Lightweight Neural Network Architecture for Pipeline Weld Crack Leakage Monitoring Using Acoustic Emission
Jing Huang, Zhifen Zhang, Rui Qin, Yanlong Yu, Guangrui Wen, Wei Cheng, Xuefeng Chen
Abstract
Wall penetration and weld crack leakage of pressure pipelines threaten its service safety. Although Acoustic Emission (AE) technology has been applied to online monitoring, the high sampling frequency poses a huge strain on data transmission and processing. A lightweight intelligent architecture for monitoring crack leakage in pipeline weld is proposed to address this, which can effectively learn high-level abstract features from compressed AE data. Aiming at the problem that the existing crack form is single and divorced from engineering practice, three well-designed experiments with different crack leaks validate the effectiveness of the proposed method. The results show that the number of sampling points is reduced by 80%, and the compressed peak voltage and energy of acoustic emission have better characterization capabilities. Compared to the other state-of-the-art methods, the proposed method has the four best performance metrics, which provides a new possibility for pipeline leak monitoring based on AE in the industrial field.