An improved variational mode decomposition method based on particle swarm optimization for leak detection of liquid pipelines
Xu Diao,Juncheng Jiang,Guodong Shen,Zhaozhao Chi,Zhirong Wang,Lei Ni,Ahmed Mebarki,Haitao Bian,Yongmei Hao
DOI: https://doi.org/10.1016/j.ymssp.2020.106787
IF: 8.4
2020-09-01
Mechanical Systems and Signal Processing
Abstract:<p>Leak detection is critical for the safety management of pipelines since leakages may cause serious accidents. The present paper aims to develop an efficient method able to detect the presence and importance of leaks in pipelines.</p><p>This method relies on adequate signal processing of acoustic emission (AE) signals, and improves the variational mode decomposition (VMD) for signal de-noising. In order to optimize the governing parameters, i.e. the penalty term and the mode number of VMD, the particle swarm optimization (PSO) algorithm is coupled to a fitness function based on maximum entropy (ME). After the signal reconstruction, based on the energy ratio of each VMD sub-mode, the waveform feature vectors for leak detection are extracted. Finally, the support vector machine (SVM) is employed for leak pattern recognition. For calibration purposes, artificial input signal is simulated. The results show that the proposed PSO-VMD method is capable of de-noising background noise.</p><p>For validation purposes, experiments have been conducted on metal pipelines, with water flow. For sensitivity analysis, a set of five different leak apertures are adopted: aperture diameters as {10; 12; 15; 20; 27} mm, whereas the pipeline diameter is 108 mm. A database of AE signals is collected for each leak aperture, and different time sequences. The proposed method appears to be efficient since the classification accuracy of the SVM method reaches up to 100% in identifying the size of the leak on the basis of the AE signals collected in the database for the same leak size, and 89.3% on the basis of the whole database.</p>
engineering, mechanical