Author(s): Tan Zhen; Guo Xinlei; Li Jiazhen

Linked Author(s):

Keywords: Pipeline leak detection; Deep learning models; Unsteady friction coefficient; Multi-scale leakage feature; Convolutional neural network

Abstract: The effective detection of pipeline leakage is critical for saving water and preventing leakage. The pipeline leak detection based on deep learning methods has gained rapid development. In this study, we propose a pipeline leak detection model based on a multi-scale one-dimensional convolutional neural network (MS1DCNN) in extracting features of leakage more accurately. The model employs multiple convolution paths with different scales to classify leak information. The classical piping system layout and the transient flow model is used to predict location, leakage, and unsteady friction coefficients under various pipeline leak conditions in our model with sample sizes of 39,601,3, 980, and 4,900, respectively. The MS1DCNN model is compared with other deep learning models, including one-dimensional convolutional neural network (1DCNN), BP neural network, support vector machine (SVM), and k-Nearest Neighbor (KNN). Simulation results show that MS1DCNN has an average classification accuracy of 99.96%, 98.49%, and 100% for leaking location, leakage, and unsteady friction coefficient, respectively, working better than other models. Furthermore, our model works better in noise environments with SNR of -4 ~ 12dB with higher scores than those of 1DCNN, BP neural network, SVM, and KNN. The model might be further applied to the synchronous prediction of pipeline leakage parameters and unsteady friction coefficient in the real world.

DOI: https://doi.org/10.3850/978-90-833476-1-5_iahr40wc-p1103-cd

Year: 2023