Computer Vision Method for Automatic Detection of Microstructure Defects of Concrete.

Area of Science:

• Materials Science
• Civil Engineering
• Computer Science

Background:

• Human visual inspection of concrete structures is prone to errors in identifying defects such as voids and pores.
• Computer vision, particularly convolutional neural networks (CNNs), offers a reliable automated solution for defect detection in building structures.
• Accurate assessment of concrete integrity is crucial for structural safety and material quality control.

Purpose of the Study:

• To develop and compare computer vision algorithms using CNNs for identifying and analyzing damaged concrete sections.
• To evaluate the performance of different CNN architectures (U-Net, LinkNet, PSPNet) in concrete defect segmentation.
• To assess the effectiveness of automated defect detection for quality control and material formulation adjustments.

Main Methods:

• Utilized U-Net, LinkNet, and PSPNet convolutional neural network architectures for image segmentation of concrete defects.
• Analyzed laboratory images of concrete samples to evaluate structural integrity and compactness.
• Monitored quality metrics including precision, recall, F1-score, IoU (Jaccard index), and accuracy during algorithm implementation.

Main Results:

• The U-Net model, enhanced with a cellular automaton algorithm, demonstrated superior performance with precision=0.91, recall=0.90, F1=0.91, IoU=0.84, and accuracy=0.90.
• Developed segmentation algorithms proved universal, effectively highlighting defects under various imaging conditions and defect sizes.
• Automated damage area calculation and critical/uncritical recommendations were achieved.

Conclusions:

• CNN-based computer vision algorithms provide a robust and accurate method for detecting and analyzing concrete structural defects.
• The U-Net model with cellular automata offers a highly effective solution for automated concrete quality assessment.
• These automated tools can inform structural condition assessments, concrete mix design, and production process adjustments.