Damage Assessment of a Cable-Stayed Bridge Based on Effective Empirical Mode Features using Empirical Wavelet Transform

Intelligent damage detection of civil infrastructures is vital to improve damage prediction performance and reduce maintenance costs. Therefore, the development of efficient techniques for detecting structural damages in an early stage is extremely important to support making decisions on structure repair. In this paper, a new damage detection method based on effective frequency band with empirical wavelet transform for a cable-stayed bridge was proposed which consists of two stages: (1) signal processing and feature extraction, (2) damage identification by combining effective features. In the first stage, structural response data was decomposed into empirical modes using empirical wavelet transform to obtain the component related to structural damage, and a set of features as damage-sensitive features were extracted from the frequency spectrum of modes. A support vector machine was applied to evaluate the results. In the second stage, by applying feature selection methods, an optimal subset of features that carries the most significant information about the structural damage was obtained as a damage index. Next, it was used in the feature extraction process. To verify the proposed damage detection method, response data obtained from a cable-stayed bridge, the Yonghe Bridge, was employed. Results showed that the second and third empirical modes obtained from the empirical wavelet analysis contain fault information and using its corresponding frequency spectrum in the feature extraction process improves detection performance by about 5% compared to conventional methods. It also increases the detection accuracy to about 94% by employing effective feature combinations rather than a single feature.