SlurryNet: Predicting Critical Velocities and Frictional Pressure Drops in Oilfield Suspension Flows

Improving the accuracy of slurry flow predictions in different operating regimes remains a major focus for multiphase research, and it is especially targeted at industrial applications such as oil gas. In this paper we develop robust integrated method consisting an artificial neural network (ANN) support vector regression (SVR) to estimate critical velocity, regime change, ultimately, frictional pressure drop solid–liquid horizontal pipe, covering wide ranges geometrical parameters. Three distinct datasets were used machine learning models with totals 100, 325, 125 data points drops heterogeneous bed-load respectively. For each dataset, 80% training rest 20% evaluating out sample performance. The K-fold technique was cross-validation. prediction results developed showed that significantly outperforms widely existing correlations literature. Additionally, proposed average absolute relative error (AARE) 0.084 outperformed model without classification AARE 0.155. not only offers reliable over range conditions first time, but also introduces general framework how utilize prior physical knowledge achieve more performances from methods.