Scholarlink

<p>Structural health monitoring of reinforced concrete (RC) structures is critical, as their durability and safety are often compromised due to cracks caused by mechanical loads and environmental exposure. Traditional crack inspection techniques are labor-intensive, subjective, and inadequate for large-scale applications. This study addresses these limitations by leveraging deep learning, specifically Convolutional Neural Networks (CNNs), for automated crack classification and segmentation in RC beams. An experimental setup was designed involving the casting and flexural testing of RC beams with varying reinforcement configurations to simulate real-world cracking scenarios. From this experimental work, a high-resolution image dataset was created, comprising categorized images of undamaged, cracked, and severely damaged beam surfaces. Four pre-trained CNN architectures AlexNet, VGG-16, ResNet-101, and GoogleNet were fine-tuned using MATLAB’s Deep Learning Toolbox and evaluated for their performance in crack classification and segmentation. Advanced image preprocessing and segmentation techniques were employed to enhance crack visibility and model accuracy. Evaluation was conducted using a confusion matrix including precision, recall, and F1-score were recorded. Among all models, VGG-16 exhibited the highest overall performance, making it the most effective for accurate crack detection. These findings underscore the potential of CNN-based methods for real-time, scalable, and reliable structural health monitoring, paving the way for intelligent infrastructure maintenance solutions.</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

Crack classification and segmentation in RC beams using convolutional neural networks and transfer learning

  • 期刊论文

  • 发表日期：

    2025年8月12日

查看全文
  • Neel S. Metha,
  • Priti R. Satarkar

摘要

Structural health monitoring of reinforced concrete (RC) structures is critical, as their durability and safety are often compromised due to cracks caused by mechanical loads and environmental exposure. Traditional crack inspection techniques are labor-intensive, subjective, and inadequate for large-scale applications. This study addresses these limitations by leveraging deep learning, specifically Convolutional Neural Networks (CNNs), for automated crack classification and segmentation in RC beams. An experimental setup was designed involving the casting and flexural testing of RC beams with varying reinforcement configurations to simulate real-world cracking scenarios. From this experimental work, a high-resolution image dataset was created, comprising categorized images of undamaged, cracked, and severely damaged beam surfaces. Four pre-trained CNN architectures AlexNet, VGG-16, ResNet-101, and GoogleNet were fine-tuned using MATLAB’s Deep Learning Toolbox and evaluated for their performance in crack classification and segmentation. Advanced image preprocessing and segmentation techniques were employed to enhance crack visibility and model accuracy. Evaluation was conducted using a confusion matrix including precision, recall, and F1-score were recorded. Among all models, VGG-16 exhibited the highest overall performance, making it the most effective for accurate crack detection. These findings underscore the potential of CNN-based methods for real-time, scalable, and reliable structural health monitoring, paving the way for intelligent infrastructure maintenance solutions.