Burst Analysis for Multi-Level Leakage Detection in Water-Filled Pipeline Based on Acoustic Emission Signal

This paper establishes a methodology to exploit the characteristics of burst waveform in acoustic emission (AE) signals and combine with the signal analysis process to enhance the accuracy of multi-level leak detection in steel pipelines. The AE bursts are signal waveforms that contain the continual imbrication transients with variable strengths in the form of impulses and these impulses include the pivotal information of the leakage syndromes. Capturing and isolation of a burst waveform against the background signal strengthen the capability of a pipeline’s fault diagnosis system. First, this research employs a method using the Enhanced Constant False Alarm Rate (ECFAR) to identify the bursts. Then, the information which is extracted from the burst waveform segment is used to recognize their various sizes of leakage in a laboratory simulated leak system. The training multi-class support vector machine in the one-against-all strategy is responsible for the leak categorization. The result of classification from the proposed method is compared with another algorithm utilizing the wavelet threshold-based burst detection algorithm, which demonstrates the ECFAR method gives an outperforms the wavelet threshold-based algorithm in classification with the accuracy of 93% for different sizes of leakage.