Scholarlink

<p>Structural Health Monitoring (SHM) methods based on vibration data have garnered significant attention from researchers due to their non-destructive nature. Identifying damage sensitive features, which are crucial for detecting damage, remains a key focus in SHM research. Resonant frequencies and mode shape vectors have been widely utilized as effective damage sensitive features as they capture the dynamic properties of a structure. These features serve as inputs for machine learning algorithms, which analyze them to predict structural damage. This study investigates the sensitivity and influence of input features derived from basic vibration data on the accuracy of damage prediction in cantilever beams. A validated Finite Element Model of a cantilever beam under free vibration is considered in the present study. Features such as modal frequencies, fundamental mode shapes and statistically modelled responses extracted from the free vibration data of 50 finite element models, given as input data for training and testing a supervised machine learning Support Vector Machines algorithm. Results indicate that the novel approach of statistically modelled modal displacements exhibit the highest sensitivity revealing that as the most damage sensitive feature offering a refined perspective on feature selection for predictive modelling in Structural Health Monitoring.</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

Impact of feature selection on predictive damage identification in beams using free vibration data-based machine learning algorithm

  • 期刊论文

  • 发表日期：

    2025年11月26日

查看全文
  • B. Rajashekharam,
  • Mallika Alapati

摘要

Structural Health Monitoring (SHM) methods based on vibration data have garnered significant attention from researchers due to their non-destructive nature. Identifying damage sensitive features, which are crucial for detecting damage, remains a key focus in SHM research. Resonant frequencies and mode shape vectors have been widely utilized as effective damage sensitive features as they capture the dynamic properties of a structure. These features serve as inputs for machine learning algorithms, which analyze them to predict structural damage. This study investigates the sensitivity and influence of input features derived from basic vibration data on the accuracy of damage prediction in cantilever beams. A validated Finite Element Model of a cantilever beam under free vibration is considered in the present study. Features such as modal frequencies, fundamental mode shapes and statistically modelled responses extracted from the free vibration data of 50 finite element models, given as input data for training and testing a supervised machine learning Support Vector Machines algorithm. Results indicate that the novel approach of statistically modelled modal displacements exhibit the highest sensitivity revealing that as the most damage sensitive feature offering a refined perspective on feature selection for predictive modelling in Structural Health Monitoring.