Scholarlink

Purpose <p>Kevlar fiber/epoxy composites are widely used in aerospace, automotive, and defense industries due to their outstanding mechanical properties. However, these materials are vulnerable to damage such as cracks, which can significantly reduce structural integrity. This study proposes a proactive approach to structural health assessment of Kevlar components by integrating experimental modal analysis (EMA), numerical modal analysis (NMA) and feedforward neural networks (FNNs).</p> Methods <p>The research methodology involves three key phases: First, EMA is conducted on damaged Kevlar beam and plate specimens under both fixed-free and fixed-fixed boundary conditions. The experimental results validate corresponding NMA simulations performed in ANSYS Composite PrepPost, demonstrating excellent agreement with average frequency errors below 5%. Second, the validated numerical models generate an extensive dataset of the first three vibration modes for various crack configurations, with crack depths ranging from 20% to 50% of the width of test specimens and locations distributed across multiple positions. Third, this dataset trains FNN models to predict crack location and depth from vibrational characteristics.</p> Results <p>The developed FNN architecture achieves outstanding predictive performance, evidenced by an R² score of 0.9312 and average localization error below 5%. The model demonstrates particular effectiveness for intermediate-depth cracks of 35% thickness, achieving an error of just 1.08%, while maintaining robust performance across all tested damage scenarios. Comparative analysis reveals the FNN approach reduces error rates by 60–75% compared to conventional vibration-based methods. The proposed FNN model demonstrates high practical reliability for crack parameter estimation, achieving an a20-index of 0.94. These results confirm the model robustness and strong predictive accuracy, supporting its suitability for real-world structural health monitoring applications.</p> Conclusion <p>Critical issues in composite structural health monitoring are addressed by this combined EMA–NMA–FNN framework, such as real-time evaluation capabilities, accurate damage identification under intricate boundary conditions, and interpretation of anisotropic vibration responses. The work shows that vibration analysis improved by machine learning offers a dependable, non-destructive method for inspecting Kevlar composites. The methodology success points to potential uses in safety-critical predictive maintenance systems.</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-Enhanced Vibration Analysis for Damage Detection and Localization in Kevlar Composites

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • S. Rama Krishna,
  • J. Sathish,
  • M. Tarun,
  • S. Raghu Vamsi,
  • S. Janu Sree

摘要

Purpose

Kevlar fiber/epoxy composites are widely used in aerospace, automotive, and defense industries due to their outstanding mechanical properties. However, these materials are vulnerable to damage such as cracks, which can significantly reduce structural integrity. This study proposes a proactive approach to structural health assessment of Kevlar components by integrating experimental modal analysis (EMA), numerical modal analysis (NMA) and feedforward neural networks (FNNs).

Methods

The research methodology involves three key phases: First, EMA is conducted on damaged Kevlar beam and plate specimens under both fixed-free and fixed-fixed boundary conditions. The experimental results validate corresponding NMA simulations performed in ANSYS Composite PrepPost, demonstrating excellent agreement with average frequency errors below 5%. Second, the validated numerical models generate an extensive dataset of the first three vibration modes for various crack configurations, with crack depths ranging from 20% to 50% of the width of test specimens and locations distributed across multiple positions. Third, this dataset trains FNN models to predict crack location and depth from vibrational characteristics.

Results

The developed FNN architecture achieves outstanding predictive performance, evidenced by an R² score of 0.9312 and average localization error below 5%. The model demonstrates particular effectiveness for intermediate-depth cracks of 35% thickness, achieving an error of just 1.08%, while maintaining robust performance across all tested damage scenarios. Comparative analysis reveals the FNN approach reduces error rates by 60–75% compared to conventional vibration-based methods. The proposed FNN model demonstrates high practical reliability for crack parameter estimation, achieving an a20-index of 0.94. These results confirm the model robustness and strong predictive accuracy, supporting its suitability for real-world structural health monitoring applications.

Conclusion

Critical issues in composite structural health monitoring are addressed by this combined EMA–NMA–FNN framework, such as real-time evaluation capabilities, accurate damage identification under intricate boundary conditions, and interpretation of anisotropic vibration responses. The work shows that vibration analysis improved by machine learning offers a dependable, non-destructive method for inspecting Kevlar composites. The methodology success points to potential uses in safety-critical predictive maintenance systems.