Imperial College London

ProfessorChristosMarkides

Faculty of EngineeringDepartment of Chemical Engineering

Professor of Clean Energy Technologies
 
 
 
//

Contact

 

+44 (0)20 7594 1601c.markides Website

 
 
//

Location

 

404ACE ExtensionSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@inproceedings{Awad:2016,
author = {Awad, M and Azizi, S and Ahmadloo, E and Ibarra, R and Zadrazil, I and Markides, C},
title = {Predicton of interface level height of stratified liquid-liquid flow using artficial neural network},
url = {http://hdl.handle.net/10044/1/78252},
year = {2016}
}

RIS format (EndNote, RefMan)

TY  - CPAPER
AB - In this study, artificial neural network (ANN) was used to predict the interface level height (ILH) of two immiscible liquids flowing in a horizontal pipe. A three-layer feed-forward back-propagation (FFBP) neural network was constructed and trained with experimental data of two different liquid-liquid flow systems reported in the literature. The all studied flow patterns were stratified flow (stratified smooth and stratified wavy with or without droplets at interface ). The input parameters of the ANN model were superficial velocity of phases, pipe diameter, the ratio of the lighter phase density to the heavier phase density (ρlp/ρhp) and the ratio of the lighter phase viscosity to the heavier phase viscosity (μlp/μhp), while the interface level height (ILH) of phases was its output. The Levenberg–Marquardt (LM) algorithm, the hyperbolic tangent sigmoid and the linear activation functions were used for training and developing the ANN. Optimal configuration of the ANN model was determined using minimizing the mean absolute percent error (MAPE) and mean square errors (MSE) between experimental and predicted ILH data by the ANN model. The results showed that the optimal configuration was a network with five neurons in hidden layer that was highly accurate in predicting the interface level. MAPE and correlation coefficient (R) between the experimental and predicted values were determined as 1.8% and 0.9962 for training, and 1.52% and 0.9996 for testing date sets, respectively.
AU - Awad,M
AU - Azizi,S
AU - Ahmadloo,E
AU - Ibarra,R
AU - Zadrazil,I
AU - Markides,C
PY - 2016///
TI - Predicton of interface level height of stratified liquid-liquid flow using artficial neural network
UR - http://hdl.handle.net/10044/1/78252
ER -