Automated crack identification in structures using acoustic waveforms and deep learning.

Area of Science:

• Civil Engineering
• Materials Science
• Artificial Intelligence

Background:

• Structural elements are susceptible to damage from environmental factors and critical loads.
• Acoustic emission (AE) waveform analysis is crucial for detecting microcracks but often relies on subjective feature selection.
• Automating damage assessment in structures is essential for safety and maintenance.

Purpose of the Study:

• To develop and validate a deep-learning model for automated prediction of damage severity and location in structural elements.
• To leverage acoustic emission (AE) waveforms for enhanced structural health monitoring.
• To overcome the limitations of traditional, subjective AE analysis methods.

Main Methods:

• A densely connected convolutional neural network (CNN) was employed for superior feature extraction from time-domain AE waveforms.
• The deep learning model was trained and validated using AE data from concrete and wooden structural elements.
• The model directly processed raw AE waveforms, eliminating the need for manual feature engineering.

Main Results:

• The proposed deep learning model accurately predicted damage severity with 92-95% accuracy.
• The model achieved 90-100% accuracy in identifying the approximate location of damage in tested structural elements.
• Validation on concrete and wooden beams and plates demonstrated the method's robustness.

Conclusions:

• The developed deep learning approach provides a robust and automated technique for damage severity prediction and localization in civil structures.
• This method enhances structural health monitoring by offering objective and accurate damage assessment.
• The CNN-based model shows significant potential for real-world applications in infrastructure safety and maintenance.