Multi-Output Prediction and Optimization of CO₂ Laser Cutting Quality in FFF-Printed ASA Thermoplastics Using Machine...

Area of Science:

• Materials Science and Engineering
• Manufacturing Technology
• Additive Manufacturing

Background:

• Fused Filament Fabrication (FFF) is a key additive manufacturing technique for thermoplastics like Acrylonitrile Styrene Acrylate (ASA).
• Laser cutting is a crucial post-processing step for FFF parts, but its optimization for ASA requires understanding complex parameter interactions.

Purpose of the Study:

• To investigate the CO2 laser cutting performance of ASA thermoplastics produced via FFF.
• To analyze the impact of plate thickness, laser power, and cutting speed on surface roughness, kerf width, and heat-affected zone.
• To develop and validate a machine learning framework for predicting laser cutting quality characteristics.

Main Methods:

• Conducted 45 experiments varying ASA plate thickness, CO2 laser power, and cutting speed.
• Employed seven machine learning models including Autoencoder variants, Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Regression, and Linear Regression.
• Utilized Analysis of Variance (ANOVA) to determine the significance of process parameters on quality metrics.

Main Results:

• Extreme Gradient Boosting (XGBoost) demonstrated the highest prediction accuracy across all quality metrics.
• Plate thickness significantly influenced surface roughness (Ra).
• Cutting speed was the primary factor for bottom kerf width (Bottom KW), while laser power dominated the bottom heat-affected zone (Bottom HAZ).
• Top kerf width (Top KW) was affected by all three parameters.

Conclusions:

• A robust prediction framework using multi-output modeling and hybrid deep learning (specifically XGBoost) was established for laser cutting quality.
• The findings provide a data-driven basis for optimizing the CO2 laser cutting of ASA and similar thermoplastics.
• This research supports intelligent manufacturing by enabling real-time quality prediction and adaptive laser post-processing.