An efficient method for identifying surface damage in hydraulic concrete buildings.

Area of Science:

• Civil Engineering
• Artificial Intelligence
• Structural Health Monitoring

Background:

• Traditional hydraulic structure inspection is manual, inefficient, and labor-intensive.
• Deep learning models often face scalability issues due to limited data size and diversity.
• Existing methods lack comprehensive decision support for maintenance.

Purpose of the Study:

• To propose an effective methodology for identifying diverse apparent damages in hydraulic structures.
• To address the limitations of data dependency and single-functionality in current deep learning models.
• To enhance the accuracy and efficiency of hydraulic structure damage recognition.

Main Methods:

• Fine-tuning lightweight pre-trained models to elucidate advanced damage features, mitigating data dependency.
• Employing ensemble learning algorithms for high-dimensional sample classification to improve accuracy and stability.
• Developing a discriminative feature selection model to reduce inference time and enhance performance.

Main Results:

• The methodology achieved high accuracies: 87.65% for cracks, 87.82% for fractures, 96.99% for holes, and 95.25% for normal structures.
• The proposed feature selection model significantly reduced inference time while improving damage recognition.
• The approach demonstrated mitigation of data dependency issues inherent in deep learning models.

Conclusions:

• The developed methodology offers a robust and efficient solution for intelligent inspection of hydraulic structures.
• This approach significantly enhances the accuracy and speed of identifying various structural damages.
• The findings highlight the practical applicability and value for improving maintenance strategies in hydraulic engineering.