Facial reconstruction

Search LJMU Research Online

Browse Repository | Browse E-Theses

Modelling and Control of Chemical Processes using Local Linear Model Networks

Abdelhadi, A (2020) Modelling and Control of Chemical Processes using Local Linear Model Networks. Doctoral thesis, Liverpool John Moores University.

[img] Text
2019-Ahmed Saad A Abdelhadi-PhD.pdf - Published Version
Restricted to Repository staff only until 7 January 2021.

Download (3MB)
[img] Text
copyright permission 1.pdf - Supplemental Material
Restricted to Repository staff only

Download (468kB)
[img] Text
copyright permission 2.pdf - Supplemental Material
Restricted to Repository staff only

Download (492kB)

Abstract

Recently, technology and research in control systems have made fast progress in numerous fields, such as chemical process engineering. The modelling and control may face some challenges as the procedures applied to chemical reactors and processes are nonlinear. Therefore, the aim of this research is to overcome these challenges by applying a local linear model networks technique to identify and control temperature, pH, and dissolved oxygen. The reactor studied exhibits a nonlinear function, which contains heating power, flow rate of base, and the flow rate of air as the input parameters and temperature, pH, and dissolved oxygen (pO2) the output parameters. The local linear model networks technique is proposed and applied to identify and control the pH process. This method was selected following a comparison of radial basis function neural networks (RBFNN) and adaptive neuro-fuzzy inference system (ANFIS). The results revealed that local linear model networks yielded less mean square errors than RBFNN and ANFIS. Then proportional-integral (PI) and local linear model controllers are implemented using the direct design method for the pH process. The controllers were designed on the first order pH model with 4 local models and the scaling factor is 20. Moreover, local linear model networks are also used to identify and control the level of dissolved oxygen. To select the best method for system identification, a gradient descent learning algorithm is also used to update the width scaling factor in the network, with findings compared to the manual approach for local linear model networks. However, the results demonstrated that manually updating the scaling factor yielded less mean square error than gradient descent. Consequently, PI and local linear model controllers are designed using the direct design method to control and maintain the dissolved oxygen level. The controllers were designed on first and second order pO2 model with 3 local models and the scaling factor is 20. The results for the first order revealed good control performance. However, the results for second order model lead to ringing poles which caused an unstable output with an oscillation in the input. This problem was solved by zero cancellation in the controller design and these results show good control performance. Finally, the temperature process was identified using local linear model networks and PI and local linear model controllers were designed using the direct design method. From the results, it can be observed that the first order model gives acceptable output responses compared to the higher order model. The control action for the output was behaving much better on the first order model when the number of local models M=4, compared with M=3 and M=5. Furthermore, the results revealed that the mean square error became less when the number of local models M=4 in the controller, compared with having number of local models M=3 and M=5.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Local Linear Model Networks
Subjects: T Technology > TP Chemical technology
Divisions: Maritime & Mechanical Engineering
Date Deposited: 09 Jan 2020 11:58
Last Modified: 09 Jan 2020 12:00
Supervisors: Gomm, B
URI: http://researchonline.ljmu.ac.uk/id/eprint/11982

Actions (login required)

View Item View Item