Feature Extraction of UGW Defect Signals Using SVD-PCA-ICA for Rail Structural Health Monitoring

The propagation characteristics of ultrasonic guided waves (UGWs) in rails, together with the inherently complex inspection environment for in-service rails, pose a significant challenge for the effective detection of rail fractures or damage using UGW. To address these issues, this work proposes a feature extraction method for defect-related UGW signals in the structural health monitoring of rails. The proposed approach integrates singular value decomposition (SVD), principal component analysis (PCA), and independent component analysis (ICA). First, to eliminate periodic interference, the constructed data matrix is processed using a Hankel matrix combined with SVD, and the analysis matrix is then reconstructed for subsequent ICA processing. Second, to reduce computational complexity, the dimensionality of the analysis matrix is reduced using PCA. Finally, the independent components and their corresponding weight vectors are obtained by applying ICA to the analysis matrix. By examining whether the resulting weight vectors exhibit significant step characteristics, it is possible to determine whether the rail is defective. Experimental validation demonstrates that the proposed feature extraction method is both reliable and effective.