Online data compression of MFL signals for pipeline inspection

The paper presents a novel three-stage algorithm for online compression of magnetic flux leakage (MFL) signals that are acquired in inspection of oil and gas pipelines. In the first stage, blocks of MFL signal are screened for useful information using a semi-robust statistical measure, Mean Absolute Deviation (mu AD). The study presents guidelines for selecting a block size to deliver robust screening and efficient compression ratios. In the second stage, a multivariate approach is used to compress the data across sensors using Principal Component Analysis (PCA). The second stage is invoked only when an anomaly is detected by sufficiently large number of sensors. In the third stage, the signal is further compressed within each sensor (univariate approach) using Discrete Wavelet Transform (DWT). Implementation on real-time MFL signals demonstrates the algorithm's ability to achieve high compression ratios with low Normalized Mean Square Error (NMSE) while being fairly robust to baseline shifts. (c) 2012 Elsevier Ltd. All rights reserved.