Data-driven predictions of pressure drop and liquid holdup in two-phase flow pipes with different inclination angles

AbstractDespite previous efforts to model multi-phase flows using artificial neural networks, there remains significant potential for improvement. This study introduces an advanced artificial neural network framework that analyzes 2651 experimental datasets. The training of the ANN models used 70% of the data, reserving the remaining 30% equally for validation and testing. The proposed network distinctively predicts two critical characteristics of two-phase flow—liquid holdup and pressure drop—with high precision using nine essential predictors. This approach deviates from traditional methods, which typically handle these parameters separately, thereby enhancing both predictive accuracy and efficiency. These include the surface tension of the liquid, the pipeline's tilt, the density and viscosity of both phases, the superficial velocities of the gas and liquid phases, and the pipe's diameter. Notably, this model can process data across all inclinations from −90 (downward) to +90 (upward) degrees and encompasses all flow patterns. This capability, coupled with a comprehensive dataset of high-quality and representative samples, significantly enhances the model's accuracy and real-world applicability, which makes it a noteworthy addition to scientists and engineers in the field of two-phase flow. Performance evaluations show that the ANN model achieved predictive accuracies of 94.38% for liquid holdup and 98.07% for pressure drop, indicating substantial predictive success.