Enhancing the reliability of marine pipeline transportation systems: A flow safety monitoring method for sand-carrying churn flows via multi-migration collision behavioral responses

Abstract

Marine pipeline transportation systems frequently encounter sand-carrying churn flows, wherein persistent sand particle-wall collisions lead to structural degradation of pipelines. This paper proposes a flow safety monitoring method for sand-carrying churn flows based on multi-migration collision behavior responses. Based on the Robust Empirical Mode Decomposition (REMD) algorithm, this study first establishes a multi-frequency scale vibration response characterization method of sand particles for sand-carrying churn flow. Then, a lightweight deep learning architecture based on Depthwise Separable Convolution (DSC) is constructed, achieving an average recognition accuracy of 87.17 % for sand features with contents ranging from 0g to 20g (in 5g increments) across three distinct datasets. Furthermore, the Bidirectional Long Short-Term Memory (BiLSTM) module and Self-Adaptive Temporal Transformer (SATT) module into the DSC framework, thereby enhancing bidirectional full-sequence time-delay feature extraction capability and adaptive weight-matching capacity for holistic particle characteristic information. The DSC-BiLSTM-SATT recognition model improves the average recognition accuracy by 8 %, achieving a final accuracy of 95.17 %. The model shows excellent generalization capability even on low signal-to-noise ratio (SNR) datasets, and the average recognition accuracy for three low SNR datasets reaches 89.73 %. The framework with high accuracy significantly contributes to improve the flow safety and reliability of marine pipeline transportation systems.