Feasibility assessment of calcium carbonate scale thickness and volume fraction estimation in three-phase flow using...

Accurate characterization of fluid volume fractions and scale deposition in multiphase flows is essential for optimizing oil and gas production and transportation. This study presents a simulation-based feasibility assessment integrating Monte Carlo neutron transport simulations with deep neural networks (DNNs) to evaluate whether prompt gamma-ray spectra contain sufficient physical information to support the estimation of volume fractions and calcium carbonate (CaCO₃) scale thickness in three-phase pipeline flows. A²⁴¹Am-Be neutron source and an array of 72 virtual detectors were modeled using MCNP6 code to generate idealized gamma-ray spectra resulting from neutron interactions with gas, oil, saltwater, and scale under annular-flow conditions. The simulated spectra were generated by scoring the energy distribution of gamma rays crossing the detector surface, without explicitly modeling the detector response function or energy resolution and were subsequently used to train DNN regression models. Under controlled simulation conditions, the developed networks demonstrated promising predictive performance. Average relative errors below 1.36% were achieved for scale thickness estimation, while volume fraction predictions exhibited average relative errors below 2.61%. More than 98% of scale thickness predictions and over 85% of volume fraction estimates presented relative errors within 5%, indicating that relevant physical correlations between neutron-induced gamma-ray spectra, scale thickness, and phase distribution can be effectively captured by data-driven models. A feature-selection strategy based on Extreme Gradient Boosting (XGBoost) enabled a systematic reduction of the input dimensionality from 13,680 spectral features to approximately 100, while preserving predictive performance comparable to that obtained using a reference spherical detector configuration.