Deep learning enhanced quantitative debonding evaluation in tile panels using Lamb waves.

Area of Science:

• Structural Health Monitoring
• Non-Destructive Evaluation
• Materials Science

Background:

• Exterior ceramic tile panels are susceptible to debonding, compromising structural integrity.
• Quantitative evaluation of debonding is crucial for timely maintenance and safety.
• Existing non-destructive evaluation (NDE) methods may have limitations in precision and scope.

Purpose of the Study:

• To develop a novel, data-driven methodology for quantitatively evaluating debonding in exterior ceramic tile panels.
• To leverage Lamb wave signals and deep learning for precise defect characterization.
• To create a comprehensive 2D damage index map for identifying debonding position and size.

Main Methods:

• Utilized Lamb wave signals and transformed them into time-frequency images using continuous wavelet transform (CWT).
• Employed a two-dimensional convolutional neural network (2D-CNN) to process time-frequency representations and extract debonding information.
• Combined simulation data with experimental validation to develop and test the damage detection model.

Main Results:

• The 2D-CNN effectively predicted debonding characteristics in ceramic tiles.
• A comprehensive 2D damage index map was generated, accurately identifying the location and size of debonding defects.
• The model demonstrated superior performance when trained with a hybrid dataset incorporating experimental data.

Conclusions:

• The proposed deep learning methodology offers a reliable approach for non-destructive evaluation of debonding in tiled structures.
• The integration of Lamb waves and 2D-CNN provides a powerful tool for quantitative damage assessment.
• This technique holds significant potential for enhancing the safety and longevity of building facades.