Answers are grounded in this paper and cite their sources.
Traditional quantitative detection methods for natural gas pipeline leakage suffer from limited visualisation and spatial distribution representation. This study proposes a dual-driven framework combining Semantic Segmentation and Fluent modelling for quantitative leakage analysis through multi-source data fusion. GasUNet was developed, a semantic segmentation model integrating Self-Calibrated Convolutions (SCConv) with nnUNetv2, achieving a Dice coefficient of 87.62% and an IoU of 77.99%. GasUNet achieved improvements of 0.75% and 2.3% in the Dice coefficient (Dice) and 1.2% and 5.1% in the Intersection over Union (IoU), respectively, compared to the two baseline models. A CFD-based dynamic leakage simulation system was implemented to predict flow rates accurately. By leveraging infrared thermography parameters, a pixel-to-volume mapping system was established, and a regression model (R²=0.90474) was developed through the integration of time-series analysis with fluid dynamics modelling. This forms a complete technical closed-loop encompassing visual detection, area calculation, flow modelling, and volume prediction, thereby enabling high-precision quantitative analysis. The proposed method provides an efficient and reliable solution for natural gas pipeline leakage monitoring, with significant theoretical and practical implications.
No takes yet. Share an insight, caveat, or question.
Ji et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: