F.G. Uler

Design optimization of electromagnetic devices using artificial neural networks

Design optimization of geometric boundaries in elecaomagnetic devices is receiving widespread aaention and interest Recently, several design optimization techniques have appeared in the literature ranging from evolution strategies to simulated annealing and search techniques for global m i ~ i m a . ~ . ~ ~ ‘ In this paper a new method which utilizes Artificial Neural Networks ( A N N s ) is presented. Results of two implementation examples are provided. The major advantages of this new technique are that the optimal design is obtained quickly ( in a matter of milliseconds) once the Neural Networks are trained with a variety of geometrical topologies. Furthermore, the p r d u r e explained in this paper can be used to provide good initial solutions for other iterative search techniques (currently in use) in order to reduce searching time. This aspect is highly desirable to increase the effectiveness of the optimal design procedure and to reduce divergence due to local minima which the majority of other optimization techniques are suffering from.

Detailed 2-D and 3-D finite element modeling of the human body for the evaluation of defibrillation fields

This paper deals with the detailed modeling of the human body for the evaluation of implanted electrode potential distributions and defibrillation fields. Modeling details have been achieved in 2-D and 3-D down to the arbitrary topologies of the human body and its organs. This type of analysis dictated for the need for 2-D and 3-D mesh generators that can be used to discretize arbitrary shapes (as those of the human organs). Developed grid generators provide an optimum finite element model of the problem. The main objective of this study is directed towards the 2-D and 3-D computation of the potential distribution across the human heart and to solving the inverse problem of determining the optimum sizes and locations of the stimulation electrodes to be used in achieving a successful defibrillation.

Utilizing Hopfield neural networks and an improved simulated annealing procedure for design optimization of electromagnetic devices

A new method for the design optimization of electromagnetic devices is presented. Improved simulated annealing (SA) is used along with the Hopfield neural network (HN) to optimize a solution obtained from a trained multilayered perceptron (MLP). This results in an overall increase in the speed of optimization. Results are presented for two examples which show the effectiveness of the proposed method in obtaining a near optimal solution. >

An intelligent system for design optimization of electromagnetic devices

An intelligent system for the design optimization of electromagnetic devices is presented. The system comprises a back-propagation neural network in conjunction with an optimizing Hopfield Net for performing search and optimization functions. An adaptive algorithm is used to improve the response of the system in a dynamic environment. Data can be input by an experienced designer in addition to a case generator from the finite element (FE) solutions. Optimal designs are obtained quickly once the artificial neural network (ANN) is trained with a variety of topologies. Results of implemented examples are provided to show the effectiveness of the proposed system. >

A state space approach and formulation for the solution of nonlinear 3-D transient eddy current problems

A novel method is presented for 3-D nonlinear transient eddy current and field computations, utilizing the state-space approach for handling the time variation. A nonlinear magnetic circuit example was successfully implemented. The results of the proposed method are compared with those of the Crank-Nicholson technique and measured data. These comparisons reveal excellent agreement, and the proposed method outperforms the Crank-Nicholson technique in CPU time, calculating the results in less than 25% of the time required by the Crank-Nicholson method. Formulation details, numerical results, and the various comparisons are given. >

3-D finite element time-varying fields and eddy currents in nonlinear thin steel channels

A method for three-dimensional nonlinear transient eddy current and field computations in thin steel channels separated by air gaps is presented. The nonlinear thin steel plates with narrow gap were placed over a coil which is excited from a time-varying source. Details of the formulation for the nonlinear problem and the time derivative are given. The current amplitude was large enough to saturate the steel parts. The time constant of the excitation function was chosen so that large eddy current density is produced. Experimental data of flux density and eddy current density at specified test points in the analyzed volume were compared with numerical results. Excellent correlation between numerical and experimental results was achieved. >

A 3-D finite element mesh generator for complex volumes

A 3-D mesh generator for finite element analysis of electromagnetic field problems has been developed. The software is capable of handling complex topologies such as the organs of the human body. The grid is optimized by performing Delaunay correction. The use of digital image processing techniques is suggested to extract topological information from images. This information is used for identifying the regions of the model by plating initial nodes on the surfaces. An application example shows a simplified model of the human thorax created using the grid generator. It should be noticed that this problem includes most of the difficulties encountered in 3-D mesh generation for the finite element method. >

Calculation of potential distributions produced by implanted electrodes in an man model by finite elements

The finite-element (FE) method is used to analyze and determine the potential distribution produced by implanted stimulating electrodes used in conjunction with an implanted defibrillator. These electrodes are used to deliver an energy shock to halt ventricular fibrillation during fatal arrhythmias in human beings. The FE solution was obtained by solving the Laplace equation using a Galerkin procedure. An infinite cylindrical model having three coaxial layers of tissue with two electrodes was used to show preliminary results of the FE model. An analytical procedure was developed and used to validate the FE results. The overall objective of this study is to develop a computer model for the study of defibrillation fields. This model can be used to estimate the potential distribution and potential gradients on the heart for various electrode sizes and configurations for a given shock strength. The most promising variation of these defibrillation-electrode configurations can be chosen for experimental testing in laboratory animals and possibly in human beings. >

A state space technique for the solution of nonlinear 3-D transient eddy current problems

The authors present a novel method for 3-D nonlinear transient eddy current and field computations in electrical devices. The method utilizes the state-space approach for handling the time variation in a volume containing excitation coils, metallic structures, and nonconducting media. A nonlinear magnetic circuit example, excited from a time-varying source, was successfully implemented. A comparison between the results of the proposed state-space method, the Crank-Nicolson technique, and measured data is provided to show the effectiveness and validity of the proposed approach. The results of the comparisons reveal excellent agreement while the proposed method outperforms the Crank-Nicolson technique in terms of CPU time. >

3-D finite element grid generation in electromagnetics

A 3-D grid generator for finite-element solutions of electromagnetic field problems is discussed. The method of dividing an entire configuration into smaller regions provides a high degree of flexibility in defining the characteristics of the different portions of the problem. The software is capable of handling a wide variety of objects and is easy to use. It only requires the information about geometries and characteristics of the different regions in the volume. The grid generator algorithm is discussed. The algorithm was applied to a general 3-D example. The example includes most of the items which create difficulty in 3-D discretization of a general volume. This example utilizes thin plates, air-gaps, and curved surfaces, as well as volume variations in all directions.<<ETX>>