Structural damage detection based on variational mode decomposition and kernel PCA-based support vector machine

This paper proposes a novel structural damage detection method by combining the advantages of variational mode decomposition algorithm and kernel principal component analysis in the presence of environmental effects. At first, decomposition of vibration response data is carried out using the variational mode decomposition algorithm to obtain intrinsic mode functions (IMFs) and features in the time–frequency domain. Then, the spectral centroid feature corresponding to the statistical characteristics of the spectrum shape in the short-time Fourier transform is extracted from selected IMF components to construct the damage feature matrix. Several IMF components containing damage information are considered for the feature extraction process to remove the redundant (or irrelevant) features. Finally, kernel principal component analysis is performed on the constructed feature matrix to overcome the operational and environmental changes and obtain damage-sensitive indices, T2 and SPE control charts, for condition monitoring. Support vector machine (SVM) is used as a decision-making tool to illustrate the effectiveness of this method compared to the existing approaches for extracting damage features. Furthermore, numerical and full-scale studies are performed in order to validate the robustness of the proposed method. The results show that the stated method can accurately identify structural damage under varying environmental conditions.