Multi-Level-Phase Deep Learning Using Divide-and-Conquer for Scaffolding Safety.

Area of Science:

• Structural Engineering
• Artificial Intelligence
• Computational Mechanics

Background:

• Traditional scaffold analysis is limited to design phases, not operational conditions.
• Automated safety prediction during operation is crucial for preventing accidents.

Purpose of the Study:

• To develop and validate a deep learning-based method for automated scaffold safety prediction during operation.
• To enhance the accuracy and scope of safety assessments for complex scaffolding systems.

Main Methods:

• Implementation of a divide-and-conquer technique combined with deep learning.
• Development of a neural network model trained on extensive simulated datasets (1,540,000 pre-training, 141,100 testing).
• Classification of 1411 safety cases into 18 failure mode categories.

Main Results:

• Achieved an overall accuracy of 99% on test datasets.
• 82.78% of safety cases demonstrated 100% accuracy.
• High precision, recall, and F1 scores indicate robust model performance.

Conclusions:

• The proposed methodology reliably assesses the safety of complex scaffolding systems.
• This approach offers significant improvements over previous methods in accuracy and classification capabilities.
• The technique is adaptable for other classification challenges in engineering.