Early Anomaly Pre-Warning of Buried Pipelines via Dynamic Acceleration Signals: An ICEEMDAN-LSTM Framework

Area of Science:

• Civil Engineering
• Mechanical Engineering
• Data Science

Background:

• Buried pipelines face structural integrity challenges from corrosion, impacts, and ground disturbances.
• Abnormal vibration responses in pipelines signal potential structural anomalies.
• Acceleration signals offer sensitive, direct measurements for monitoring pipeline dynamics.

Purpose of the Study:

• To develop an intelligent framework for buried pipeline condition recognition using acceleration data.
• To enhance early anomaly identification in pipelines through advanced signal processing and machine learning.
• To provide a robust method for pre-warning pipeline anomalies for safe infrastructure operation.

Main Methods:

• Utilized Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) for signal decomposition.
• Extracted multi-scale features from intrinsic mode functions (IMFs).
• Employed a Long Short-Term Memory (LSTM) network for supervised health state classification.

Main Results:

• Achieved an F1-score of 0.70 and a Precision-Recall AUC of 0.72 in anomaly identification.
• Demonstrated stable performance and physical interpretability using multi-source field data (acceleration and strain).
• Confirmed robustness under significant noise interference.

Conclusions:

• The proposed ICEEMDAN-LSTM framework is effective for buried pipeline anomaly pre-warning.
• Combining advanced signal decomposition with deep learning offers a viable approach for structural health monitoring.
• The methodology provides a rigorous basis for ensuring the safe operation of critical energy infrastructure.