AI-Powered Thermal Fingerprinting: Predicting PLA Tensile Strength Through Schlieren Imaging

Area of Science:

• Additive Manufacturing
• Materials Science
• Machine Learning

Background:

• Fused Deposition Modeling (FDM) prints exhibit unpredictable mechanical properties, necessitating advanced quality assurance.
• Current methods like destructive testing and post-process inspection are costly and inefficient.
• Existing machine learning models often overlook real-time thermal environments, focusing instead on printing parameters.

Purpose of the Study:

• To develop a low-cost, non-destructive framework for predicting tensile strength during FDM printing.
• To utilize real-time convective thermal gradients surrounding the print for quality prediction.
• To establish a methodological framework for real-time, non-contact quality assurance in FDM.

Main Methods:

• Introduced "thermal fingerprinting," combining Background-Oriented Schlieren (BOS) imaging with machine learning.
• Captured thermal gradient fields around PLA specimens (n=30) using consumer-grade equipment.
• Processed BOS imaging data from critical layers into features for machine learning analysis.

Main Results:

• Achieved 100% classification accuracy for controlled cooling conditions.
• Demonstrated promising initial correlations with tensile strength (R² = 0.808).
• Highlighted the need for larger datasets for robust generalization in machine learning models (five-fold cross-validation R² = 0.301).

Conclusions:

• This work is the first to apply Schlieren imaging to polymer additive manufacturing.
• Established a novel framework for real-time, non-contact quality prediction in FDM.
• The method enables on-the-fly identification of mechanically unreliable prints without interrupting production.