A Method for Pipeline Leak Detection Based on Acoustic Imaging and Deep Learning

Area of Science:

• Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Pipeline integrity is crucial for safety and environmental protection.
• Traditional acoustic emission (AE) signal analysis for leak detection is hindered by noise.
• Extracting reliable leak signatures from noisy AE data remains a challenge.

Purpose of the Study:

• To develop a robust pipeline leak detection technique using acoustic emission signals.
• To overcome the limitations of traditional AE feature extraction methods.
• To improve the accuracy of leak detection under varying conditions.

Main Methods:

• Acoustic images (AE images) were generated from AE time-series signals using continuous wavelet transform.
• Convolutional autoencoder (CAE) and convolutional neural network (CNN) were employed for feature extraction from AE images.
• Global features from CAE and local features from CNN were merged into a single feature vector.
• A shallow artificial neural network (ANN) was used for final leak state classification.

Main Results:

• AE images effectively represented leak-related information with high energy compared to noise.
• The combined global and local features improved leak detection discriminability.
• The proposed method achieved high classification accuracy on an industrial pipeline testbed dataset.
• Effective leak detection was demonstrated across different leak sizes and fluid pressures.

Conclusions:

• The proposed technique reliably detects pipeline leaks using AE signals and advanced signal processing.
• Integrating CAE and CNN for feature extraction from AE images enhances leak detection performance.
• The method shows significant promise for real-world industrial pipeline monitoring.