Structural damage identification method based on threshold-free cross-recurrence plot and feature dimension reduction

Abstract

As an important structural form in the fields of architecture and mechanics, the accurate identification of damage in frame structures is crucial for ensuring operational safety. The recurrence plot (RP) of structural responses can finely characterize the recurrence characteristics of structural damage features in the time-frequency domain and serve as an effective damage identification indicator. However, previous research typically relies on supervised learning for feature extraction. To overcome this limitation and further enhance its ability to extract damage information while suppressing redundant noise, an unsupervised damage identification method based on a recurrence principal component (PC) energy indicator is proposed in this article. This method innovatively combines threshold-free cross-recurrence analysis with PC dimensionality reduction techniques. While enriching and expanding the damage information contained in the signal, it effectively eliminates redundant information. Through comparative analysis of multi-degree-of-freedom structural systems, it is shown that, compared with traditional indicators such as wavelet packet energy indicators and recurrence rate change ratio, more precise damage localization is achieved. Compared to the classical recurrence quantification analysis method, the damage identification reliability of the RP is significantly enhanced under noise interference after PC dimensionality reduction. Further numerical simulations and damage experiments have verified the wide applicability of the proposed method to various damages in frame structures. Combined with a mechanical mechanism analysis, the comprehensive advantages of this method in terms of noise resistance, robustness, and engineering applicability are systematically demonstrated.