An automated vibration-based structural damage localization strategy using filter-type feature selection. (1st May 2023)

• Record Type: Journal Article
• Authors: Alves, Victor; Cury, Alexandre
• Is Part Of: Mechanical systems and signal processing. Volume 190(2023)
• Publication Date: 2023-05-01
• Subjects: Damage localization -- Damage quantification -- Feature selection -- Environmental conditions -- Traffic variation -- Automated; Structural dynamics -- Periodicals; Vibration -- Periodicals; Constructions -- Dynamique -- Périodiques; Vibration -- Périodiques; Structural dynamics; Vibration; Periodicals; 621
• DOI: 10.1016/j.ymssp.2023.110145
• Languages: English
• ISSNs: 0888-3270

Abstract

Abstract: Damage detection techniques have been widely explored over the last years driven by the advances of computational intelligence technologies. To understand the structure's dynamic behavior, some studies use modal-based methods, which are traditionally applied to reflect changes caused by damage. More recently, strategies using raw vibration measurements have gained notoriety and are proving to be a promising approach. In this context, the aim of this paper is to present a novel automated methodology for damage localization based on the extraction of features from dynamic data using domain knowledge and on an unsupervised filtering process. Feature extraction is performed simultaneously in time, frequency, and quefrency domains to diversify information retrieval. In machine learning (ML), this filtering procedure is called feature selection (FS) and is applied herein with the goal to decrease redundancy and increase the relevance of the feature set by eliminating a portion of it in view of a predefined criterion. The main concept is that the proposed method can customize its features concerning the structure, providing generality about any type of geometry, material, and excitation it encounters. The damage-sensitive index is computed through an outlier analysis, where percentile intervals are created based on the filtered features from the structure's healthy state. Thus, any anomalies surrounding this initial state should be automatically located. The method was successfully tested in five applications, two of them being full-scale bridges, which shows a promising performance for real-world monitoring situations. Some of its significant achievements are single and multiple damage location, structural reinforcement detection, quantification of severity in progressive damage scenarios, and robustness to noise due to varying traffic and/or environmental conditions. Such a reliability is tied to an efficient default parameters setup, enabling an automated damage localization algorithm.