Scholarlink

<p>Quadratic Frequency-Modulated Thermographic Wave Imaging (QFMTWI) is a powerful nondestructive testing (NDT) technique based on thermographic principles that has been proven to detect defects in various materials. However, its nonlinear thermal stimulus generates high-dimensional responses, challenging defect detection. Traditional dimensionality reduction methods, such as Principal Component Analysis (PCA), rely on linear transformations, which limit their ability to capture complex thermal patterns. While linear autoencoders offer an alternative, they often suffer from uncorrelated encoded features, non-orthogonal encoder-decoder weights, and a lack of unit norm constraints, leading to suboptimal defect representation. This work introduces a constrained and regularised shallow autoencoder that enforces correlated encoding, orthogonality, and unit norm constraints, enhancing feature extraction. Carbon Fibre Reinforced Polymer (CFRP) specimens with delamination defects are simulated to validate the proposed approach, demonstrating a significant improvement in the defect signal-to-noise ratio (SNR) and superior performance compared to traditional methods.</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

Autoencoder with Constraints for Subsurface Defect Detection in Quadratic Frequency Modulated Thermal Wave Imaging

  • 期刊论文

  • 发表日期：

    2025年8月18日

查看全文
  • Enosh Bandari,
  • Siva Subrahmanyam Oleti,
  • Y. Naga Prasanthi,
  • G. T. Vesala,
  • V. S. Ghali

摘要

Quadratic Frequency-Modulated Thermographic Wave Imaging (QFMTWI) is a powerful nondestructive testing (NDT) technique based on thermographic principles that has been proven to detect defects in various materials. However, its nonlinear thermal stimulus generates high-dimensional responses, challenging defect detection. Traditional dimensionality reduction methods, such as Principal Component Analysis (PCA), rely on linear transformations, which limit their ability to capture complex thermal patterns. While linear autoencoders offer an alternative, they often suffer from uncorrelated encoded features, non-orthogonal encoder-decoder weights, and a lack of unit norm constraints, leading to suboptimal defect representation. This work introduces a constrained and regularised shallow autoencoder that enforces correlated encoding, orthogonality, and unit norm constraints, enhancing feature extraction. Carbon Fibre Reinforced Polymer (CFRP) specimens with delamination defects are simulated to validate the proposed approach, demonstrating a significant improvement in the defect signal-to-noise ratio (SNR) and superior performance compared to traditional methods.