Intelligent Testing Method for Multi-Point Vibration Acquisition of Pile Foundation Based on Machine Learning.

Area of Science:

• Civil Engineering
• Geotechnical Engineering
• Artificial Intelligence

Background:

• Conventional low-strain reflected wave methods for pile foundation testing have limitations.
• Assessing the integrity of in-service, high-cap pile foundations requires advanced techniques.

Purpose of the Study:

• To propose an intelligent multi-point vibration acquisition testing model for evaluating pile foundation integrity.
• To overcome the limitations of existing pile testing methodologies.

Main Methods:

• Development of a machine learning-based model utilizing multi-point vibration acquisition.
• Comparative evaluation of different model frameworks, including Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) neural networks.
• Preprocessing of multi-sensor fusion signals using the time series stacking method.

Main Results:

• Both CNN and LSTM models demonstrated high accuracy in locating the first reflection point in the pile shaft.
• Achieved R-squared values > 0.98, Mean Absolute Error < 0.41 m, and Variance Accounted For > 98%.
• The models exhibited strong predictive capability, test stability, and practical utility.

Conclusions:

• The intelligent multi-point vibration acquisition model is effective for assessing in-service pile foundation integrity.
• CNN is recommended for analyzing pile foundation integrity using preprocessed multi-point vibration and multi-sensor fusion signals.
• The findings support operator decision-making in geotechnical engineering applications.