MVSAPNet: A Multivariate Data-Driven Method for Detecting Disc Cutter Wear States in Composite Strata Shield...

Area of Science:

• Engineering
• Geotechnical Engineering
• Machine Learning

Background:

• Disc cutters are critical components in shield tunnel construction.
• Monitoring disc cutter wear is essential for operational safety and efficiency.
• Directly observing cutter wear within the soil silo presents significant challenges.

Purpose of the Study:

• To propose an end-to-end detection method for disc cutter wear state.
• To address the limitations of existing classification methods, particularly data imbalance.
• To improve the accuracy and efficiency of disc cutter wear detection.

Main Methods:

• Developed the Multivariate Selective Attention Prototype Network (MVSAPNet).
• Introduced an attention prototype network for variable selection from multiple input parameters.
• Utilized a prototype network to learn class centers for normal and worn states, detecting wear by feature distance comparison.

Main Results:

• MVSAPNet achieved an accuracy of 0.9187 and an F1 score of 0.8978 on data from the Ma Wan Cross-Sea Tunnel project.
• The proposed method demonstrated superior performance compared to other classification models.
• Successfully addressed the challenge of data imbalance in wear state detection.

Conclusions:

• MVSAPNet offers an effective solution for accurate and efficient disc cutter wear detection.
• The method enhances safety and efficiency in shield tunnel construction by overcoming data imbalance issues.
• The approach shows significant promise for real-world applications in underground engineering projects.