Fatigue Crack Evaluation with the Guided Wave-Convolutional Neural Network Ensemble and Differential Wavelet...

Area of Science:

• Structural Health Monitoring (SHM)
• Non-destructive Testing (NDT)
• Signal Processing

Background:

• Online fatigue crack evaluation is vital for structural safety and cost-effective maintenance of critical systems.
• Guided wave (GW)-based SHM is a promising technique, but traditional methods rely on expert-defined features susceptible to uncertainties.
• Existing machine learning approaches also require manual feature engineering, limiting their robustness.

Purpose of the Study:

• To develop an automated and robust framework for online fatigue crack evaluation using GW-based SHM.
• To overcome the limitations of manual feature extraction in traditional and machine learning methods for GW analysis.
• To accurately determine fatigue crack lengths in complex structures.

Main Methods:

• A novel framework combining a convolutional neural network (CNN) ensemble with differential wavelet spectrograms is proposed.
• Complex Gaussian wavelet transform is used to generate differential time-frequency spectrograms from baseline and monitoring GW signals.
• An ensemble of CNNs is trained to directly process these spectrograms for crack length estimation.

Main Results:

• The proposed method achieved a root mean square error (RMSE) of 1.4 mm on the training dataset.
• Validation on complex lap joint structures yielded an RMSE of 1.7 mm for evaluated crack lengths.
• The framework demonstrated effectiveness in real-world fatigue tests.

Conclusions:

• The GW-CNN ensemble with differential wavelet spectrograms offers an effective and automated solution for online fatigue crack evaluation.
• This approach reduces reliance on expert knowledge and enhances robustness against uncertainties in SHM.
• The method shows significant potential for improving structural safety and reducing maintenance costs in safety-critical applications.