Pipeline systems take a significant role in the transportation of important resources for urban cities. However, these vital pipeline systems are vulnerable to damage. Corrosion and cracks can lead to blockages and leakages in pipeline systems. It is desirable to have a proactive and preventive monitoring system for assessing pipe wall thickness, pipeline radius and pipeline material which can all degrade and be the forerunner to pipeline faults. Recent advances in wireless sensor networks have made it possible to develop proactive pipeline health monitoring systems. Due to limited communication resources, data storage and centralized processing capability of wireless sensor networks, it is highly desirable for each sensor node to extract low dimension features before further da...[ Read more ]

Pipeline systems take a significant role in the transportation of important resources for urban cities. However, these vital pipeline systems are vulnerable to damage. Corrosion and cracks can lead to blockages and leakages in pipeline systems. It is desirable to have a proactive and preventive monitoring system for assessing pipe wall thickness, pipeline radius and pipeline material which can all degrade and be the forerunner to pipeline faults. Recent advances in wireless sensor networks have made it possible to develop proactive pipeline health monitoring systems. Due to limited communication resources, data storage and centralized processing capability of wireless sensor networks, it is highly desirable for each sensor node to extract low dimension features before further data analysis. To address such a feature extraction problem, we exploit the hidden sparsity of acoustic sensor readings in the space of dispersion modes and transform the parameter estimation problem into a sparse signal recovery problem.

We propose an off-grid SBL approach combined with inexact MM algorithm to extract the acoustic mode information from the raw measurements of the sensors. The proposed algorithm has superior performance than standard sparse recovery approaches due to the adjustable off-grid parameter and the inherit data-driven learning capability (no prior knowledge about the parameter, e.g., noise level, of the system model is required) of SBL. It is proved that our proposed algorithm generates sequences of estimates converge to stationary points. Simulation results reveal that our proposed method has substantial performance gain over baseline methods with respect to the recovery of sparse mode information. Moreover, we design a neural network to predict pipeline condition from the extracted information. Simulation results reveal that pipeline health learning based on the recovered mode information extracted by our proposed method can achieve high accuracy.