Predictive Models for the Characterization of Internal Defects in Additive Materials from Active Thermography...

Area of Science:

• Materials Science
• Non-Destructive Testing
• Computational Modeling

Background:

• Additive manufacturing (AM) enables complex geometries but poses challenges in defect detection.
• Internal defects can compromise the structural integrity of AM components.
• Active transient thermography is a promising non-destructive testing (NDT) method for subsurface defect characterization.

Purpose of the Study:

• To develop and compare predictive models for assessing the thickness and length of internal defects in AM materials.
• To evaluate the performance of various machine learning algorithms integrated with finite element simulation.
• To identify optimal predictive methods for defect characterization using thermographic data.

Main Methods:

• Generation of predictive models using data from active transient thermography numerical simulations.
• Integration of finite element (FE) simulation with machine learning (ML) models, including regression, Gaussian regression, support vector machines (SVM), multilayer perceptron (MLP), and random forest (RF).
• Statistical analysis and comparison of model performance based on predictive accuracy, processing time, and outlier sensitivity.

Main Results:

• Interaction linear regression was the best model for predicting defect thickness using six thermal features.
• Gaussian process regression excelled in predicting both defect length and thickness.
• Support vector machines demonstrated significant advantages in processing time and performance for specific feature sets.

Conclusions:

• The developed predictive models, leveraging active thermography and machine learning, can effectively detect and measure internal defects in additive manufacturing materials.
• The choice of the best predictive model depends on specific requirements, balancing accuracy, processing time, and feature set.
• This hybrid approach offers a viable non-destructive testing solution for quality control in additive manufacturing.