Vibration-based structural health monitoring of bridges based on a new unsupervised machine learning technique under varying environmental conditions

The most significant steps in vibration-based structural health monitoring (SHM) are to extract reliable damage sensitive features from the responses of structure and to make a decision about the safety and serviceability of the structure using the extracted features. However, in most real-world applications, adverse influences caused by multiple sources of environmental variability conditions such as traffic loading, wind, and, most importantly, temperature variations can mask extracted features and may lead to false positive and/or false negative indications of structural damage. Hence, it is thus fundamentally significant to understand the relationship between extracted features and environmental variations and to investigate the effects of these variations on the damage-related features and damage detection procedure. This article proposes a new hybrid unsupervised machine learning technique for early damage detection of bridge structures, which are always exposed to environmental variability conditions. The proposed method is based on a data dependent dissimilarity measure with the focus on effectively investigating and accurately suppressing the effects of environmental variability conditions from extracted features. The main merit of this method is to enable a machine learning technique to highly reduce the variations caused by environmental factors and increase damage detectability in an unsupervised manner. At last, the effectiveness and robustness of the proposed approach are assessed and verified through the well-known Tianjin-Yonghe Bridge; additionally, the proposed unsupervised machine learning methodology succeeds in early detecting damage under variability of environmental conditions.