Multi-Harmonic Nonlinear Ultrasonic Fusion with Deep Learning for Subtle Parameter Identification of Micro-Crack...

Area of Science:

• Materials Science
• Mechanical Engineering
• Non-Destructive Testing

Background:

• Fatigue cracks in metallic materials critically shorten component lifespan.
• Current methods often focus on single, larger cracks, overlooking micro-crack clusters.
• Identifying subtle parameters of micro-crack groups from complex ultrasonic signals is challenging.

Purpose of the Study:

• To develop a novel method for identifying subtle parameters of micro-crack groups.
• To integrate multi-harmonic nonlinear ultrasonic responses with a deep learning model.
• To enhance the accuracy of non-destructive testing for fatigue crack defects.

Main Methods:

• Training a 1D convolutional neural network (1D CNN) with finite element method (FEM) signals.
• Analyzing harmonic nonlinear ultrasonic responses for sensitivity to micro-crack parameters.
• Fusing high harmonic responses for decoupled identification of multiple parameters.
• Enhancing Dempster-Shafer (DS) evidence theory with sensitivity considerations and sensor weighting for decision fusion.

Main Results:

• Achieved 93.73% identification accuracy using fused high harmonics and enhanced DS theory.
• Further improved accuracy to 95.68% by incorporating sensor weighting based on a novel conflict measurement.
• Demonstrated the effectiveness of the deep learning approach in analyzing complex ultrasonic data.

Conclusions:

• The proposed multi-harmonic nonlinear response fusion method with deep learning offers a robust solution for micro-crack group parameter identification.
• The integration of enhanced DS theory and sensor weighting significantly boosts identification accuracy.
• This approach advances non-destructive testing capabilities for assessing the remaining useful life of metallic components.