Prediction in FDM 3D Printing

Area of Science:

• Materials Science and Engineering
• Additive Manufacturing
• Machine Learning Applications

Background:

• Process parameters significantly influence the mechanical and physical properties of Fusion Deposition Modeling (FDM) prints.
• Accurate prediction of 3D print quality is crucial for optimizing manufacturing processes and material usage.
• Machine learning (ML) offers a powerful approach for predicting print quality based on various input factors.

Purpose of the Study:

• To develop and validate a machine learning framework for predicting multi-output quality characteristics of 3D prints.
• To quantitatively analyze the impact of material type on print quality.
• To integrate optimization algorithms for deriving optimal processing parameters for different materials.

Main Methods:

• A comprehensive dataset was generated under strict standards, including data sanitization using the Interquartile Range (IQR) method.
• Tensile, hardness, and surface roughness tests were performed on 500 unique 3D printed sample combinations across 10 material types.
• A Hybrid Multi-Material Quality-Ensemble System-Stacking-Gradient Boosting Regressor (HMQ-ES-Stack-GBR) architecture was employed for quality prediction.

Main Results:

• The proposed HMQ-ES-Stack-GBR framework demonstrated effective multi-output quality prediction for tensile strength, hardness, and surface roughness.
• A multi-objective optimization pipeline, integrating Non-dominated Sorting Genetic Algorithm II (NSGA-II), Particle Swarm Optimization (PSO), and Grey Wolf Optimizer (GWO), successfully derived material-specific optimal parameters.
• Open-system printers showed higher prediction errors compared to closed-system printers, indicating system-induced variability.

Conclusions:

• The developed framework serves as a valuable diagnostic decision-support tool for pre-print quality estimation in FDM.
• The study highlights the potential for proactive interventions to minimize material, time, and energy losses in 3D printing.
• The findings underscore the importance of considering system-specific characteristics in print quality prediction and optimization.