Adaptive abnormal signal extraction based on iterative computations

An adaptive approach of abnormal signal extracting is proposed for pipeline leak detection based on acoustic method in this paper. The dynamic pressure signal is divided into many intervals based on zero crossing points, and each frame of signal is regarded as the superposition of the interval signals. The characteristics of the interval signals and their Signal-to-Noise Ratios (SNRs) are analyzed first. Based on the analysis, the number of abnormal signal intervals is determined depending on the position of inflection point in the Root-Mean-Square error sequence obtained by SNR difference sequences, and then abnormal signals are isolated. In this method, the feature extraction of signals and leakage diagnosis model training are all cancelled, and the problem of missing alarm can be overcome. The field experiment testified the effectiveness of the approach.