Scholarlink

Abstract <p>The quality of girth welds in long-distance oil and gas pipelines is critical for the safe operation of pipeline systems. This study investigates time-of-flight diffraction (TOFD) ultrasonic inspection of X70-grade pipeline steel butt welds containing typical defects such as porosity, lack of fusion, cracks, slag inclusions, and lack of penetration. A multifrequency dual-probe array scanning strategy was developed to increase blind-zone coverage to 95%. A multimodal joint filtering algorithm was applied to suppress impulse noise while preserving high-frequency details, resulting in a peak signal-to-noise ratio of 38.6 dB and improved defect contrast. For defect classification, an AlexNet-SVM hybrid model was constructed by replacing the Softmax classifier with an RBF-kernel SVM and optimizing hyperparameters. The proposed method achieved a classification accuracy of 98.67%, which is 9.97% higher than that of the baseline AlexNet model and meets the requirements of ASME B31.8S-2021. The results demonstrate that the combined scanning strategy, image preprocessing, and optimized classifier can significantly improve the reliability and automation of TOFD-based weld 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

TOFD-Based Ultrasonic Inspection of Pipeline Welds Using Multifrequency Array Scanning and AlexNet-SVM Classification

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Ting Zou,
  • Changjun Liu,
  • Guangyu Chen,
  • Chen Ma,
  • Dongao Huo

摘要

Abstract

The quality of girth welds in long-distance oil and gas pipelines is critical for the safe operation of pipeline systems. This study investigates time-of-flight diffraction (TOFD) ultrasonic inspection of X70-grade pipeline steel butt welds containing typical defects such as porosity, lack of fusion, cracks, slag inclusions, and lack of penetration. A multifrequency dual-probe array scanning strategy was developed to increase blind-zone coverage to 95%. A multimodal joint filtering algorithm was applied to suppress impulse noise while preserving high-frequency details, resulting in a peak signal-to-noise ratio of 38.6 dB and improved defect contrast. For defect classification, an AlexNet-SVM hybrid model was constructed by replacing the Softmax classifier with an RBF-kernel SVM and optimizing hyperparameters. The proposed method achieved a classification accuracy of 98.67%, which is 9.97% higher than that of the baseline AlexNet model and meets the requirements of ASME B31.8S-2021. The results demonstrate that the combined scanning strategy, image preprocessing, and optimized classifier can significantly improve the reliability and automation of TOFD-based weld defect detection.