An Automated Image-Based Multivariant Concrete Defect Recognition Using a Convolutional Neural Network with an...

Area of Science:

• Civil Engineering
• Computer Science
• Materials Science

Background:

• Deterioration of buildings and infrastructure in urban areas is a significant concern.
• Traditional manual inspection methods for concrete defects are subjective and limited in accuracy.
• Existing structural flaws like cracks, spalling, and delamination require efficient and objective assessment techniques.

Purpose of the Study:

• To develop an automated, image-based multivariant defect recognition technique for concrete structures.
• To enhance the objective and efficient assessment of structural health issues in concrete.
• To categorize various concrete defects, including surface cracks, delamination, and spalling.

Main Methods:

• A dataset of 3650 images of concrete defects was utilized.
• A convolution-based multivariant defect recognition neural network model was developed.
• The model was trained to differentiate between non-defective concrete and various defect types.

Main Results:

• The developed model achieved a 98.8% accuracy in defect detection.
• The system successfully categorized multiple types of concrete defects.
• The automated technique demonstrated high efficiency in recognizing structural flaws.

Conclusions:

• The proposed automated system offers a reliable solution for concrete defect recognition.
• This technology can accelerate the evaluation of existing infrastructure conditions.
• The developed method promotes advancements in defect detection and recognition for structural health monitoring.