Online monitoring of fatigue crack under variable amplitude loading driven by multispectral features fusion of FBG sensors

This paper proposed an online fatigue crack propagation quantitative monitoring method based on fiber Bragg grating (FBG) sensors by introducing the multispectral features fusion and convolutional neural network (CNN) approach. Fatigue crack propagation under variable amplitude loading is simulated by ABAQUS- FRANC3D co-simulation to obtain FBG sensed strain information. FBG distortion spectra were reconstructed using the transfer matrix method (TMM), and the relationship between five damage indicators and fatigue crack, such as center wavelength, spectral area, full width at quarter maximum (FWQM), fractal dimension, and overlapping area, was investigated. Based on this, the FBG spectrum damage feature matrix was constructed to perform multispectral feature fusion. A quantitative monitoring model between damage features and crack length was developed using CNN method to realize real-time monitoring of fatigue crack length, and crack length simulation analysis and fatigue crack propagation test were carried out on aluminum alloy. The results show that the monitoring average absolute error of fatigue crack length is 0.067 mm by proposed method in simulation analysis, and the monitoring average absolute error of fatigue crack length is 0.52 mm in fatigue crack propagation experiment, which verifies the accuracy and effectiveness of this method.