الفهرس 
PrefaceOnly 14 pages are availabe for public view
 |  | from | 123 |  |  |
| | | from | 123 | | |
Abstract
This thesis introduces a new method for accurate field calculation as well as electrode field optimization. The method is based on combining the Artificial Neural Network (ANN) with the numerical field method under use. Data of wellchosen or generated equipotentials and their resulting electric field intensities are used to train ANN. The trained ANN is then used to determine the electric field for any predetermined electrode configuration in the specified region with acceptable accuracy. Moreover the trained ANN can be used for field optimization by evaluating the desired, but originally unknown, equipotential surfaces with constant electric field in that region. The method has been applicated on Finite Element program (FEP), which are not able to simulate any rounded electrode contours. The trained ANN could give the electric field values for rounded electrode configurations with high accuracy. Specially, by the Charge Simulation Method (CSM), the use of ANN has been proved to be very advantageous. Using a simple charge simulating system for generating a number of equipotentials, in the high stressed zones, to train the ANN leads to the accurate field calculation of any predetermined electrode configuration in that zone. Hereby, the accuracy is direct field strength accuracy (not an accuracy of potential simulation). The obtained results for accurate simulation as well as field optimization have been proved to be very acceptable and very encourage by two dimensional and rotationally symmetric configurations.