MLP ANN Equipped Approach to Measuring Scale Layer in Oil-Gas-Water Homogeneous Fluid by Capacitive and Photon Attenuation Sensors

Abstract

Metering of various parameters is a very imperative task in the gas and oil industries. Therefore, many studies can be found that focus on measuring the volume fractions of multiphase flows without any interruption or separation in the process. One of the key factors highly impacting on the accuracy of the measurements is the scale layer formed in the pipelines. When there is a scale in the transmission lines, it significantly affects measurement accuracy, sensor performance, and fluid dynamics. In this paper, a new approach, including two distinct sensors, photon-attenuation-based and capacitance-based, in conjunction with an Artificial Neural Network (ANN), is presented to measure scale thickness in multiphase oil-gas-water homogeneous fluids. The intelligent model has 2 inputs. While the first input is generated by simulating a capacitive sensor, the concave type, in the COMSOL Multiphysics software, the second input comes from counting rays traveling from a Cobalt-60 source to a detector. This counting is calculated using the Beer-Lambert equations. By considering an interval equal to 10% of material in each ratio, in total, 726 data are accumulated resulting in collecting enough data to measure the scale thickness with a high level of precision. The investigated range for the thickness of the metering scale inside a pipe with a gas-oil-water homogeneous fluid is from 0 cm to 1 cm. Moreover, to reach the lowest amount of Mean Absolute Error (MAE), a number of networks with various hyperparameters were run in MATLAB software, and the best model had MAE equal to 0.46 illustrating the accuracy of the proposed metering system in predicting scale thickness.