Reconstruction of capacitance tomography data using entropic thresholding techniques.

Recently, the standards for the industries regarding safety, environmental protection, energy conservation, quality assurance and complexity of the products have increased significantly, necessitating more complex and detailed measurement techniques for the design and control of process plants and transportation systems. The desired quality of measuring instruments for such applications is that they must have robust non-invasive sensors, fast dynamic response, require minimum maintenance or calibration, and be compact and intrinsically safe. Industrial process tomography, which is new field of science, has the potential of offering solutions to such demands. It involves using tomography imaging techniques to obtain both space and time variant data rapidly and efficiently. Such data are extremely useful in improving the modelling and design of many complex processes, which are very often difficult to understand using other sensing techniques. The extensive use of the soft-field based sensory system relative to the traditional hard-field based sensory system in process tomography systems, has resulted in a new data set with its associated problems on how best to utilize the non-linear information generated from these systems. Although they have relatively poor resolution, they are however, cost effective, easy to install and maintain, intrinsically safe and have a fast dynamic response. The first soft-field based process tomography system to be developed is the Electrical Capacitance Tomography (ECT). It is the most advanced system thus immediately applicable for research and industrial use. The ECT is mostly suitable for processes involving a mixture of insulating compounds with different permittivity. For instance, it can be used to image component concentration distributions and detect transient dynamic changes of a multiphase process. However, information generated from the system has been useful for qualitative and not quantitative analysis. This is because there are inherent limitations from the system, which produces distortions in the generated information. These distortions affect the accuracy of the measurements obtained from the system. In order to extract meaningful information for decision making by man or machine, further improvement on the quality of tomograms from the system is required. The proposed solution in this thesis is to use thresholding technique, which is based on the principle of maximum entropy. Basically, it involves further processing of the Linear Back Projection (LBP) reconstructed tomograms. By using this approach, segmentation of the tomograms contents into two components of interest can be easily done. This simplification of the tomogram contents is very useful, because a decision can be made without further processing of tomogram data. Different thresholding algorithms based on the principle of maximum entropy have been developed. These algorithms have been found to improve the accuracy of the measurement data at insignificant computational cost, hence useful in many industrial situations. These algorithms have been evaluated for their use in estimating the component fractions in the gas-liquid multiphase flow as applicable to the oil industry. The obtained performance results are encouraging and considered as a step forward toward using capacitance tomography data for quantitative analysis in industrial applications. Although the focus of algorithm evaluation has been for oil industry applications the developed algorithms can be easily extended to other applications.