Uncertainty-Quantified Abnormal Detection for Pressure Pipelines Based on Acoustics Hybrid Fusion

Abstract With the development of energy and transportation fields, pressure pipelines are being widely used to transport materials, such as natural gas and water. Working at complex conditions and harsh environments over long time periods, pressure pipelines are prone to damage, which affects the safe and stable operation of the transportation system. Based on its advantage of nondestructive testing, acoustic sensors are employed to detect anomalies in a pressure pipeline. However, the detection accuracy is unsatisfactory due to the uncertainty caused by noise interference and minor damage. Therefore, an uncertainty-quantified abnormal detection method is proposed for pressure pipelines, which quantifies and reduces the uncertainty based on acoustic hybrid fusion. Notably, the data-level fusion method is proposed by dividing and fusing raw acoustic signals at low- and high-frequency bands. Further, the 1D convolutional neural network is employed to obtain preliminary results for two raw and data-level fusion acoustic signals. The uncertainty detection results are modified and corrected through the proposed decision-level method, which can achieve accurate abnormal detection of pressure pipelines with high accuracy and reliability. To validate the proposed method, the natural gas and water pressure pipeline experimental platforms are established with noisy and normal environments. The results demonstrate that the propose method can accurately detect anomalies of pressure pipelines with accuracy over 90% for two study cases. It is superior to other methods based on the ablation experiments and comparative results.