Data and decision-level hybrid fusion method for pressure pipeline pattern recognition

Operating in harsh environments such as high temperature and high pressure for a long time, pressure pipelines are prone to corrosion, wear, and other damages. It can cause abnormal vibrations and radiated noise, leading to malfunctions and seriously affecting the safe and stable operation of gas transmission systems. To timely and effectively recognize and locate abnormal sounds in pressure pipelines, the data and decision-level hybrid fusion is proposed for pressure pipeline abnormal sound detection and pattern recognition. First, the acoustic information deep fusion framework is constructed by deeply fusing multichannel raw acoustic signals at two levels, including data level fusion based on sensor bandwidth and decision level fusion based on information weight allocation. Then, the acoustic information deep fusion framework is applied to multichannel acoustic signals collected from the pressure pipelines for pattern recognition, and it achieves abnormal sound recognition and localization. Specially, the proposed method is tested using publicly datasets GPLA-12 and pressure pipeline experimental datasets. The results illustrate that the proposed method can accurately recognize and locate the abnormal sounds for pressure pipelines pattern recognition. The accuracy exceeded 92% under different working conditions. The superiority of the proposed method is further highlighted by comparing it with other methods.