Pipeline leakage detection based on MTF-dual channel multiscale CNN

Aiming at the problem that the subtle feature extraction of one-dimensional(1D) vibration data is not obvious and the real-time detection speed is slow in pipeline leak detection. In this paper, a Markovian transition field(MTF) based pipeline leak detection algorithm with dual-channel multiscale convolutional neural networks(CNN) is proposed. First, the negative pressure wave signal collected by the pressure sensor is converted into a time-correlated two-dimensional(2D) feature image through the MTF encoding method. Then, the feature image is used as the input of a two-channel multi-scale CNN for automatic feature extraction for leak detection. Through comparative experiments, the results show that the average recognition rate of no leakage, 2, 5, and 8 mm leakage holes is 95.37%. Compared with SVM and 1DCNN, the method proposed in this paper has significant improvement and can be applied to pipeline leak detection.