Data anomaly segmentation for structural health monitoring using denoising diffusion probabilistic models

Structural health monitoring systems are often challenged by various patterns of data anomalies, making the identification of these anomalies a crucial preprocessing step before employing monitoring data for assessing structural conditions or making informed decisions. In this paper, an advanced data anomaly segmentation approach based on denoising diffusion probabilistic models is introduced, which represents a significant enhancement over existing anomaly detection techniques. This method employs an adapted framework of denoising diffusion probabilistic models and meticulously designs the architecture of the model. Through testing on a dataset derived from a benchmark of a long span bridge, the performance of the proposed approach is validated to accurately determine the precise location and pattern of anomalous data points with an accuracy of 98.98 %. Comparative analysis with existing research further demonstrates the superior performance of the proposed approach, underscoring its potential to conduct the detailed anomaly detection and localization in the field of structural health monitoring.