Acoustic Signal-Based Defect Identification for Directed Energy Deposition-Arc Using Wavelet Time-Frequency Diagrams.

Area of Science:

• Additive Manufacturing
• Materials Science
• Acoustic Signal Processing

Background:

• Directed Energy Deposition-arc (DED-arc) offers high deposition rates and low costs.
• Defects like discontinuity and pores can occur during DED-arc manufacturing.
• Effective defect identification is crucial for quality control in additive manufacturing.

Purpose of the Study:

• To propose a novel acoustic signal-based method for defect identification in DED-arc additive manufacturing.
• To utilize wavelet time-frequency diagrams for defect analysis.
• To evaluate the performance of Convolutional Neural Network (CNN) models for defect detection.

Main Methods:

• Acquisition of in situ acoustic signals during DED-arc manufacturing.
• Conversion of 1D acoustic signals into 2D time-frequency diagrams using continuous wavelet transform.
• Training, validation, and testing of CNN models (AlexNet, ResNet-18, VGG-16, MobileNetV3) on the generated diagrams.

Main Results:

• The proposed method effectively identified defects in the DED-arc manufacturing process.
• MobileNetV3 achieved the highest accuracy at 98.31%, followed by ResNet-18 at 97.92%.
• Significant differences in energy distribution between normal and abnormal acoustic signals were observed in time and frequency domains.

Conclusions:

• The acoustic signal-based method with wavelet time-frequency diagrams is a viable approach for defect identification in DED-arc.
• The study demonstrates the potential of CNN models for real-time quality assessment in additive manufacturing.
• The proposed technique advances defect identification time, enabling quicker quality control interventions.