Real-Time Gas Flow Leakage Detection: A Machine Learning Approach to Sensitivity and Uncertainty Analysis
Leakage monitoring in flow lines and pipelines is highly important in gas plants due to the relevance of such a system to safety and efficiency. This work will, therefore, attempt to resolve the uncertainties in flow monitoring by integrating machine learning techniques in conducting sensitivity tests on real-time detection mechanisms. In this paper, the effectiveness of pressure-based indicators compared with volume changes has been considered with variations in flow rate and lifting processes. The findings obtained showed that the conventional assumption of the leakage being represented by the difference between initial and final gas volumes is unsatisfactory, especially during the initial pumping phase where inflow rates may appear to be less than outflow rates because of the purging of residual gases. In addition, the ramp-up and plateau stages exhibited a fair amount of variation in inflow and outflow pressure readings, further adding to the leak detection uncertainties. It has, therefore, been deduced that a variable tolerance window will be effective for leak detection based on the differential pressure data analysis between the inlet and outlet gauges. According to the result of the data analysis, the population variance is 5.38; the sample variance varies across different stages of operation, while the maximum tolerance and pressure are 0.166 and 7.9 bars, respectively. The work automates leak detection and simulates the range of variations, showing the potentiality of AI-ML modeling in enhancing real-world applications. In this work, we are pointing out how machine learning integration may enable a completely new way to define the variable tolerance windows that dramatically improve conventional leak detection.
