基于EMD-Transformer-GMM的桥梁结构异常识别方法OA

Acceleration data collected by bridge structural health monitoring systems provide essential information for structural anomaly detection and early damage identification.However,raw acceleration signals are complex and contain substantial noise interference,which significantly degrades the separability of anomalous features and increases the risk of misjudgment.To address this issue,this paper proposes a self-supervised anomaly detection framework that integrates multi-frequency decomposition with multi-scale temporal modeling,along with a novel anomaly-sensitive metric:the Gaussian kernel-weighted residual sum of squares(GKWRSS).The proposed method first applies empirical mode decomposition(EMD)to extract frequency-sensitive components,then utilizes a Trans-former to capture long-range temporal dependencies and reconstruct the input signals.Finally,a Gaussian mixture model(GMM)is used to cluster the GKWRSS values of the original and reconstructed signals,thereby achieving anomaly detection for bridge structures.Experimental results demonstrate that the proposed EMD-Transformer-GMM architecture effectively suppresses cross-frequency interference and enhances the separability of anomalous features.Moreover,owing to the transposed convolutional decoder incorporated into the Transformer layer,the architecture enables self-supervised learning,preserving fine-grained temporal features and frequency-domain specificity during signal reconstruction.Furthermore,the method requires only classification based on signal differences,without the need for labeled anomaly samples.Compared with the conventional residual sum of squares(RSS),GKWRSS significantly improves the separability between normal and anomalous states while maintaining sensitivity to local anomalies.Validation using real bridge data and numerical simulation tests confirms the reliability and practicality of the proposed method for bridge structural anomaly detection.