Acoustic Emission-Based Offshore Pipeline Valve Leakage Detection Toward Enhanced Process Safety

Area of Science:

• Marine engineering
• Signal processing
• Artificial intelligence

Background:

• Valve leakage in marine oil and gas pipelines poses significant risks to safety, environment, and economics.
• Existing Acoustic Emission (AE) based diagnosis methods lack accuracy, speed, and noise immunity for practical offshore deployment.
• Traditional methods rely on manual feature engineering, limiting adaptability.

Purpose of the Study:

• To develop a lightweight and robust deep learning model for accurate, real-time valve leakage localization in marine pipelines.
• To overcome the limitations of existing AE-driven methods in terms of detection accuracy, inference latency, and noise immunity.
• To enhance the safety and ecological protection capacity of offshore operations.

Main Methods:

• Proposed a Parameter-free Star-shaped Attention Fusion Network (SAFNet) utilizing AE signals.
• Incorporated Temporal Pyramid Encoder (TPE) and Progressive Lightweight Star-shaped Attention (PLSA) modules.
• Integrated Dual Bilinear Star Mapping (DBSM), Energy-Driven Feature Refiner (EDFR), and Multi-Scale Gated Attention Fusion (MS-GAF) for efficient feature extraction and fusion.

Main Results:

• SAFNet achieved high detection accuracy (above 95%) across variable pipeline pressures (2-5 MPa).
• The network demonstrated excellent stability and performance in extreme marine noise environments.
• Experimental results validated the model's balance of accuracy, compact size, and low inference latency.

Conclusions:

• SAFNet provides an efficient, lightweight solution for marine pipeline valve leakage localization using AE signals.
• The proposed method significantly improves intelligent monitoring technologies for offshore systems.
• The study promotes enhanced offshore operational safety and marine ecological protection.