Pixel-Level Fatigue Crack Segmentation in Large-Scale Images of Steel Structures Using an Encoder-Decoder Network

Area of Science:

• Civil Engineering
• Computer Vision
• Materials Science

Background:

• Fatigue cracks in steel structures pose significant risks, potentially leading to catastrophic failure.
• Current inspection methods like visual inspection are subjective and labor-intensive, while non-destructive testing is expensive.
• Accurate fatigue crack detection is crucial for structural health monitoring, maintenance, and extending service life.

Purpose of the Study:

• To develop an automated, pixel-level fatigue crack segmentation framework for large-scale steel structure images.
• To address the limitations of existing manual and non-destructive testing methods for crack detection.
• To improve the efficiency and accuracy of fatigue crack assessment in civil infrastructure.

Main Methods:

• A modified U-net encoder-decoder network was employed for pixel-level fatigue crack segmentation.
• Large-resolution images were cropped into smaller segments for efficient training and testing.
• Image post-processing techniques, including opening and closing operations, were applied to refine segmentation maps.

Main Results:

• The proposed U-net based method achieved acceptable accuracy in automatic fatigue crack segmentation, measured by average intersection over union (mIOU).
• Comparative analysis showed superior performance of the U-net model over a Fully Convolutional Network (FCN) with ResNet34 backbone.
• The U-net model demonstrated better segmentation results with fewer training epochs and a simpler architecture.

Conclusions:

• The developed framework offers an effective and efficient solution for fatigue crack segmentation in large-scale steel structures.
• The study provides valuable insights and recommendations for applying image-based fatigue crack detection in civil infrastructure engineering.
• This automated approach supports better condition assessment, maintenance planning, and lifecycle management of existing structures.