Robust damage localization in plate-type structures by using an enhanced robust principal component analysis and data fusion technique

Damage localization in plate-type structures via full-field vibration measurements has attracted much more attention. Traditionally, the damage-induced local shape singularities at a certain mode are harnessed for damage localization, but this is not reliable and robust for multi-damage localization. Therefore, a general strategy is that the damage features in different modes should be accurately extracted and integrated for a robust damage localization. However, the damage features are naturally contaminated by the measurement noise and the baseline-data on pristine state is commonly unavailable, which degrade the accuracy of damage feature extraction. Furthermore, the damage features in different modes normally contain conflicting damage location evidence, which leads to misleading damage localization results. To address these issues, an enhanced robust principal component analysis (RPCA) with contiguous outlier constraint is proposed to accurately extract the damage-caused local features without requiring the baselinedata of healthy state. Moreover, a novel data fusion approach based on cosine similarity measure is developed to effectively integrate the damage features of different modes for robust damage localization. In addition, a multiscale denoising approach is proposed to evaluate the noise-robust full-field vibration measurements for damage localization. Finally, numerical and experimental studies of cantilever plates with two damage zones are studied to verify the feasibility and effectiveness of the proposed damage localization method. It is found that the proposed damage localization method is robust in two aspects: damage feature extraction from noisy measurements and detecting all the possible damage zones.

Automated summary

Damage localization in plate-type structures using full-field vibration measurements has gained increased attention. A general strategy involves accurately extracting and integrating damage features from different modes for robust localization. However, measurement noise and lack of baseline data may degrade the accuracy of feature extraction, leading to conflicting evidence for damage localization. An enhanced robust principal component analysis and data fusion approach are proposed to address these challenges, resulting in a robust method for detecting multiple damage zones.