Working mode detection method based on bidirectional LSTM for pipe jacking inertial automatic guidance system.

Area of Science:

• Geotechnical Engineering
• Robotics and Control Systems
• Machine Learning

Background:

• Pipe-jacking inertial guidance is crucial for complex projects but suffers from accumulated gyroscope errors.
• Accurate zero-velocity detection is needed to correct these errors, but slow pipe-jacking speeds hinder traditional methods.

Purpose of the Study:

• To develop a robust method for detecting zero-velocity intervals (ZVI) in pipe-jacking.
• To improve the accuracy and reliability of inertial guidance systems for long-distance and curved pipe-jacking projects.

Main Methods:

• A Bidirectional Long Short-Term Memory (Bi-LSTM) deep learning model was developed for working pattern detection.
• A data collection system was established to train and validate the Bi-LSTM model.
• A zero-velocity correction model was integrated with the Bi-LSTM detection results.

Main Results:

• The Bi-LSTM model achieved 98.54% accuracy on the test set for detecting pipe-jacking working patterns.
• Experimental validation showed accurate ZVI detection, correcting inclination error by 0.06° and azimuth error by 0.18°.
• The model demonstrated strong robustness and adaptability in a real-world pipeline project.

Conclusions:

• The proposed Bi-LSTM based method significantly enhances the accuracy of pipe-jacking working pattern detection.
• This approach effectively addresses the limitations of traditional zero-velocity detection in slow-moving applications.
• The findings provide a foundation for applying improved inertial guidance systems in challenging pipe-jacking scenarios.