Scholarlink

<p>Performance loss occurs in civil engineering infrastructure (including bridges, buildings, and geo-construction), which will eventually require sustained maintenance through the use of Structural Health Monitoring (SHM) to prevent a catastrophic failure of the structure. While SHM has become increasingly popular, visual inspections and the interpretation of manually collected sensor data have proven to be inadequate for complex rheology. In addition to testing using fiber bragg grating (FBG) accelerometer systems to determine that a resonance frequency of 708&#xa0;Hz exists, a large amount of quality data has been collected for use with supervised learning techniques. Using acoustic emission (AE), unsupervised cluster analysis was performed to identify patterns related to crack propagation in the specimen. In the case of surface acoustic wave (SAW) sensors, a number of regression-based models have been developed for the purpose of predicting the remaining useful service life of concrete structures based on measurements of corrosion on rebar.Results from experimental tests using FBG accelerometers demonstrate a resonance frequency of 708&#xa0;Hz with coefficients of regression greater than 0.99, thus demonstrating the potential of the test data to provide highly accurate input for supervised learning classifications. Results from this study also demonstrated the ability to process AE signals using unsupervised cluster analysis to isolate crack growth at high energy release rates, which is difficult to accomplish using manual collection methods. The AE feature extraction was accomplished by determining the peak amplitude and signal duration for AE. In addition, the FFT method was employed for vibration feature extraction in order to measure the shift in the 708&#xa0;Hz peak due to changes in vibration levels. Using this quantifiable amount of rebar corrosion that is determined through surface acoustic wave (SAW) sensors as the input variable to regression-based machine learning models allowed the researchers to estimate the remaining service life of a structure composed of concrete. Therefore, the multi-modal assessment of the structure was achieved through the integration of the SAW corrosion data with the AE crack growth data. The data fusion process also provides a reduction in error resulting from environmental noise since a structural anomaly can be confirmed by at least one additional sensor type thereby increasing the accuracy of detection. In conclusion, the authors suggest that the combination of smart sensors with mechanism knowledge algorithms will enable a shift from monitoring to analytical maintenance for civil infrastructure.</p>

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

Machine learning-based feature extraction from smart sensors for quantifying rebar corrosion and structural integrity

  • 期刊论文

  • 发表日期：

    2026年4月6日

查看全文
  • Parthiban V,
  • Madhumathi R.,
  • Pranesh C.,
  • Om Shri M.

摘要

Performance loss occurs in civil engineering infrastructure (including bridges, buildings, and geo-construction), which will eventually require sustained maintenance through the use of Structural Health Monitoring (SHM) to prevent a catastrophic failure of the structure. While SHM has become increasingly popular, visual inspections and the interpretation of manually collected sensor data have proven to be inadequate for complex rheology. In addition to testing using fiber bragg grating (FBG) accelerometer systems to determine that a resonance frequency of 708 Hz exists, a large amount of quality data has been collected for use with supervised learning techniques. Using acoustic emission (AE), unsupervised cluster analysis was performed to identify patterns related to crack propagation in the specimen. In the case of surface acoustic wave (SAW) sensors, a number of regression-based models have been developed for the purpose of predicting the remaining useful service life of concrete structures based on measurements of corrosion on rebar.Results from experimental tests using FBG accelerometers demonstrate a resonance frequency of 708 Hz with coefficients of regression greater than 0.99, thus demonstrating the potential of the test data to provide highly accurate input for supervised learning classifications. Results from this study also demonstrated the ability to process AE signals using unsupervised cluster analysis to isolate crack growth at high energy release rates, which is difficult to accomplish using manual collection methods. The AE feature extraction was accomplished by determining the peak amplitude and signal duration for AE. In addition, the FFT method was employed for vibration feature extraction in order to measure the shift in the 708 Hz peak due to changes in vibration levels. Using this quantifiable amount of rebar corrosion that is determined through surface acoustic wave (SAW) sensors as the input variable to regression-based machine learning models allowed the researchers to estimate the remaining service life of a structure composed of concrete. Therefore, the multi-modal assessment of the structure was achieved through the integration of the SAW corrosion data with the AE crack growth data. The data fusion process also provides a reduction in error resulting from environmental noise since a structural anomaly can be confirmed by at least one additional sensor type thereby increasing the accuracy of detection. In conclusion, the authors suggest that the combination of smart sensors with mechanism knowledge algorithms will enable a shift from monitoring to analytical maintenance for civil infrastructure.