Experimental and Data-driven approach of investigating the effect of parameters on the fluid flow characteristic of nanosilica enhanced two phase flow in pipeline

This study investigates the effect of various parameters on the fluid flow characteristic of nanosilica enhanced two phase (oil-water) flow in the pipeline using experimental and data driven approach. Levenberg-Marquardt (LM) algorithm and Scaled Conjugate gradient (SC) were used for training 20 Artificial Neural Network (ANN) model configurations. The ANN model configurations were optimized using 1 to 20 hidden neurons. Optimized ANN architecture of 3-6-3 and 3-17-3 was obtained for the LM and SC trained ANN. The performance of both the LM and SC trained ANN models were adjudged using mean square error (MSE) and the coefficient of determinant (R). Both the optimized LM and SC trained ANN architecture accurately modeled the prediction of kinematic viscosity, dynamic viscosity, and the pressure of the nanosilica enhanced crude oil. A very low MSE of 9.35x10(-7) and 5.62x10(-2) were obtained for the optimized LM and SC trained ANN architecture, respectively with R values of 0.999. This is an indication of the robustness of the ANN technique used in modeling the effect of the various predictors on the fluid flow characteristic of nanosilica enhanced mixture in pipelines. The sensitivity analysis revealed that the nanosilica concentration has the most significant influence on the various outputs from the ANN model. (C) 2021 THE AUTHORS. Published by Elsevier BV on behalf of Faculty of Engineering, Alexandria University.

Automated summary

This study investigates the effect of various parameters on nanosilica enhanced two phase flow in pipelines using experimental and data-driven approach. Artificial Neural Network (ANN) models accurately predicted fluid parameters with low MSE and high R values, demonstrating the robustness of the ANN technique in modeling fluid flow characteristics. Sensitivity analysis showed that nanosilica concentration had the most significant influence on the outputs of the ANN model.