Acoustic emission corrosion feature extraction and severity prediction using hybrid wavelet packet transform and...

Area of Science:

• Materials Science and Engineering
• Structural Health Monitoring
• Machine Learning Applications

Background:

• Corrosion in carbon-steel pipelines is a primary cause of failure and breakdown maintenance in the oil and gas industries.
• Acoustic emission (AE) signal analysis is crucial for modern Structural Health Monitoring (SHM) systems for corrosion detection and classification.
• Extracting effective AE features and classifying corrosion severity remain significant challenges.

Purpose of the Study:

• To propose a hybrid machine learning approach for multiresolution feature extraction and corrosion severity prediction.
• To overcome limitations in existing AE feature extraction and corrosion classification methods.
• To enhance the reliability and accuracy of corrosion assessment in SHM applications.

Main Methods:

• A hybrid approach combining Wavelet Packet Transform (WPT) and Fast Fourier Transform (FFT) for multiresolution AE feature extraction.
• Linear Support Vector Classifier (L-SVC) for predicting corrosion severity levels.
• Laboratory-based Linear Polarization Resistance (LPR) tests for AE data acquisition on carbon-steel samples.

Main Results:

• Simulation demonstrated a linear relationship between extracted AE features and the corrosion process.
• Three distinct corrosion severity stages were identified based on corrosion rate and AE activity.
• The L-SVC classifier achieved a high prediction accuracy of 99.0%, outperforming other benchmarked classifiers.

Conclusions:

• The proposed hybrid machine learning approach effectively extracts multiresolution AE features for corrosion analysis.
• The integrated WPT-FFT and L-SVC model accurately predicts corrosion severity levels.
• This method offers a promising solution for reliable corrosion detection and assessment in SHM applications.