Multi-Priors Tensor Completion for Highly Under-sampled Magnetic Flux Leakage Signal Reconstruction

This is the authors' abstract. We don't add key points for this paper.

Magnetic flux leakage (MFL) detection has been extensively employed in the safety maintenance of oil and gas pipelines. However, MFL signals are frequently rendered incomplete owing to missing entries during signal acquisition. The presence of missing values in MFL signals can significantly impair subsequent tasks, including defect localization, classification, and inversion. A novel signal reconstruction method is proposed, which leverages the inherent priors of pipeline MFL signals. The reconstruction process enhances three priors of the observed MFL signals: tensor low-rankness, local smoothness, and non-local similarity. The resulting constrained optimization problem is addressed within the framework of the Alternating Direction Method of Multipliers (ADMM). The proposed algorithm is evaluated using MFL signals obtained from pipeline pull-through testing. Experimental results indicate that the proposed algorithm accurately reconstructs the features of complete MFL signals. Compared with under-sampled signals, the reconstructed signals substantially improve the accuracy of subsequent defect detection and classification tasks. Furthermore, the proposed method exhibits competitive reconstruction quality and generalization capability compared with existing reconstruction algorithms.