Scholarlink

<p>Metals and composite structures are widely used in various industries due to their high mechanical strength and durability. However, defects generated during the manufacturing and operating phase limit their future usefulness and are recommended through non-invasive inspection. Quadratic frequency-modulated thermography (QFMT) is a non-destructive testing technique applicable to many materials due to its high-energy deposition at lower frequencies for enhanced defect signatures. The recent past in QFMT is advanced with machine learning and deep learning-based techniques. In contrast, the highly class-imbalanced and scarce nature of thermal profiles has recently gained interest in anomaly detection models. A stacked denoising convolution autoencoder (SDCAE) driven Local Outlier Factor (LOF) is proposed in the present article to identify defects in mild steel and carbon fiber reinforced polymer specimens inspected by QFMT. Deep features extracted from the temporal thermal profiles using pre-trained SDCAE are further fed to LOF for automatic defect detection. A quantitative comparison with recently introduced deep anomaly detection models and other autoencoder models for performance analysis strengthens the suitability of proposed method for automatic defect detection.</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

Automatic Deep Anomaly Detection using Thermograpy based Convolution Autoencoder Framework

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Naga Prasanthi Yerneni,
  • L. Sainath,
  • V. S. Ghali,
  • Sk. Aashik,
  • G. T. Vesala,
  • V. Dhanunjana Chari,
  • Fei Wang

摘要

Metals and composite structures are widely used in various industries due to their high mechanical strength and durability. However, defects generated during the manufacturing and operating phase limit their future usefulness and are recommended through non-invasive inspection. Quadratic frequency-modulated thermography (QFMT) is a non-destructive testing technique applicable to many materials due to its high-energy deposition at lower frequencies for enhanced defect signatures. The recent past in QFMT is advanced with machine learning and deep learning-based techniques. In contrast, the highly class-imbalanced and scarce nature of thermal profiles has recently gained interest in anomaly detection models. A stacked denoising convolution autoencoder (SDCAE) driven Local Outlier Factor (LOF) is proposed in the present article to identify defects in mild steel and carbon fiber reinforced polymer specimens inspected by QFMT. Deep features extracted from the temporal thermal profiles using pre-trained SDCAE are further fed to LOF for automatic defect detection. A quantitative comparison with recently introduced deep anomaly detection models and other autoencoder models for performance analysis strengthens the suitability of proposed method for automatic defect detection.