An Inductive Sensing-Based Pipeline Inspection System With Machine Learning for Defect Reconstruction

Internal metal-loss defects are a leading cause of unplanned pipeline outages, highlighting the critical need for accurate and high-resolution in-line inspection systems. In this paper, an inductive sensing-based inspection system is presented for detecting and characterizing internal metal-loss defects in metallic pipelines. The proposed system utilizes inductance-to-digital converters coupled with planar sensor coils, eliminating the need for separate dedicated transmitter coil/modules or magnetizer assemblies. This integrated transceiver architecture provides a compact, cost-effective, and low-power solution suitable for prolonged pipeline inspections. To automate defect detection and reconstruction, a hybrid neural network (HNN) is introduced as an inversion scheme. The HNN combines convolutional and recurrent subnetworks to effectively extract spatial and sequential features from inductance measurements. Experimental and simulation validations demonstrate the system’s capability to reliably detect and accurately characterize internal defects as small as 4 mm in diameter and 1 mm in depth.