Acoustic Emission-Based Pipeline Leak Detection and Size Identification Using a Customized One-Dimensional DenseNet

Area of Science:

• Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Industrial pipeline integrity relies on effective leak detection.
• Traditional methods face limitations in noise sensitivity, adaptability, and computational cost.
• Real-time monitoring requires robust and efficient leak identification solutions.

Purpose of the Study:

• To develop a novel acoustic emission (AE)-based approach for precise pipeline leak detection and size classification.
• To overcome the limitations of traditional monitoring techniques.
• To enhance the safety and longevity of industrial pipelines.

Main Methods:

• Utilized Empirical Wavelet Transform (EWT) for adaptive frequency decomposition and signal segmentation.
• Applied adaptive thresholding and denoising to improve signal quality.
• Employed a customized one-dimensional DenseNet deep learning model for feature extraction and classification.

Main Results:

• Achieved an exceptional leak detection accuracy of 99.76% on real-world AE data.
• Demonstrated reliable differentiation between normal operation and various leak severities.
• Showcased reduced computational costs with robust performance across diverse environments.

Conclusions:

• The proposed AE-based method with EWT and DenseNet offers a highly accurate and efficient solution for pipeline leak monitoring.
• This approach enhances operational safety and integrity by enabling precise leak detection and severity classification.
• The technique is adaptable to various operating conditions, offering a significant advancement over traditional methods.