Scholarlink

Civil structures are valuable assets of every society with significant influences on human life, economics, transportation networks, and energy supply. These assets are always susceptible to natural and man-induced hazards, aging, and material deterioration. The adverse consequences of these events are the occurrence and growth of different patterns of structural damage, failure, and even collapse. Vibration-based structural health monitoring (SHM) supported by sensory data and artificial intelligence is an emerging and innovative technology for assessing the safety and functionality of various civil structures, especially bridges. Although this technology can be implemented in short- and long-term programs, some limitations do not allow civil engineers to benefit from the Dunn long-term SHM, in which case an automated short-term monitoring program is an effective and practical alternative. However, some challenges make this program problematic. The most significant challenge relates to encountering limited vibration data during a short period of monitoring. The other important issue is profound effects of operational and environmental variability on vibration data such as modal frequencies, which cause false alarm and mis-detection errors. To address these challenges, this chapter proposes an innovative SHM approach suitable for short-term monitoring programs based on the concept of integrated unsupervised learning. Using limited modal frequencies as the main vibration features, this approach comprises three steps of local feature segmentation, unsupervised feature selection, and unsupervised anomaly detection. First, agglomerative hierarchical clustering in conjunction with the Dunn’s index is employed to segment the limited set of modal frequencies into local subsets (clusters). Second, a filter-based unsupervised feature selector is proposed to find the most appropriate cluster with the features insensitive to environmental/operational conditions. Third, these features are used to develop an anomaly detector based on the local outlier factor and determine anomaly indices for damage assessment. A concrete cable-stayed bridge is considered to testify the proposed approach. It is observed that this approach succeeds in alarming damage with limited data under a short-term monitoring program.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Short-Term Health Monitoring and Damage Assessment of a Concrete Cable-Stayed Bridge by an Integrated Unsupervised Learning Approach

  • 章节

  • 发表日期：

    2025年

查看全文
  • Hassan Sarmadi,
  • Alireza Entezami

摘要

Civil structures are valuable assets of every society with significant influences on human life, economics, transportation networks, and energy supply. These assets are always susceptible to natural and man-induced hazards, aging, and material deterioration. The adverse consequences of these events are the occurrence and growth of different patterns of structural damage, failure, and even collapse. Vibration-based structural health monitoring (SHM) supported by sensory data and artificial intelligence is an emerging and innovative technology for assessing the safety and functionality of various civil structures, especially bridges. Although this technology can be implemented in short- and long-term programs, some limitations do not allow civil engineers to benefit from the Dunn long-term SHM, in which case an automated short-term monitoring program is an effective and practical alternative. However, some challenges make this program problematic. The most significant challenge relates to encountering limited vibration data during a short period of monitoring. The other important issue is profound effects of operational and environmental variability on vibration data such as modal frequencies, which cause false alarm and mis-detection errors. To address these challenges, this chapter proposes an innovative SHM approach suitable for short-term monitoring programs based on the concept of integrated unsupervised learning. Using limited modal frequencies as the main vibration features, this approach comprises three steps of local feature segmentation, unsupervised feature selection, and unsupervised anomaly detection. First, agglomerative hierarchical clustering in conjunction with the Dunn’s index is employed to segment the limited set of modal frequencies into local subsets (clusters). Second, a filter-based unsupervised feature selector is proposed to find the most appropriate cluster with the features insensitive to environmental/operational conditions. Third, these features are used to develop an anomaly detector based on the local outlier factor and determine anomaly indices for damage assessment. A concrete cable-stayed bridge is considered to testify the proposed approach. It is observed that this approach succeeds in alarming damage with limited data under a short-term monitoring program.