Loughborough University
Leicestershire, UK
LE11 3TU
+44 (0)1509 263171
Loughborough University

Loughborough University Institutional Repository

Please use this identifier to cite or link to this item: https://dspace.lboro.ac.uk/2134/9516

Title: Artifical neural network modelling of reverse osmosis process
Authors: Al Shaalan, Hakem
Issue Date: 2012
Publisher: © Hakem Al Shaalan
Abstract: With the increase in population and the scarcity of fresh water in the Middle East desalination has taken an important role in the provision of water for everyday use and for industrial purposes. Reverse osmosis water treatment process is of particular interest as it is one of the key processes in a desalination plant. The modelling of this process and the prediction of permeate flow is useful in better understanding the process. In the present study, an artificial neural network based model was developed based on plant data for the prediction of permeate flow performance. Plant data was collected and a number of variables determined. Principal component analysis was then carried and factor loadings obtained to identify the main variables. Once the main input variables were obtained a statistical analysis of the data was done in order to remove outliers present in the data. This was done because the presence of outliers in data to be analysed using ANN models renders the models ineffective in prediction of an output. Once the removal of outliers was done, the data was then analysed using the developed model. 1081 sets of data were originally used with twelve input variables. After principal component analysis was done the input variables were reduced to five with one output variable. With the removal of outliers 981 sets of data were obtained and these were then used in the model. The model was able to predict the output accurately with r2 at 0.97. Key factors determined from the process were that to obtain an optimum network one has to consider the epoch size, the transfer function, the learning rate and finally the number of nodes in the hidden layers. The number of hidden layers also had an effect on the overall prediction of the data. It is also important when using ANN models to obtain the correct input variables and to remove any outliers that are present in the data in order to be able to predict the output. The use of plant data severely limited optimisation of the process due to it already being heavily optimised.
Description: This thesis is restricted until 31 December 2015. A Doctoral Thesis. Submitted in partial fulfilment of the requirements for the award of Doctor of Philosophy of Loughborough University.
URI: https://dspace.lboro.ac.uk/2134/9516
Appears in Collections:Closed Access PhD Theses (Chemical Engineering)

Files associated with this item:

File Description SizeFormat
Thesis-2012-Al-Shaalan.pdf2.72 MBAdobe PDFView/Open
Form-2012-Al-Shaalan.pdf1.4 MBAdobe PDFView/Open

 

SFX Query

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.