International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017
p-ISSN: 2395-0072
www.irjet.net
Artificial Neural Network modelling for pressure drop estimation of oil-water flow for various pipe diameters Shekhar Pandharipande1, Saurav Surendrakumar2 1Associate
professor, Dept. of Chemical Engineering, LIT, RTM Nagpur University, Maharashtra, India 4th SEM student, Dept. of Chemical Engineering, LIT, RTM Nagpur University, Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------2M.Tech.
Abstract – The flow of two immiscible liquids in pipeline
occurs many times in chemical industries. Oil-water mixture is dispersion and estimation of pressure gradient for flow through pipeline using empirical equation is tedious and less accurate. The present work is aimed at development of artificial neural network models for estimation of pressure gradient as a function of pipe diameter, flow rate, composition of oil-water mixture and angle of elevation of the pipe. 175 experimental runs have been conducted by varying process parameter combinations. Three ANN models have been developed using elite-ANN© based on the experimental data generated. Comparison among actual values with predicted values using ANN models is carried out. Based on the results and discussion, it can be said that all the ANN models developed have excellent accuracy level of prediction for both the training as well as test data sets. The relative error of prediction is in the range of 5 to 20%, highlighting the success of the present work. The work is demonstrative and it is felt that many such models can be developed for various combinations of input and output parameters that is readily available in process industry. Key Words: Artificial Neural Network modelling, two phase flow, oil-water dispersion, pressure gradient and angle of elevation of pipe.
1. INTRODUCTION 1.1 liquid-liquid two phase flow The flow of two immiscible liquids occurs in a pipeline many times in chemical industries, mostly in the petrochemical industry where oil and water are pumped from the wells and transported together [1]. The interactions between the two liquids, with respect to interfacial tension and the wetting properties of the pipe material, means that phenomena are more complex compared to gas-liquid systems [2]. There are large differences in the test fluids used in liquid-liquid experimental studies so that it is difficult to draw any definitive rules for flow pattern boundaries and properties [3]. The flow patterns of liquid-liquid flow are divided into three main categories that are separated flow, dual continuous and dispersed flow. In separated flow both liquids retain their continuity at the top and bottom of the pipe. It consists of © 2017, IRJET
|
Impact Factor value: 5.181
|
stratified flow, where the oil flows above the water and the interface between the two liquids is smooth; and stratified wavy flow where the flow is still stratified but the interface has large waves. Dual continuous flow is flow pattern where both phases remain continuous, but there is a degree of dispersion of one phase into the other. There are limited experimental studies for pressure gradient of liquid-liquid two phase flow in pipelines. There are mainly two types of pressure measuring devices manometers and mechanical gauges. Mechanical gauges are the most used pressure gauges for industrial purposes. Types of mechanical gauges are Bourdon tube gauge, Bellow gauge, Diaphragm gauge and dead weight gauge.
1.2 Artificial Neural Network Artificial Neural network is derived from biological neural network. It can be compared with a black box having multiple inputs and multiple outputs which operates using large number of data which have non-linear relationship with each other. Artificial neurons behave like biological neurons. It accepts signals from adjoining neurons and process to give output signals [4]. There are various types of ANN & Error Back-propagation is one amongst them. It requires at least two layers of nodes. One is the hidden layer and second is output layer. The nodes of two layers are interconnected by the constants called weights. In error back-propagation learning the weights in output layer are corrected first and after having these weights corrected together, the errors have been evenly distributed to the last hidden layer. Then the weights of last hidden layer are corrected and so on [5]. Learning of error back-propagation is in cycles called epochs. The period in which all inputs are presented once to the network is one epoch. After each epoch, RMS (root mean square) error is reported. RMS value decides the accuracy of the model [6]. The aim of all the researchers is to reach as small RMS value as possible. Various applications of ANN in modeling, simulation and optimization of chemical processes have been reported in literature, these include, Estimation of Pressure Drop of Packed Column Using Artificial Neural Network [7], Modeling of Artificial Neural Network for Leak Detection in Pipe Line [8], Artificial Neural Network Modeling of Equilibrium Relationship for Partially Miscible Liquid-Liquid Ternary System [9], Modeling of Packed Bed Using Artificial Neural Network [10], Developing Optimum ANN Model for ISO 9001:2008 Certified Journal
|
Page 3400