Artificial Neural Network-Based Detection of Fluids and Scale in Petroleum Pipelines via Capacitive and Electromagnetic Sensors

Abstract

Accurate measurement of volume fractions (VF) in multiphase flows is often challenged by the presence of scale layer deposits on the internal pipe walls. This study presents a robust, non-invasive intelligent metering system designed for the simultaneous estimation of scale thickness and VFs of oil, water, and gas phases. The methodology integrates data from two distinct sensing modalities: a twin rectangular fork-like capacitive sensor, simulated using COMSOL Multiphysics, and gamma-ray attenuation (Co-60 source) calculated via the Beer-Lambert law. To process this hybrid dataset, various Multilayer Perceptron (MLP) neural network architectures were developed and optimized in MATLAB. The proposed model achieves high precision, predicting the VFs of gas, water, and oil with Mean Absolute Errors (MAE) of 0.027, 0.032, and 0.033, respectively. Furthermore, the scale layer thickness was estimated with a remarkable MAE of 0.244 mm. These numerical results confirm the efficiency of the proposed sensor fusion technique in mitigating the adverse effects of scale on measurement accuracy, offering a significant advancement for real-time monitoring in the petroleum industry.