Full text
| |
| Author(s): |
Grazieli Luiza Costa Carosio
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Press: | São Carlos. |
| Institution: | Universidade de São Paulo (USP). Escola de Engenharia de São Carlos (EESC/SBD) |
| Defense date: | 2008-12-15 |
| Examining board members: |
Paulo Seleghim Junior;
Vanessa Rolnik Artioli;
André Carlos Ponce de Leon Ferreira de Carvalho;
Marcelo José Colaço;
Flavio Donizeti Marques
|
| Advisor: | Paulo Seleghim Junior |
| Abstract | |
Tomography by electrical sensing represents a technique of great potential for the optimization of processes usually associated with petroleum and chemical industries. However, the employment of tomographic techniques in industrial processes involving multiphase flows still lacks robust and computationally efficient methods. In this context, the main objective of this thesis is to contribute to the development of solution methods based on specific genetic algorithms for the phenomenology of the tomographic problem (interaction between electric and hydrodynamic fields), as well as the adaptation of the algorithm to parallel processing. From qualitative images provided by a direct imaging probe, the basic idea is to generate a model of electric contrast internal distribution and refine it repeatedly until control variables resulting from the numerical model equalize their counterparts, determined experimentally. It can be performed by using an error functional to quantify the difference between non-intrusive external measurements (actual electric current flow) and measurements calculated in a numerical model (approximate electric current flow). According to the functional approach, the numerical reconstruction of the electrical contrast can be treated as a global minimization problem in which the fitness function is an error functional conveniently defined and the global minimum corresponds to the sought image. The major difficulty lies in the nonlinear and ill-posed nature of the problem, which reflects on the topology of the minimization surface, demanding a specialized optimization method to escape from local minima, saddle points, boundary minima and almost plane regions. Although powerful optimization methods, such as genetic algorithms, require high computational effort to obtain the sought image, they are best adapted to the problem in question, therefore parallel genetic algorithms were employed in master-slave, island, cellular and hybrid models (combining island and cellular). The computational performance of the developed algorithms was tested in a tomographic image reconstruction problem of vertical bubble flow. According to the results, the hybrid model can obtain the sought image with a better computational performance, when compared with the other models. Besides, strategies to improve the algorithm efficiency, such as the introduction of a priori information derived from the physical knowledge of the tomographic problem (void fraction and symmetry coefficient of the flow), the insertion of a hash table to avoid the calculation of solutions already found, the use of predation and local search operators were proposed. According to the results, it is possible to conclude that the hybrid model is an appropriate method for solving the electrical impedance tomography problem of multiphase flows. (AU) | |