Mode shape prediction based on convolutional neural network and autoencoder

A novel version of hierarchical genetic algorithm and its application for hyperparameters optimization in CNN models for structural delamination identification

“…Numerous approaches have been proposed in academic works to detect, quantify, and locate delamination in laminated composites, while vibration-based methods have demonstrated greater suitability [4][5][6], these methods can be classified into two classes: those based on low-frequency structural vibrations and those based on high-frequency guided waves. In the former category, modal parameters like modal damping, natural frequency, modal shapes [7], as well as modal shape curvature, wavelet coefficient, frequency response function (FRF), auto-regressive parameters, and power spectral density (PSD) of transient response [8][9][10][11] are typically employed for delamination evaluation [12,13], conversely, frequency centroid, phase change, correlation coefficient, the difference of amplitudes, and wavelet energy [14][15][16][17][18] are utilized for high-frequency guided wave-based methods. Generally, low-frequency structural vibration response captures the global behavior of a structure, whereas high-frequency vibration response focuses on its local aspects [19], however, high-frequency guided wave-based methods have practical limitations, such as requiring too many sensors, data acquisition at a higher rate, and complex signal processing [20], which can be challenging to implement.…”