A Hybrid Convolutional and Recurrent Neural Network for Multi-Sensor Pile Damage Detection with Time Series.

Area of Science:

• Civil Engineering
• Structural Health Monitoring
• Machine Learning Applications

Background:

• Machine learning (ML) is increasingly used for structural health monitoring (SHM).
• Applying ML to pile damage detection (PDD) is challenging due to problem complexity.
• Existing methods may lack the sophistication for automatic and detailed PDD.

Purpose of the Study:

• To propose a novel multi-sensor pile damage detection (MSPDD) method.
• To extend the application of ML algorithms for automatic PDD.
• To enhance the accuracy and detail in classifying pile quality.

Main Methods:

• Utilizing time-series signals from multiple sensors during pile integrity tests.
• Processing signals with traveling wave decomposition (TWD) theory.
• Employing a hybrid one-dimensional (1D) convolutional and recurrent neural network for multi-task identification.

Main Results:

• The hybrid neural network successfully performs automatic multi-task identification of pile damage.
• The model accurately analyzes time-series data from MSPDD.
• Performance evaluation using an analytical solution-based sample set validates the model's effectiveness.

Conclusions:

• The proposed MSPDD method effectively extends ML applications in PDD.
• The hybrid neural network provides detailed pile quality descriptions.
• This approach offers strong support for accurate pile quality classification.