Deep Learning-Assisted Porosity Assessment for Additive Manufacturing Components Using Ultrasonic Coda Waves

Area of Science:

• Materials Science
• Non-Destructive Testing
• Artificial Intelligence

Background:

• Porosity in additive manufacturing components critically affects mechanical properties, hindering engineering applications.
• Current porosity assessment methods are primarily destructive, necessitating advanced non-destructive testing (NDT) solutions.
• Accurate in situ NDT is crucial for reliable quality control in additive manufacturing.

Purpose of the Study:

• To develop a novel deep learning-assisted NDT method for precise porosity assessment in additive manufacturing components.
• To overcome limitations of conventional NDT parameters in mapping complex porosity features.
• To enhance the accuracy and reliability of porosity evaluation in additively manufactured parts.

Main Methods:

• Utilized ultrasonic coda waves for their high sensitivity to internal material variations and porosity.
• Employed a deep learning approach, specifically a coda-convolutional neural network with a multi-head attention mechanism.
• Integrated ultrasonic coda wave analysis with deep learning for enhanced feature extraction and complex mapping.

Main Results:

• Ultrasonic coda waves demonstrated sensitivity to porosity variations in additive manufacturing components.
• The deep learning network effectively extracted porosity-related features from ultrasonic coda wave signals.
• The proposed method achieved a high prediction accuracy of 98% for component porosity.

Conclusions:

• Ultrasonic coda waves and deep learning offer a complementary approach for porosity assessment.
• This integrated framework overcomes challenges in feature extraction and complex relationship mapping.
• The developed method provides a new, high-precision solution for non-destructive testing of additive manufacturing components.