Development Of Intelligent Gas-Oil Flow Process Interpreter Based On Generic Primary Electrode Of Electrical Capacitance Tomography Sensor

• Main Author: Mokhtar, Khursiah Zainal
• Format: Thesis
• Language: English
• Published: 2013
• Subjects: TK1-9971 Electrical engineering. Electronics. Nuclear engineering
• Online Access: http://eprints.usm.my/43535/1/Khursiah%20Zainal%20Mokhtar24.pdf; http://eprints.usm.my/43535/

Electrical Capacitance Tomography (ECT) is a technique used to obtain information about the distribution of materials inside a vessel by measuring variations in the dielectric properties of the material distributions. Previous research works on ECT flow regime classification and material fraction estimation have employed Artificial Neural Networks (ANNs) approach focusing on fixed ECT sensor parameters, and hence producing inefficient process interpreter systems. Therefore, this research aims to develop intelligent process interpreter systems which function to accommo- date a range of ECT primary electrode sensor sizes. For the purpose, Multilayer Perceptron (MLP) ANNs have been trained with different types of datasets to investigate the best method in producing generic intelligent gas-oil flow regime classifier and oil fraction estimator. The Principal Component Analysis (PCA) technique has also been used to reduce the dimensionality of input, reduce training time and improve the systems’ performances. The developed intelligent gas-oil classifier has given 93.93% average correct classification accuracy from ECT data of generic primary electrode. This accuracy value is higher than the average classification accuracy of intel- ligent classifier trained with fixed ECT primary electrode size which is 37.45%, for the same test dataset. The developed intelligent oil fraction estimator has produced 3.05% mean absolute er- ror (MAE) for generic ECT data of various flow regimes. This MAE is 3.25% lower than the MAE produced by the best non-generic intelligent oil fraction estimator, based on the same dataset. The satisfactory research results reveal that the performances of generic intelligent gas-oil clas- sifier and oil fraction estimator are better than the non-generic gas-oil classifier and estimator for process interpretation tasks.