Detection Method for Bolted Connection Looseness at Small Angles of Timber Structures based on Deep Learning

Area of Science:

• Structural Engineering
• Computer Vision
• Machine Learning

Background:

• Bolt looseness is a critical failure factor in timber structures.
• Existing detection methods often lack a balance between cost and accuracy.
• Accurate monitoring of bolted connections is essential for structural safety.

Purpose of the Study:

• To develop a precise and cost-effective method for detecting small bolt loosening angles in timber structures.
• To leverage deep learning and machine vision for enhanced structural health monitoring.
• To validate the proposed method's effectiveness on multi-bolted connections.

Main Methods:

• Utilized deep learning and machine vision technology for bolt looseness detection.
• Trained the Single Shot MultiBox Detector (SSD) algorithm on image datasets.
• Evaluated three recognition schemes targeting nut specification numbers, rectangular marks, and circular marks.

Main Results:

• The scheme identifying circular marks achieved the smallest identification angle error of 0.38°.
• Further improvements led to a minimum recognition angle accuracy of 1°.
• The method demonstrated feasibility and high accuracy on four- and eight-bolted connections.

Conclusions:

• The proposed deep learning and machine vision approach offers a viable solution for detecting bolt looseness in timber structures.
• This method achieves a favorable balance between low operating cost and high accuracy.
• The technique is effective for monitoring both single and multi-bolted connections, enhancing structural safety.