Hybrid Experimental-Machine Learning Study on the Mechanical Behavior of Polymer Composite Structures Fabricated via
Osman Ulkir1, Sezgin Ersoy2,3
1Department of Electric and Energy, Mus Alparslan University, Mus 49210, Türkiye.
Polymers
|August 14, 2025
Summary
Fused Deposition Modeling (FDM) specimens showed varying mechanical strengths based on material, infill, and print direction. Carbon fiber-reinforced polyphthalamide (PPA/Cf) printed flat offered superior tensile and flexural strength compared to ABS.
Area of Science:
- Materials Science
- Additive Manufacturing
- Mechanical Engineering
Background:
- Fused Deposition Modeling (FDM) is a widely used additive manufacturing technique for polymers and composites.
- Optimizing FDM process parameters is crucial for achieving desired mechanical properties in printed parts.
- Understanding the influence of material type, infill pattern, and printing direction is key to material performance.
Purpose of the Study:
- To investigate the mechanical behavior (tensile and flexural strength) of FDM-printed specimens.
- To evaluate the impact of material type, infill pattern, and printing direction on mechanical performance.
- To develop and compare machine learning models for predicting mechanical properties.
Main Methods:
- Specimens of Acrylonitrile Butadiene Styrene (ABS), PPA/Cf, and a sandwich composite were fabricated using FDM.
- Box-Behnken Design (BBD) was employed to systematically study the effects of material type (MT), infill pattern (IP), and printing direction (PD).
- Bayesian Linear Regression (BLR) and Gaussian Process Regression (GPR) were utilized for predictive modeling, with results compared against BBD.
Main Results:
- PPA/Cf material printed 'Flat' with a 'Cross' infill pattern exhibited the highest mechanical strength (75.8 MPa tensile, 102.3 MPa flexural).
- ABS printed 'Upright' with a 'Grid' pattern showed the lowest strength (37.8 MPa tensile, 49.5 MPa flexural).
- Gaussian Process Regression (GPR) demonstrated superior prediction accuracy (R² > 0.99) and lower errors (MAPE < 13%) compared to BBD and BLR.
Conclusions:
- Material type, infill pattern, and printing direction significantly impact the mechanical properties of FDM parts.
- PPA/Cf offers superior mechanical performance in FDM compared to ABS.
- Machine learning models, particularly GPR, provide highly accurate predictions of mechanical behavior, outperforming traditional design of experiments methods.
